# Gold Star AI > Gold Star AI builds AI Operating Systems: it maps every action in a business, connects the tools and knowledge already there, and builds AI agents to do the work. Source: https://goldstarlabs.ai Machine view: https://goldstarlabs.ai/machine We Build AI Operating Systems. We learn how your business works, connect the tools and knowledge you already have, and build AI agents to do the work. Each one has a clear job and knows when to bring you in. ## At a glance - **Name:** Gold Star AI - **Also written:** Gold Star - **Builds:** AI Operating Systems: connected tools, shared knowledge, and AI agents with one clear job each - **Offerings:** - Components - Industry OS - Custom AI OS - Labs - **Founders:** Thomas Baker & Michael Krebs - **Industries served:** Commercial Real Estate. Deep Tech. Financial Services. Tech Startups. Local SMBs. - **Contact:** michael@goldstarworkflow.com - **Website:** https://goldstarlabs.ai ## What we build - [Components](https://goldstarlabs.ai/components): An AI agent for a specific job, built to work with the tools you already use. - [Industry OS](https://goldstarlabs.ai/industry): An operating system shaped around the workflows and needs of your industry. - [Custom AI OS](https://goldstarlabs.ai/custom): An operating system built around your business, connecting its tools, knowledge and work. - [Labs](https://goldstarlabs.ai/labs): Bring us the problem, and we’ll study it in the lab and build a solution. ## A business, mapped A business is a set of actions. We map each one. What starts it, what it needs, what it can do, who else it talks to, and where it stops. A shared map of the people, systems and actions that make your business work. ### Sales From the first signal to a signed agreement. Every action has a defined job. #### Outbound Find the right opportunities and prepare a relevant approach. #### Cold email Research, draft, review and send are separate actions with separate permissions. #### Prospect research - **Trigger:** A prospect enters the queue. - **Inputs:** Company website, public signals and your ideal customer criteria. - **Work:** Research the company and prepare a brief with sources. - **Boundary:** Cannot contact the prospect or alter their CRM record. - **Output:** A sourced brief for the email-drafting agent. #### Draft an email - **Trigger:** A reviewed prospect brief is ready. - **Inputs:** Research brief, your offer and examples of your writing. - **Work:** Prepare one relevant, personalized introduction. - **Boundary:** Cannot send the message or invent claims. - **Output:** An email draft ready for human review. #### Review & send - **Trigger:** A draft is submitted for review. - **Inputs:** The draft, supporting sources and your sending rules. - **Work:** Present the draft for approval, then send the approved version. - **Boundary:** Sending requires explicit approval and a valid recipient. - **Output:** A sent message and a recorded approval. #### Find buying signals - **Trigger:** The daily signal check begins. - **Inputs:** Your signal criteria and permitted public sources. - **Work:** Find changes that make a prospect worth a closer look. - **Boundary:** Cannot infer unsupported facts or initiate outreach. - **Output:** A shortlist of sourced opportunities. #### Inbound Turn an incoming conversation into a prepared, useful next step. #### Qualify an inquiry - **Trigger:** A new inquiry arrives. - **Inputs:** The inquiry, account history and qualification criteria. - **Work:** Summarize the need and suggest the appropriate next step. - **Boundary:** Cannot promise pricing or accept a project. - **Output:** A qualification brief for your team. #### Prepare a meeting - **Trigger:** A meeting is added to the calendar. - **Inputs:** The invite, CRM history, notes and relevant documents. - **Work:** Prepare context, open questions and previous commitments. - **Boundary:** Cannot edit the meeting or message attendees. - **Output:** A meeting brief available before the call. ### Operations Make the work between teams as deliberate as the work inside them. #### Project delivery Keep commitments visible and prepare repeatable deliverables. #### Prepare a deliverable - **Trigger:** A deliverable request is approved. - **Inputs:** Project context, prior examples and the required format. - **Work:** Produce a first draft in the agreed structure and style. - **Boundary:** Cannot publish, share externally or fabricate missing information. - **Output:** A formatted draft with open questions marked. #### Prepare a handoff - **Trigger:** A project changes stage. - **Inputs:** Scope, decisions, owners and outstanding commitments. - **Work:** Assemble the context the next owner needs. - **Boundary:** Cannot change scope or assign unapproved commitments. - **Output:** A handoff brief for the responsible person. ### Finance Prepare the paperwork and surface exceptions, with approvals where they belong. #### Billing A clear path from completed work to reviewed paperwork. #### Prepare an invoice - **Trigger:** A billable milestone is confirmed. - **Inputs:** Approved scope, billing terms and the completed milestone. - **Work:** Prepare an invoice from the agreed commercial terms. - **Boundary:** Cannot change rates, charge a card or send an unapproved invoice. - **Output:** An invoice draft for approval. ### Relationships Keep context close so the next conversation picks up where the last one left off. #### Follow-through Remember the commitments that matter to your clients. #### Prepare a follow-up - **Trigger:** An agreed follow-up date arrives. - **Inputs:** Conversation history, commitments and relationship context. - **Work:** Draft a relevant check-in in your voice. - **Boundary:** Cannot send automatically or make new commitments. - **Output:** A contextual draft in your review queue. ### Shared foundation The existing tools, connected systems, shared data and knowledge that every part of the business builds on. #### Existing tools & systems We start with the software the business already uses and establish the connections agents need. #### CRM & business records - **Trigger:** Connect your existing systems. - **Inputs:** Your CRM, operational databases and approved access. - **Work:** Make relevant records available through controlled connections. - **Boundary:** Agents only see and change the records their role permits. - **Output:** Connected business records with clear ownership. #### Email & calendars - **Trigger:** Connect your team's communication tools. - **Inputs:** Existing mailboxes, calendars and communication software. - **Work:** Provide the context and channels needed for each approved workflow. - **Boundary:** Reading, drafting and sending are separate permissions. - **Output:** Tools agents can use within their assigned role. #### Shared data & knowledge Build a common picture of the business so agents can work from the same context. #### Documents & knowledge - **Trigger:** Bring the business's knowledge together. - **Inputs:** Documents, procedures, project history and existing files. - **Work:** Organize useful context and make it retrievable with its sources. - **Boundary:** Respect source access; flag missing or conflicting information. - **Output:** A shared knowledge foundation agents can refer to. #### Connected data - **Trigger:** Map the records that belong together. - **Inputs:** Customer, project and operational data across existing tools. - **Work:** Define shared entities, relationships and sources of truth. - **Boundary:** Preserve provenance and surface uncertain matches for review. - **Output:** A consistent understanding of people, work and relationships. #### Access & operating rules Define how the connected system is allowed to operate before giving agents work. #### Permissions & approvals - **Trigger:** Define each role's authority. - **Inputs:** Your existing permissions, business policies and decision owners. - **Work:** Specify what an agent may read, prepare or execute, and when approval is required. - **Boundary:** Actions outside the assigned scope require an authorized person. - **Output:** Explicit boundaries shared across the operating system. #### History & traceability - **Trigger:** Connect actions to their evidence. - **Inputs:** Inputs, tool results, approvals and changes. - **Work:** Record what happened and preserve the context needed to review it. - **Boundary:** Keep records within the business's access and retention rules. - **Output:** An inspectable history of the system's work. --- # Components > Gold Star's AI components: single-job agent suites that work with the tools a business already uses. Source: https://goldstarlabs.ai/components Machine view: https://goldstarlabs.ai/machine/components ## AEO Suite See how AI finds your business. Track the questions buyers ask, inspect how models answer, and review content built around the gaps. - **Works with:** Your website, model answers, published content - **Try in the demo:** Inspect a question and open the content it suggests. We work with a lot of local small businesses, and for twenty years SEO decided how they showed up. Now that AI is here, SEO is no longer the only game in town, especially with Google traffic falling and traffic from AI climbing. So we asked ourselves the question “if we hired one person whose only job was to make us rank on AI, what would they do all day?” The answer was clear, we’d build agents to figure out exactly what people are searching for any given business, build a routing system to run each of the top 25 queries through the top 12 LLMs each day, then track the results day over day. But we took it a step further, building agents to write AEO optimized content each day that would have the highest chance of moving the needle on ranking. We ended up with a self contained suite that would check what people are searching for, see how each of the top LLMs are answering for each query, then write the content that improved the ranking. All of this while tracking the progress day over day and printing the results on a shared dashboard. ## Signal Engineering and Action Find the right moment to reach out. Turn specific buying signals and opportunities into a shortlist, a relevant contact, and a message ready for your review. - **Works with:** Public sources, contact research, your offering - **Try in the demo:** Open an opportunity and review the outreach it prepared. We tried every cold outreach method in the book and after a year of experimentation we realized one thing that rings true above the rest. People don’t buy things, people buy moments. It's not rocket science, but historically it's been a huge challenge to find the right signals and then figure out exactly how to source them on a regular basis. Even when doing this through tools like Apollo or ZoomInfo, everyone else is mining the exact same signals at the same time which leaves the approach worthless. We realized that you could figure out what makes your buyers tick, figure out the signals that they throw off when they are most likely to buy, and then build a bunch of AI agents to scour the internet on a daily basis to find the right people at the exact right time and reach out with a clear message. No two signals are the same. In fact, we’ve used this to source speaking engagements in the same way that we’ve sourced prospects based on whether or not they are in an active lawsuit in a particular county. Instead of just identifying the signal, we took it a step further. We built a suite of agents to find the right person to reach out to for that particular signal, their contact information, and then draft a custom relevant message that perfectly positions your offering to them for exactly that moment. The result is a simple shortlist of signals to act on each day, the exact contact to reach out to, and the exact right thing to say to maximize the chance of converting that signal. The only thing left for you to do is hit send. ## Paid Ads Suite Turn ad research into campaign drafts. Inspect public ad evidence, explore original creative, and review a campaign package before anything goes live. - **Works with:** Ad libraries, your brand, campaign data - **Try in the demo:** Inspect an ad, refine your creative, and preview the landing page. AI is really good at analyzing data and acting on it, which is the core of running paid ad campaigns on search engines and social media. The age old adage of scale the winners and kill the losers is easier said than done when you have 1000 other priorities and can’t get around to testing 100 variants of the same ad and tracking which one converts better. Rather than paying an ad agency $10k per month to run this for you (plus ad spend), we opted to break it down to its core to see if we could build AI agents to do it. Turns out, you can. The premise is simple, start with seed data on what already works, tweak it with your messaging and offer, try out a few variants, and A/B test yourself into oblivion until the phone is ringing off the hook. Disparate AI systems were already able to do all of the individual pieces of this, so we just put them together in a way that would be useful for this. You just point this suite at your Google or Meta ads, tell it a little bit about your business, confirm the competitors that you want to look at, and then approve/reject potential new campaigns that were generated based on the data. The engine is designed to scale the winners and quickly kill the losers so every dollar of paid ad spend goes as far as it possibly can. ## Matching and Action Engine Find the fit. Prepare the next move. Match an opportunity against your records, see why each result belongs, and prepare the action that follows. - **Works with:** Your people, product catalogs, project briefs - **Try in the demo:** Compare the requirements, inspect the matches, and review the proposed action. So many modern businesses are just different variations of finding a needle in a haystack faster than anyone else can and then doing something with that needle better than anyone else once you find it. We built this engine to help with those businesses. On one side we consolidate all of the data that could go into finding the perfect match (CRM, historical database of customer work, list of capabilities, etc) into one repository. We then build an engine that can find the best match from that collection based on a user input, which could be a real estate listing, a job candidate, or a million other things. This engine takes the input and finds the 5 best matches for it and tells you exactly why it matched them. From there, we build AI agents to be able to execute on whatever action makes the most sense based on that action. For some that could be sourcing the match's contact information and drafting up an email. In other cases it's drafting v1 of a proposal document to apply for a grant. This turns a week’s worth of digging into fifteen minutes of reviewing and editing. And the same engine applies anywhere two sets of information need matching fast and action to be taken immediately. From recruiting to government contracting. ## Deliverable Builder Get the first draft off your desk. Bring project context and examples of past work. Review a structured deliverable shaped around your process. - **Works with:** Project notes, past deliverables, your templates - **Try in the demo:** Choose a project and preview its scope of work. Most businesses produce some variation of the same few deliverables over and over. In a few rare cases, it's as simple as just swapping out a name and date from a template. But most deliverables are just different enough that they need to be rewritten each time with the unique context that feeds it, even though to the author it feels like the same document that they’ve written a hundred times. Since most deliverable heavy businesses have a deep catalog of work, we figured it would be the perfect job for an AI agent. The deliverable builder agent reads every document you’ve ever produced and learns how to write them the way you would. Once it’s trained on your style, we build a portal where you can give it relevant context like email threads, call transcripts, and previous deliverables that inform the context of the present one. This could be PowerPoints, financial models, or any text document as long as you have a catalog of previous similar deliverables. When the next one is due, you point the agent at the relevant context and it hands you a formatted document the way you would have written it. It’s not perfect, but it gets you 95% of the way there in 2 minutes so you can spend your time on the human only finishing touches instead of drafting the same standard document over and over again. ## Golden Nuggets The important things, already prepared. Surface the messages that deserve your attention, gather the context, and prepare the next action. You make the call. - **Works with:** Email, calendar, relationship history - **Try in the demo:** Open a message, edit the reply, and try the approval flow. People are juggling more balls at the same time than ever before. Regardless of any one person’s competence, balls inevitably get dropped and important things slip through the cracks. The amount of noise and AI slop clouds the things that actually need your attention, which is why we built this. So many of the inbox assistants lack context, ability to take action, and can never seem to capture your personality quite right. We think this sucks so we decided to build something that actually works in a way that’s actually helpful. Golden Nuggets connects to your email, calendar, and the rest of your known universe then sits on a dashboard every morning with the stuff that actually matters. It figures out which emails actually need an answer, reads the whole thread behind it, searches in your CRM for any past communication with that person, looks at previous meeting notes and anything else in your known universe. With that context in mind, it drafts a response in your voice and all you need to do is approve it and it’ll send it. It also connects to your calendar and gives you context informed briefs for every meeting. While useful on its own, this feature is at its most powerful when connected to a full AI Operating System. ## Social Media Suite and Trend Spotter Know what to post. Have it ready. Connect relevant trends to your point of view, review platform-specific content, and plan what goes out next. - **Works with:** Social signals, brand guidance, publishing calendar - **Try in the demo:** Pick a trend, compare the captions, and schedule a sample post. An active social media presence is one of the highest leverage marketing channels available, yet it's always the thing that gets dropped when you get busy. With algorithm dominated socials, knowing what to post and how to post it is harder than it's ever been, especially if you don’t have meaningful time to dedicate to it. After scrolling socials for hours on end to try to make sense of what to post, we realized that an agent can watch for us while we handle our busy schedule, distill what’s actually important, and tell us exactly what we should post based on our brand and what the algorithm is pushing that day. That is how the trend spotter was born. In typical Gold Star fashion, we didn’t stop at building the agent to identify the right move, we built a suite of agents to actually execute on it. In this case that means generating images, writing captions and hashtags, and actually scheduling the post to your account on your behalf once you approve it. This also works for longer form content and talking head videos, where the agents will generate you a script that fits your content and brand goals in with what the algorithm is rewarding that day so you have mathematically the best chance to go viral. ## Smart Outbound Prospector Make every introduction specific. Start with your business and who it serves. Inspect five researched prospects and outreach shaped around their situation. - **Works with:** Your ideal customer profile, research, email sequencer - **Try in the demo:** Open a prospect and read the research behind their email. Every B2B business on the planet could benefit from a well run outbound sales motion. When executed correctly, it completely removes chance from the growth equation and puts any business owner in the driver's seat. At Gold Star, we recognized this early and built the smart outbound prospector as our first agent, which resulted in a 12% cold email response rate. When building this, we asked ourselves what we’d do for each prospect if we had unlimited time, money, energy and resources to make sure we’re framing our offering in the best way possible. All you do with this agent is give it some context on your business and a few sentences about your ideal customer profile. The agent takes care of the rest, researching each prospect and learning everything it possibly can about them, then writing them a fully personalized email that is most likely to appeal to that person based on their situation. You can then either load that into an email sequencer like Instantly or Smartlead, or send a few emails per day by hand. Or if you want Gold Star to manage the whole process for you, we can do that too. --- # Industry AI OS > Operating systems Gold Star has already mapped for specific industries alongside their top operators. Source: https://goldstarlabs.ai/industry Machine view: https://goldstarlabs.ai/machine/industry ## Built with the people who know the business best We don’t know what we don’t know, so we sought out the experts. An AI Operating System is only as good as its map, and no one has a better map of an industry than someone who has dedicated their life to it. So instead of guessing what to build, we found the experts and asked them to lend their expertise. What actions actually matter? What interfaces really move the needle? How are things really done? What are the most important systems? How do you measure value? And importantly, what great looks like in the context of this specific industry. We took what they know and built it into an AI OS that was not just helpful for them, but also anyone else operating in their industry. Battle tested with expert operators and ready to deploy in a matter of weeks. Because the industry map is already built, you skip the longest part of the process. ## Faster than custom, deeper than generic Gold Star’s forward deployed engineering team integrates your systems into the existing AI OS, tunes the system to your unique preferences and quirks, and gets you operational in a matter of weeks. Because we’ve already handled the hard stuff, it's a lot cheaper and easier to get started than a fully custom AI OS. ## Industries we cover today Commercial real estate. Roofing. Independent consulting. Residential real estate. Adventure parks and government contracting are coming soon. ## The maps - [Roofing](https://goldstarlabs.ai/industry.md#roofing): 20 actions. Storm to signature to supplement to sweep, run at the speed the first 72 hours after hail demands. - [Residential Real Estate](https://goldstarlabs.ai/industry.md#residential): 20 actions. An agent's whole business, from the lead that came in at 10pm to the client who calls you again in six years. - Adventure Parks: Coming soon. - [Commercial Real Estate](https://goldstarlabs.ai/industry.md#cre): 18 actions. Requirements, availabilities, comps and deal dates, held in one place that is right on the day it changes. - Government Contracting: Coming soon. - [Independent Consulting](https://goldstarlabs.ai/industry.md#consulting): 19 actions. The practice around the work: pipeline that keeps moving while you deliver, and money that actually gets collected. ## Roofing Contractors Storm to signature to supplement to sweep, run at the speed the first 72 hours after hail demands. ### Storm Response The 72 hours after hail where the year's revenue is decided. #### Stand up storm intake - **Trigger:** Hail or high wind hits a zip code in the service area. - **Ceiling:** Within an hour of the storm passing, the office knows which neighborhoods were hit, the phones and web forms are routed to people who can book, and every call is answered live instead of going to voicemail. The competitor who answered the phone stops getting the whole street. - **Touches:** - Storm and hail swath data - Call routing - Inbound lead inbox - **Never:** - Never tells a homeowner their roof is damaged before an inspection - Never promises the insurance will pay - **Agent:** Complex agent that reads the swath, ranks the neighborhoods and opens the intake queue with live answering. - **Measured by:** Percent of storm calls answered live within 60 seconds - **Execution:** Complex agent - **Hands off to:** Route the canvass; Answer an inbound lead; Schedule the inspection #### Route the canvass - **Trigger:** A storm swath is mapped. - **Ceiling:** Crews get door lists ordered by hail size, roof age and homeowner status instead of driving a neighborhood at random. Streets already knocked are marked so nobody doubles back, and every knock result lands in the system the same day. - **Touches:** - Property and roof age records - Storm swath map - Canvass result log - **Never:** - Never routes a canvasser to an address on the do-not-knock list - Never knocks outside permitted hours or without the local solicitation permit - **Agent:** Complex agent that scores and sequences the streets, and keeps the knocked map current. - **Measured by:** Inspections booked per canvasser hour - **Execution:** Complex agent - **Hands off to:** Schedule the inspection; Triage reported damage; Stand up storm intake #### Triage reported damage - **Trigger:** A homeowner calls about a leak or visible damage. - **Ceiling:** Within minutes the call is sorted into active leak, storm damage or age, with photos requested, the address checked against the swath and the emergency tarp crew dispatched when water is coming in. The house with a hole in it stops waiting behind eleven inspection calls. - **Touches:** - Inbound call and photo intake - Storm swath data - Emergency tarp crew schedule - **Never:** - Never diagnoses the cause of damage over the phone - Never quotes a price without the estimator - **Agent:** Complex agent that interviews the caller, classifies urgency and dispatches or books accordingly. - **Measured by:** Median hours from active leak reported to tarp on the roof - **Execution:** Complex agent - **Hands off to:** Schedule the inspection; Build the claim file; Stand up storm intake ### Sales and Estimating From the first call to a signed contract with a real number on it. #### Answer an inbound lead - **Trigger:** A form, a call or a message arrives from a homeowner. - **Ceiling:** Every lead gets a real answer in under five minutes, seven days a week, with the next two inspection slots offered. In a trade where the first contractor to respond wins most of the time, nothing sits in an inbox until Monday. - **Touches:** - Lead inbox and phone queue - Inspection calendar - Service area map - **Never:** - Never quotes a price without the estimator - Never promises a specific insurance outcome - **Agent:** Complex agent that replies, qualifies the address and books the inspection. - **Measured by:** Median minutes to first response on inbound leads - **Execution:** Complex agent - **Hands off to:** Schedule the inspection; Build the estimate; Stand up storm intake #### Schedule the inspection - **Trigger:** A homeowner agrees to an inspection. - **Ceiling:** The appointment is booked against the right rep's real calendar, routed so they cover a neighborhood rather than a county, confirmed the night before and reminded two hours out. No-shows fall and windshield time falls with them. - **Touches:** - Rep calendars - Territory map - Homeowner contact record - **Never:** - Never books an inspection outside the service area without a manager approving it - Never schedules a roof walk during active storm conditions - **Agent:** Simple agent that books, routes by geography, confirms and reminds. - **Measured by:** Inspections completed per rep day, and no-show rate - **Execution:** Simple agent - **Hands off to:** Build the estimate; Build the claim file; Answer an inbound lead #### Build the estimate - **Trigger:** An inspection is complete with measurements and photos. - **Ceiling:** The same day, the rep has a priced estimate built from the measurement report and current material and labor costs, with good, better and best options and every photo attached. Nobody hand-types a squares number into a spreadsheet at 9pm and loses four hundred dollars of margin to a typo. - **Touches:** - Measurement report - Material and labor price list - Inspection photos - **Never:** - Never sends a price to a homeowner without the estimator approving it - Never prices below the company's margin floor - **Agent:** Complex agent that builds the priced scope from the measurements and the current price list. - **Measured by:** Hours from inspection to a priced estimate in the rep's hands - **Execution:** Complex agent - **Hands off to:** Follow up an open proposal; Hand off to financing; Order the material #### Follow up an open proposal - **Trigger:** An estimate is three days old with no decision. - **Ceiling:** Every open proposal gets a follow-up sequence that actually runs, with the photos, the warranty terms and the answer to whatever they hesitated on. Homeowners stop going dark because nobody called back, and the close rate on quoted work is a number the company can see. - **Touches:** - Open proposal list - Homeowner contact record - Estimate and photo package - **Never:** - Never discounts to close - Never contacts a homeowner who asked to stop - **Agent:** Simple agent that runs the sequence and drafts each message against the stated objection. - **Measured by:** Close rate on quoted work - **Execution:** Simple agent - **Hands off to:** Build the estimate; Hand off to financing; Order the material #### Hand off to financing - **Trigger:** A homeowner says the price works but the cash does not. - **Ceiling:** The financing application goes out while the rep is still in the driveway, with the right lender for that credit profile and the monthly payment shown next to the price. Deals stop dying at the kitchen table over a number that was always financeable. - **Touches:** - Lender application portals - Approved estimate amount - Homeowner contact record - **Never:** - Never states a rate or an approval outcome - Never collects or stores a social security number outside the lender's own form - Never advises a homeowner on whether to borrow - **Agent:** Simple agent that sends the right lender application and tracks it to a decision. - **Measured by:** Percent of price-objection deals that close with financing - **Execution:** Simple agent - **Hands off to:** Follow up an open proposal; Build the estimate ### Insurance and Supplements Getting paid for the work the carrier's first scope left out. #### Build the claim file - **Trigger:** A homeowner decides to file a claim. - **Ceiling:** Photos by elevation, test squares, the hail report for that date and address, the measurement report and the damage narrative are assembled into one file within 24 hours. The carrier receives a complete package the first time, so the first scope comes back closer to reality. - **Touches:** - Inspection photo set - Weather verification data - Measurement report - **Never:** - Never files the claim for the homeowner - Never negotiates with the carrier on the homeowner's behalf - Never states what the policy covers - **Agent:** Complex agent that assembles and checks the documentation package against what carriers actually require. - **Measured by:** Percent of claim files complete on first submission - **Execution:** Complex agent - **Hands off to:** Prepare for the adjuster meeting; Compare the carrier scope; Triage reported damage #### Prepare for the adjuster meeting - **Trigger:** An adjuster appointment is set. - **Ceiling:** The rep arrives with the marked test squares, the photo set organized by slope, the code requirements for that jurisdiction and the line items this carrier has historically missed. The meeting takes forty minutes and the scope comes back right. - **Touches:** - Claim file - Local code requirements - Carrier history by line item - **Never:** - Never speaks for the homeowner on coverage - Never marks damage that was not found on inspection - **Agent:** Complex agent that builds the meeting pack and the line-item checklist from this carrier's history. - **Measured by:** Percent of claims approved at the adjuster meeting - **Execution:** Complex agent - **Hands off to:** Compare the carrier scope; Track the supplement; Build the claim file #### Compare the carrier scope - **Trigger:** The carrier's estimate arrives. - **Ceiling:** Within a day the carrier scope is lined up against the company's measured scope, line by line, with every missing item named: drip edge, ice and water, ridge vent, code upgrades, steep and high charges. The gap is a document, not a feeling. - **Touches:** - Carrier estimate - Company measured scope - Local code requirements - **Never:** - Never adds a line item that the inspection does not support - Never sends the comparison to the carrier without the supplement writer reviewing it - **Agent:** Complex agent that reconciles the two scopes and produces the itemized difference with evidence. - **Measured by:** Dollars of missed scope identified per claim - **Execution:** Complex agent - **Hands off to:** Track the supplement; Prepare for the adjuster meeting; Order the material #### Track the supplement - **Trigger:** A supplement is submitted to the carrier. - **Ceiling:** Every open supplement has a submitted date, a carrier contact, a next follow-up and an age. Nothing sits 45 days because the desk adjuster changed. The company stops writing off money it already earned and the supplement approval rate becomes a number on a wall. - **Touches:** - Open supplement list - Carrier correspondence - Claim file - **Never:** - Never accepts a carrier's final position without the supplement writer deciding - Never invoices the homeowner for a supplement that has not been approved - **Agent:** Simple agent that ages the queue, chases the carrier and escalates by days open. - **Measured by:** Supplement approval rate and average days to approval - **Execution:** Simple agent - **Hands off to:** Compare the carrier scope; Handle a change order; Close the job ### Production Material on the roof, crew on the job, permit on the door. #### Order the material - **Trigger:** A job is signed and scheduled. - **Ceiling:** The order goes to the supplier from the measured scope with the right shingle, color, underlayment, ventilation and accessory counts plus the agreed waste factor, delivered the day before the crew and staged on the right side of the house. No crew ever sits in a driveway because the ridge cap did not come. - **Touches:** - Measured scope - Supplier ordering system - Crew schedule - **Never:** - Never orders above the job's material budget without a manager approving - Never substitutes a product line without the homeowner agreeing in writing - **Agent:** Simple agent that builds the order from the scope and times delivery to the crew date. - **Measured by:** Percent of jobs starting with complete material on site - **Execution:** Simple agent - **Hands off to:** Schedule the crew; Pull the permit; Handle a change order #### Schedule the crew - **Trigger:** Material delivery and permit are both confirmed. - **Ceiling:** Crews are booked against real capacity, real weather and real drive time, with steep and complex roofs going to the crews that handle them. The homeowner gets a start date that holds, and the week is not rebuilt every Monday morning by phone. - **Touches:** - Crew capacity and skills - Weather forecast - Job complexity and material status - **Never:** - Never schedules a tear-off into a forecast with rain inside the work window - Never assigns a steep or complex roof to a crew not certified for it - **Agent:** Complex agent that builds the schedule from capacity, weather and job type, and reshuffles when weather moves. - **Measured by:** Percent of jobs started on the promised date - **Execution:** Complex agent - **Hands off to:** Pull the permit; Capture the job photo record; Order the material #### Pull the permit - **Trigger:** A job is signed in a jurisdiction that requires one. - **Ceiling:** The permit application goes in the day after signing with the right forms for that city, the license and insurance attached, and the inspection scheduled to the completion date. No job is ever torn off without a permit on the door and no final inspection is discovered six weeks after the crew left. - **Touches:** - Jurisdiction permit portals - License and insurance documents - Job schedule - **Never:** - Never starts a tear-off before the permit is issued - Never submits under a license that does not cover that jurisdiction - **Agent:** Script that files the application, tracks issuance and calendars the required inspections. - **Measured by:** Percent of jobs with permit issued before tear-off - **Execution:** Deterministic script - **Hands off to:** Schedule the crew; Close the job; Order the material #### Capture the job photo record - **Trigger:** A crew is on a roof. - **Ceiling:** Before, during and after photos come off the job the same day, tagged to the address and the stage: decking, underlayment, flashing, ventilation, final. The warranty claim two years out is answered from the file, and the homeowner sees the work they cannot climb up to look at. - **Touches:** - Crew photo uploads - Job record - Homeowner update channel - **Never:** - Never sends a photo to the homeowner that shows unfinished flashing as complete - Never closes a stage without the required photos on file - **Agent:** Script that collects, tags and files the photo set, and blocks stage sign-off when photos are missing. - **Measured by:** Percent of jobs with a complete staged photo record - **Execution:** Deterministic script - **Hands off to:** Close the job; Register the warranty; Handle a change order #### Handle a change order - **Trigger:** A crew finds rotten decking or an extra layer during tear-off. - **Ceiling:** The find is photographed, priced from the standard rate sheet and in front of the homeowner for approval within the hour, while the crew keeps working on the rest. Nobody eats four sheets of decking and nobody argues about it at the final invoice. - **Touches:** - Crew field report and photos - Standard change rate sheet - Signed contract terms - **Never:** - Never performs extra work before the homeowner approves in writing - Never prices a change outside the agreed rate sheet without a manager - **Agent:** Simple agent that prices the change from the photos and the rate sheet and routes it for signature. - **Measured by:** Percent of change orders approved before the work is done - **Execution:** Simple agent - **Hands off to:** Capture the job photo record; Track the supplement; Close the job ### Warranty and Aftercare The last five percent of the job that decides the next referral. #### Close the job - **Trigger:** A crew marks the roof complete. - **Ceiling:** The magnet sweep, the gutter clean, the final photos, the city inspection and the homeowner walkthrough all happen before the job closes, and the final invoice goes out the same day. No job lingers for three weeks between finished and billed. - **Touches:** - Completion checklist - Final photo set - Inspection and invoice records - **Never:** - Never closes a job with an open punch item or a failed inspection - Never invoices before the homeowner walkthrough is done - **Agent:** Simple agent that runs the checklist, confirms each item and releases the invoice when it passes. - **Measured by:** Days from crew complete to invoiced - **Execution:** Simple agent - **Hands off to:** Register the warranty; Ask for the review; Capture the job photo record #### Register the warranty - **Trigger:** A job closes with a manufacturer system installed. - **Ceiling:** The manufacturer warranty is registered inside its window every time, the workmanship warranty document goes to the homeowner with the photo record, and both are in a file that answers a call five years later in one minute. - **Touches:** - Manufacturer registration portals - Job and material records - Homeowner document delivery - **Never:** - Never registers a system that was not installed to the manufacturer's spec - Never extends a warranty term beyond the contract - **Agent:** Script that files the registration from the job record and delivers the homeowner package. - **Measured by:** Percent of eligible jobs registered inside the manufacturer window - **Execution:** Deterministic script - **Hands off to:** Close the job; Ask for the review; Capture the job photo record #### Ask for the review - **Trigger:** A job closes clean and the homeowner is satisfied. - **Ceiling:** The ask goes out the day the driveway is swept and the homeowner is standing there happy, naming the crew and making the review one tap. In a trade where homeowners pick on reviews, the rating tells the truth and new reviews arrive every week instead of every renewal. - **Touches:** - Closed job records - Review platform links - Homeowner contact record - **Never:** - Never offers anything in exchange for a review - Never asks a homeowner with an open punch item or a warranty complaint - Never posts a public reply to a negative review without a manager approving it - **Agent:** Simple agent that watches for the moment, drafts the ask and holds negative-review replies for approval. - **Measured by:** Reviews earned per hundred completed jobs - **Execution:** Simple agent - **Hands off to:** Close the job; Register the warranty ## Residential Real Estate An agent's whole business, from the lead that came in at 10pm to the client who calls you again in six years. ### Lead Generation Every inquiry, sign call and open house name, answered before it cools. #### Answer a portal lead - **Trigger:** An inquiry lands from a listing portal, often at 10pm. - **Ceiling:** Every lead gets a real reply in under two minutes with the price, the taxes, the HOA number and two times to see it. Nobody waits until morning. The agent wakes up to a conversation already in progress instead of a name that went cold. - **Touches:** - The lead inbox - Public listing data - Showing calendar availability - **Never:** - Never answers whether a neighborhood is safe or who lives there - Never states a commission or offers a rebate - **Agent:** Complex agent that replies in plain language, answers property facts and hands off a live conversation. - **Measured by:** Median seconds to first reply on inbound leads - **Execution:** Complex agent - **Hands off to:** Qualify a new lead; Run buyer intake; Book a tour route #### Qualify a new lead - **Trigger:** A lead replies to the first message. - **Ceiling:** Timeline, price range, financing status, current home and who else is deciding are all captured in the first conversation, not on a call three weeks later. The agent spends Saturday with the two buyers who can actually close, not the six who are browsing. - **Touches:** - Lead record - Conversation history - Lender referral list - **Never:** - Never asks about race, religion, family status or national origin - Never pushes one specific lender as a requirement - **Agent:** Simple agent that asks the qualifying questions conversationally and writes the summary to the record. - **Measured by:** Percent of new leads with timeline and financing captured in 48 hours - **Execution:** Simple agent - **Hands off to:** Run buyer intake; Prepare a listing appointment; Run the buyer's search #### Capture and work an open house - **Trigger:** An open house ends on Sunday afternoon. - **Ceiling:** Every visitor is in the system before the agent leaves the driveway, tagged by whether they have an agent, what they said about the price and what they are actually looking for. Follow-up goes out that evening, personal to what they said in the kitchen. No stack of paper sheets in a car door. - **Touches:** - Sign-in records - Visitor notes - Follow-up templates - **Never:** - Never contacts a visitor who is already represented by another agent - Never shares visitor comments with the seller by name - **Agent:** Simple agent that cleans the sign-in list, drafts a personal note each, and files the seller feedback separately. - **Measured by:** Percent of open house visitors contacted within 24 hours - **Execution:** Simple agent - **Hands off to:** Send the weekly seller update; Qualify a new lead; Run the buyer's search #### Prepare a listing appointment - **Trigger:** A seller books a time to talk about listing. - **Ceiling:** The agent walks in with the tax record, the permit history, the last two sales of that house, six real comps, the likely net sheet and what the seller posted about the remodel. The conversation is about strategy, not about looking things up. Listings are won on the first visit. - **Touches:** - Public records - Closed sale history - Active competing listings - **Never:** - Never commits to a list price before the agent sees the house - Never quotes a commission - **Agent:** Complex agent that builds the property research file and the comp set with sources shown. - **Measured by:** Listing appointment to signed agreement rate - **Execution:** Complex agent - **Hands off to:** Scope the pre-listing work; Launch a listing; Qualify a new lead ### Listings and Sellers From the listing appointment to the day the sign comes down. #### Scope the pre-listing work - **Trigger:** A listing agreement is signed. - **Ceiling:** Inside a day the seller has a room-by-room punch list with costs, a paint and declutter plan, the stager booked and the photographer on the calendar behind them. Nothing goes live with a dark living room and a laundry basket in frame because the schedule slipped. - **Touches:** - Property photos and notes - Vendor calendars - Listing timeline - **Never:** - Never books work over the seller's stated budget - Never promises a repair will raise the price by a specific number - **Agent:** Simple agent that builds the punch list from the walkthrough notes and books the vendor sequence backward from the go-live date. - **Measured by:** Days from signed agreement to photo-ready - **Execution:** Simple agent - **Hands off to:** Launch a listing; Send the weekly seller update #### Launch a listing - **Trigger:** Photos are delivered and the price is agreed. - **Ceiling:** Remarks, room dimensions, disclosures, features and the syndication feed all go live in one pass, written for that specific house, with the copy checked for fair housing language before anyone sees it. The listing hits the portals Thursday morning with the video, not Friday afternoon with three photos. - **Touches:** - Listing entry system - Photo and video assets - Disclosure documents - **Never:** - Never publishes without the seller approving the remarks and the price - Never uses language describing the kind of buyer a home would suit - **Agent:** Complex agent that writes the remarks, assembles the media and runs the compliance pass before a human publishes. - **Measured by:** Hours from photo delivery to live listing with full media - **Execution:** Complex agent - **Hands off to:** Collect showing feedback; Send the weekly seller update; Prepare the price conversation #### Collect showing feedback - **Trigger:** A buyer agent shows the listing. - **Ceiling:** Feedback is asked for within two hours while the showing is fresh, chased once, and rolled into a pattern the seller can act on. Ten showings and no offers stops being a mystery and becomes a sentence: eight of ten said the kitchen, four said the price. - **Touches:** - Showing service records - Buyer agent contacts - Feedback log - **Never:** - Never passes a personal or insulting remark to the seller verbatim - Never asks a buyer agent about their client's finances - **Agent:** Script that requests feedback, chases once and groups the answers into themes. - **Measured by:** Feedback response rate per showing - **Execution:** Deterministic script - **Hands off to:** Send the weekly seller update; Prepare the price conversation; Launch a listing #### Send the weekly seller update - **Trigger:** Every Friday a listing is active. - **Ceiling:** Every seller gets the same honest page each week: views, saves, showings, feedback themes, what the three competing homes did, and the recommendation. Sellers stop calling to ask if anything is happening because they already know before they wonder. - **Touches:** - Listing performance data - Showing and feedback logs - Competing active listings - **Never:** - Never recommends a price change without the agent approving the number - Never shares a buyer agent's name or a visitor's identity - **Agent:** Simple agent that builds the update from the week's data and writes the plain English note. - **Measured by:** Percent of active listings with an update delivered every week - **Execution:** Simple agent - **Hands off to:** Collect showing feedback; Prepare the price conversation; Capture and work an open house #### Prepare the price conversation - **Trigger:** A listing passes 21 days with no offer. - **Ceiling:** The agent walks into that call with the showing curve, the feedback themes, the three homes that sold instead of this one, and what each price band would likely do to days on market. It is a conversation with evidence, not a plea. The house stops sitting at the wrong number for two months. - **Touches:** - Listing activity history - Feedback themes - Closed and pending comps - **Never:** - Never changes the list price - Never sends the recommendation to the seller before the agent reads it - **Agent:** Complex agent that diagnoses which part of the funnel is failing and builds the case with sources. - **Measured by:** Median days on market against the local average - **Execution:** Complex agent - **Hands off to:** Send the weekly seller update; Collect showing feedback; Launch a listing ### Buyers What they say they want, what they actually tour, and the offer that wins. #### Run buyer intake - **Trigger:** A qualified buyer agrees to work with the agent. - **Ceiling:** Must-haves, deal-breakers, schools, commute, budget ceiling versus comfort, lender status and who else signs are all written down in one place before the first tour. The agent stops showing houses that were never going to work. - **Touches:** - Buyer record - Search criteria - Lender pre-approval status - **Never:** - Never steers toward or away from any neighborhood on demographic grounds - Never sets a budget above what the pre-approval supports - **Agent:** Simple agent that runs the intake conversation and writes the criteria into a working search. - **Measured by:** Homes toured per buyer before an accepted offer - **Execution:** Simple agent - **Hands off to:** Run the buyer's search; Book a tour route; Prepare an offer #### Run the buyer's search - **Trigger:** A new listing or price change matches a buyer's criteria. - **Ceiling:** The buyer hears about the right house within minutes of it hitting the market, with a one-line note on why this one and not the other four. Buyers stop finding their own house on a portal and asking the agent to catch up. - **Touches:** - Active listing feed - Buyer criteria - Tour history and reactions - **Never:** - Never sends listings outside the agreed criteria without saying why - Never comments on the character of an area or its residents - **Agent:** Simple agent that watches the feed, filters against real tour behavior and writes the short reason. - **Measured by:** Percent of buyer tours sourced by the agent rather than the client - **Execution:** Simple agent - **Hands off to:** Book a tour route; Run buyer intake; Recap the tour #### Book a tour route - **Trigger:** A buyer picks four homes to see Saturday. - **Ceiling:** Access is confirmed with every listing side, the route is ordered so the drive makes sense, the buyer gets addresses and times, and each home comes with its taxes, HOA and days on market on one page. Nobody stands on a porch waiting for a lockbox code. - **Touches:** - Showing service - Buyer and agent calendars - Listing access instructions - **Never:** - Never books a showing without confirmed listing-side approval - Never shares a lockbox code with the buyer - **Agent:** Simple agent that requests access, sequences the route and sends the tour sheet. - **Measured by:** Percent of booked tours with confirmed access before departure - **Execution:** Simple agent - **Hands off to:** Recap the tour; Run the buyer's search; Prepare an offer #### Recap the tour - **Trigger:** A tour ends. - **Ceiling:** Each home gets a yes, no or maybe with the reason in the buyer's own words, filed the same afternoon. By the fourth tour the search has narrowed because the pattern is visible: they say they want the yard and they keep picking the kitchen. - **Touches:** - Tour notes - Buyer criteria - Listing records - **Never:** - Never changes the buyer's stated criteria without telling them - Never shares a buyer's reaction with the listing side beyond standard feedback - **Agent:** Simple agent that captures the reactions and updates the working search. - **Measured by:** Tours per accepted offer, trending down - **Execution:** Simple agent - **Hands off to:** Run the buyer's search; Prepare an offer; Book a tour route #### Prepare an offer - **Trigger:** A buyer says they want to write on a house. - **Ceiling:** Within the hour the buyer has the comps for that exact house, what similar homes closed at over or under ask, the seller's likely timeline from public record, and a clear picture of what each term costs them. Offers go out the same day with the pre-approval, the proof of funds and every date filled correctly. - **Touches:** - Comparable sales - Offer templates - Pre-approval and funds documents - **Never:** - Never sets the offer price or waives a contingency - Never submits an offer without the buyer signing - Never advises on the legal effect of a contract term - **Agent:** Complex agent that builds the pricing case, assembles the packet and fills the form for review. - **Measured by:** Offer acceptance rate, and hours from decision to submitted offer - **Execution:** Complex agent - **Hands off to:** Build the contract timeline; Recap the tour; Prepare the inspection response ### Under Contract Thirty days of dates and documents where deals die quietly. #### Build the contract timeline - **Trigger:** An offer is accepted. - **Ceiling:** Every date in the contract becomes a tracked deadline the same hour it is signed: earnest money, inspection, loan commitment, appraisal, walkthrough, closing. Everyone on the deal sees the same calendar. No contingency ever expires because somebody was counting days in their head. - **Touches:** - Executed contract dates - Transaction calendar - Party contact list - **Never:** - Never waives or extends a deadline - Never interprets a contract clause for a client - **Agent:** Script that reads the dates, builds the deadline set and alerts the file owner ahead of each one. - **Measured by:** Contingency deadlines missed per hundred contracts - **Execution:** Deterministic script - **Hands off to:** Prepare the inspection response; Watch the appraisal and the loan; Run the closing week #### Prepare the inspection response - **Trigger:** An inspection report is delivered. - **Ceiling:** The forty page report becomes a one page list within two hours: safety items, real repairs, cosmetic noise, with rough cost ranges and what is typical to ask for in this market. The client decides on facts instead of panicking at the word mold. - **Touches:** - Inspection report - Repair cost references - Contract repair terms - **Never:** - Never advises a client to accept or reject a repair - Never estimates a structural or environmental cost without a licensed opinion - Never sends the request without the client approving every line - **Agent:** Complex agent that summarizes and categorizes the report and drafts the request for review. - **Measured by:** Percent of inspection responses delivered inside the contract window - **Execution:** Complex agent - **Hands off to:** Build the contract timeline; Run the closing week; Prepare an offer #### Watch the appraisal and the loan - **Trigger:** The appraisal is ordered. - **Ceiling:** The lender gets a comp package the day the appraisal is scheduled, the agent knows the result the day it lands, and a low value becomes a plan within hours instead of a scramble on day 25. Loan conditions are chased weekly, not discovered at the closing table. - **Touches:** - Comparable sales - Lender contact record - Loan condition checklist - **Never:** - Never contacts the appraiser to influence a value - Never renegotiates price after a low appraisal - **Agent:** Simple agent that sends the comp package, tracks the milestone and chases open loan conditions. - **Measured by:** Percent of closings delayed by a loan or appraisal surprise - **Execution:** Simple agent - **Hands off to:** Build the contract timeline; Run the closing week; Prepare an offer #### Run the closing week - **Trigger:** Closing is seven days out. - **Ceiling:** Utilities, the walkthrough, wiring instructions verified by phone, the settlement statement reviewed line by line, keys and the after-closing follow-up are all handled on a checklist that never varies. Nobody wires money to a spoofed address and nobody arrives at closing missing a document. - **Touches:** - Closing checklist - Title and settlement contacts - Client contact record - **Never:** - Never sends or forwards wiring instructions - Never approves a settlement statement on the client's behalf - **Agent:** Script that runs the checklist, confirms each item and escalates anything still open at 48 hours. - **Measured by:** On-time closing rate - **Execution:** Deterministic script - **Hands off to:** Build the contract timeline; Keep a past client warm; Ask for the review and the referral ### Sphere and Repeat The people who already bought from you, kept warm on purpose. #### Keep a past client warm - **Trigger:** A purchase anniversary, a rate move, or a neighbor's home sells. - **Ceiling:** Every past client hears from the agent four times a year with something actually about their house: what it is worth now, what sold on their street, what their tax assessment did. Not a mass newsletter. The second sale comes back to the same agent. - **Touches:** - Past client records - Public sale and assessment data - Contact preferences - **Never:** - Never sends a home value estimate as a formal opinion of value - Never contacts a client who asked to be left alone - **Agent:** Simple agent that watches for the moment and drafts the personal note with the local numbers. - **Measured by:** Percent of closings sourced from past clients and sphere - **Execution:** Simple agent - **Hands off to:** Ask for the review and the referral; Run the closing week; Qualify a new lead #### Ask for the review and the referral - **Trigger:** A closing goes well, or a client sends a thank you. - **Ceiling:** The ask lands in the week the client is happiest, names the specific thing that went well, and makes leaving the review one tap. Reviews stop being something the agent chases once a year before renewal. - **Touches:** - Closed transaction records - Review platform links - Client contact preferences - **Never:** - Never offers anything of value in exchange for a review - Never asks a client with an open dispute or complaint - Never posts a public reply about a transaction without the agent approving it - **Agent:** Simple agent that watches for the moment and drafts the ask. - **Measured by:** Reviews earned per closing, and referral share of new business - **Execution:** Simple agent - **Hands off to:** Keep a past client warm; Run the closing week ## Commercial Real Estate Requirements, availabilities, comps and deal dates, held in one place that is right on the day it changes. ### Tenant Representation A requirement in someone's head turned into a signed lease in a real building. #### Intake a tenant requirement - **Trigger:** A company says they need space. - **Ceiling:** Size, use, timing, current lease expiry, parking, power, ceiling height, budget, who signs and what would kill the deal are all captured in the first meeting and written into one requirement document everyone works from. No survey is ever built against a number somebody half remembered. - **Touches:** - Requirement document - Client contact record - Existing lease abstract - **Never:** - Never commits to a rental rate or a term - Never signs a representation agreement - **Agent:** Simple agent that runs the intake, structures the requirement and flags what is still unanswered. - **Measured by:** Percent of requirements fully specified before the first survey - **Execution:** Simple agent - **Hands off to:** Build the survey; Book the tour; Capture a comp #### Build the survey - **Trigger:** A requirement document is signed off. - **Ceiling:** Within a day the client has every qualifying availability in the market, not just the ones in one database: size, rate, TI, parking ratio, floor plate, delivery date and a photo, laid out the same way for each. The buildings that were quietly taken off the market are already struck. - **Touches:** - Availability databases - Building records - Requirement criteria - **Never:** - Never publishes a rate the listing side has not confirmed - Never includes a building the client has already been shown by another broker - **Agent:** Complex agent that assembles, verifies and formats the survey with sources on each line. - **Measured by:** Hours from signed requirement to survey delivered - **Execution:** Complex agent - **Hands off to:** Book the tour; Build the proposal comparison; Refresh availabilities #### Book the tour - **Trigger:** The client shortlists buildings from the survey. - **Ceiling:** Listing brokers are confirmed, building access and COIs are arranged, the route runs in a sensible order, and the client gets a tour book with the plan, the rate and the questions to ask at each stop. Nobody stands in a lobby waiting on a security desk. - **Touches:** - Listing broker contacts - Building access requirements - Client and broker calendars - **Never:** - Never tours a building without the listing side registering the client - Never discusses the client's budget with a listing broker - **Agent:** Simple agent that confirms access, sequences the route and produces the tour book. - **Measured by:** Percent of tours confirmed with access arranged 24 hours ahead - **Execution:** Simple agent - **Hands off to:** Build the proposal comparison; Build the survey; Track the LOI #### Build the proposal comparison - **Trigger:** Two or more landlord proposals are in hand. - **Ceiling:** Every proposal is normalized to the same math: net effective rent over the term, free rent, TI allowance, escalations, operating expense base year, the option language and the real cost per seat per month. The client picks on the actual number, not on the headline rate. - **Touches:** - Landlord proposals - Requirement criteria - Market rate benchmarks - **Never:** - Never counters a landlord proposal without the client deciding - Never interprets a lease clause as legal advice - Never recommends signing anything - **Agent:** Complex agent that models each proposal on the same basis and writes the differences in plain language. - **Measured by:** Net effective rent achieved against the market benchmark - **Execution:** Complex agent - **Hands off to:** Track the LOI; Book the tour; Abstract the lease ### Landlord Agency Space to fill, a stacking plan to keep honest, and an owner to report to. #### Keep the stacking plan current - **Trigger:** A lease is signed, amended or terminated in a represented building. - **Ceiling:** The stacking plan matches reality the day it changes: who is on each floor, how much they have, when they expire, what options they hold. An owner call never starts with three people disagreeing about what is on the fourth floor. - **Touches:** - Lease abstracts - Building floor plans - Rent roll - **Never:** - Never alters a rent roll figure from the owner's accounting - Never shares one tenant's economics with another tenant - **Agent:** Script that updates the plan from executed lease documents and flags any conflict with the rent roll. - **Measured by:** Days between a lease event and the stacking plan reflecting it - **Execution:** Deterministic script - **Hands off to:** Refresh availabilities; Watch lease expirations; Report to the owner #### Refresh availabilities - **Trigger:** Monday morning, and whenever a space status changes. - **Ceiling:** Every listing the firm controls shows a current rate, a current available date and a current floor plan on every platform at once. Nobody calls about a suite that leased in March, and nobody loses a tour because the listing still said negotiable. - **Touches:** - Listing platforms - Space status records - Marketing materials - **Never:** - Never changes an asking rate without the owner's approval - Never marks a space available before the current tenant's term ends - **Agent:** Script that syncs status across platforms and reports anything out of date. - **Measured by:** Percent of controlled listings accurate across all platforms - **Execution:** Deterministic script - **Hands off to:** Keep the stacking plan current; Market an availability; Build the survey #### Market an availability - **Trigger:** A space comes available or has had no tours in 30 days. - **Ceiling:** New availabilities go out the same week with a flyer, the plan, the rate and a targeted broker email to the people who have shown requirements that fit. Stale space gets a diagnosis instead of another blast: wrong rate, wrong size, or the wrong brokers hearing about it. - **Touches:** - Broker contact lists - Marketing materials - Tour and inquiry history - **Never:** - Never states a rate or a concession the owner has not approved - Never emails a broker list outside the firm's opt-out policy - **Agent:** Complex agent that builds the materials, targets the broker list and writes the stale-space diagnosis. - **Measured by:** Tours generated per available space per month - **Execution:** Complex agent - **Hands off to:** Report to the owner; Refresh availabilities; Publish the market snapshot #### Report to the owner - **Trigger:** The month ends on any agency assignment. - **Ceiling:** The owner gets the same page every month: tours, proposals out, LOIs live, what the competing buildings did on rate and concession, and one recommendation. Owners stop calling to ask what is happening, and the leasing strategy changes on evidence. - **Touches:** - Tour and proposal activity - Competing building data - Stacking plan - **Never:** - Never recommends a rate change without the broker approving the number - Never discloses a prospect's identity before they consent - **Agent:** Complex agent that assembles the report and writes the recommendation with the comps behind it. - **Measured by:** Percent of assignments with a monthly owner report delivered - **Execution:** Complex agent - **Hands off to:** Market an availability; Publish the market snapshot; Keep the stacking plan current ### Market Research Comps, availability and absorption. The reason anyone believes your number. #### Capture a comp - **Trigger:** A lease signs anywhere in the covered market. - **Ceiling:** The deal is in the comp set within a week with the size, term, starting rate, escalations, free rent, TI and the source of each figure. The firm's comp database is better than the paid one because it holds the terms nobody publishes. - **Touches:** - Comp database - Deal records - Public and broker sources - **Never:** - Never records a confidential term the client has not released - Never publishes a comp sourced from a confidential document - **Agent:** Simple agent that extracts terms from deal records and sources, and marks confidence on each field. - **Measured by:** Comps captured per quarter, and percent with verified economics - **Execution:** Simple agent - **Hands off to:** Publish the market snapshot; Build the proposal comparison; Prepare a broker opinion of value #### Publish the market snapshot - **Trigger:** The quarter closes. - **Ceiling:** Availability, asking rates, net absorption, concession trends and the notable deals are out within five business days of quarter end, by submarket and by class. The firm's numbers are in the client's hands before the national report prints. - **Touches:** - Comp database - Availability data - Absorption history - **Never:** - Never publishes a figure without a countable source - Never includes a deal under a confidentiality hold - **Agent:** Complex agent that computes the statistics, writes the narrative and cites each number. - **Measured by:** Business days from quarter end to published snapshot - **Execution:** Complex agent - **Hands off to:** Capture a comp; Watch the submarket turn; Report to the owner #### Watch the submarket turn - **Trigger:** Weekly, across every covered submarket. - **Ceiling:** When a submarket shifts, the firm sees it in weeks rather than at the next quarterly report: three sublets hit at once, concessions stretch past twelve months, a large block comes back. Advice changes on the turn, not after it. - **Touches:** - Availability changes - Sublet listings - Concession data - **Never:** - Never issues a market call to clients without a broker reviewing it - Never attributes a change to a named tenant's plans without confirmation - **Agent:** Complex agent that tracks the leading indicators and writes a short note when one moves. - **Measured by:** Weeks between a submarket inflection and the firm naming it - **Execution:** Complex agent - **Hands off to:** Publish the market snapshot; Capture a comp; Canvass an owner ### Deal Execution LOI to lease to commission, with the dates that decide whether it closes. #### Track the LOI - **Trigger:** A letter of intent goes out or comes back. - **Ceiling:** Every live LOI has a version history, the open points listed, who owes the next response and by when. Nothing sits eleven days because each side thought the other had it. The broker can answer where any deal stands in one sentence. - **Touches:** - LOI versions - Open point list - Party contact record - **Never:** - Never sends a counter without the client approving every term - Never agrees to a term on the client's behalf - Never signs anything - **Agent:** Simple agent that diffs versions, lists open points and chases the party who owes the next move. - **Measured by:** Median days per LOI round trip - **Execution:** Simple agent - **Hands off to:** Chase the deal to signature; Abstract the lease; Build the proposal comparison #### Abstract the lease - **Trigger:** A lease is executed. - **Ceiling:** Within 48 hours the eighty page document becomes a one page abstract: term, commencement, rent schedule, escalations, operating expense treatment, options, notice dates, assignment and restoration. Every critical date is in the calendar. Option deadlines stop being missed by a week. - **Touches:** - Executed lease document - Abstract template - Critical date calendar - **Never:** - Never interprets an ambiguous clause; it flags it for counsel - Never abstracts from an unexecuted draft - **Agent:** Complex agent that extracts the terms, builds the date set and flags every clause it could not read cleanly. - **Measured by:** Percent of executed leases abstracted within 48 hours - **Execution:** Complex agent - **Hands off to:** Keep the stacking plan current; Watch lease expirations; Invoice and track the commission #### Chase the deal to signature - **Trigger:** A deal is in documentation with an open item older than three days. - **Ceiling:** Legal comments, the estoppel, the guaranty, the insurance certificate and the space plan each have an owner and a date, and the broker sees the oldest open item every morning. Deals stop dying in the six weeks between agreed LOI and executed lease. - **Touches:** - Open item list - Party contacts - Document status - **Never:** - Never pushes a client's counsel on a legal position - Never commits to a signing date the parties have not agreed - **Agent:** Simple agent that tracks each open item, nudges the owner and escalates by age. - **Measured by:** Median days from LOI agreement to lease execution - **Execution:** Simple agent - **Hands off to:** Track the LOI; Abstract the lease; Invoice and track the commission #### Invoice and track the commission - **Trigger:** A lease executes. - **Ceiling:** The invoice goes out the same week against the commission agreement, with the calculation shown, the split recorded and the second installment dated to the commencement trigger. No fee ever quietly goes uncollected because it was due nine months later. - **Touches:** - Commission agreement - Executed lease terms - Split records - **Never:** - Never invoices an amount outside the signed commission agreement - Never releases a split payment before the firm collects - **Agent:** Script that computes the fee from the lease terms, issues the invoice and calendars the deferred installment. - **Measured by:** Percent of commissions collected within terms, and aged fee balance - **Execution:** Deterministic script - **Hands off to:** Abstract the lease; Chase the deal to signature ### Coverage Knowing every owner and every lease expiry in your market before anyone calls. #### Canvass an owner - **Trigger:** A property changes hands, or an owner has had no contact in 90 days. - **Ceiling:** Every owner in the coverage area hears from the broker with something about their building: a comp on their street, a competing concession, the tenant two floors up whose lease is expiring. The call comes before the assignment goes out for bid, not after. - **Touches:** - Ownership records - Property transaction data - Comp database - **Never:** - Never contacts an owner represented by another broker under an exclusive - Never quotes a fee or a listing rate - **Agent:** Complex agent that finds the reason, builds the one-page property view and drafts the outreach. - **Measured by:** Coverage contact rate, and new assignments won per quarter - **Execution:** Complex agent - **Hands off to:** Watch lease expirations; Prepare a broker opinion of value; Watch the submarket turn #### Watch lease expirations - **Trigger:** Any tracked lease reaches 24 months from expiry. - **Ceiling:** Every tenant in the market is contacted at the right moment in their cycle, which for a large requirement is two years out and not six months. The pipeline is a calendar rather than a hope, and no requirement is discovered after somebody else was hired. - **Touches:** - Lease abstract dates - Tenant contact records - Coverage territory - **Never:** - Never contacts a tenant the firm represents the landlord against - Never uses a confidential lease term in the outreach - **Agent:** Script that surfaces the expiry list each month and queues the right outreach window per tenant. - **Measured by:** Percent of market expirations contacted in the right window - **Execution:** Deterministic script - **Hands off to:** Canvass an owner; Intake a tenant requirement; Abstract the lease #### Prepare a broker opinion of value - **Trigger:** An owner asks what the building is worth or what it would lease for. - **Ceiling:** The opinion goes back within three business days with the rent roll analysis, the lease comps, the sale comps, the submarket position and the assumptions written out. It reads like something the owner can take to their lender. - **Touches:** - Comp database - Rent roll - Submarket data - **Never:** - Never presents the opinion as a certified appraisal - Never delivers without a broker reviewing every assumption - **Agent:** Complex agent that assembles the comps and the analysis and drafts the opinion for review. - **Measured by:** Business days from request to delivered opinion - **Execution:** Complex agent - **Hands off to:** Canvass an owner; Capture a comp; Publish the market snapshot ## Independent Consulting The practice around the work: pipeline that keeps moving while you deliver, and money that actually gets collected. ### Pipeline Where the next engagement comes from while the current one eats the week. #### Triage an inbound inquiry - **Trigger:** Someone emails asking whether you can help with something. - **Ceiling:** Every inquiry gets a reply the same day with three clarifying questions and a link to a call, and the sender is already researched: what the company does, who they are, what changed recently. The consultant walks into the first call knowing more than the prospect expected. - **Touches:** - The inbox - Public company research - The calendar - **Never:** - Never quotes a rate or a fee - Never commits to a start date or availability - **Agent:** Complex agent that researches the sender, drafts the reply and offers real calendar times. - **Measured by:** Median hours to first substantive reply on inbound - **Execution:** Complex agent - **Hands off to:** Prepare a discovery call; Work the warm list; Check the terms against the floor #### Work the warm list - **Trigger:** A prospect goes 30 days without contact, or their company makes news. - **Ceiling:** The twelve people most likely to hire you next hear from you with something useful, not a check-in. A funding round, a new hire in their function, a regulation that changes their problem. Nobody goes quiet for six months and then hires somebody else. - **Touches:** - Contact records - Public company news - Prior conversation notes - **Never:** - Never sends more than one message per contact per month - Never contacts anyone who asked to stop hearing from you - **Agent:** Complex agent that watches for a real reason, drafts the note and holds it for a human to send. - **Measured by:** Warm conversations reopened per month - **Execution:** Complex agent - **Hands off to:** Triage an inbound inquiry; Catch a network signal; Ask for the referral #### Catch a network signal - **Trigger:** A past client changes jobs, or a former colleague is promoted. - **Ceiling:** You know within a day when somebody who already trusts you lands somewhere with a budget. The congratulation goes out while it is still news. Most independent work comes from these moves and they stop being something you notice a year late. - **Touches:** - Contact records - Public profile changes - Engagement history - **Never:** - Never pitches in the same message as the congratulation - Never scrapes a source whose terms forbid it - **Agent:** Simple agent that watches for role changes across the network and drafts the note. - **Measured by:** Percent of network job changes caught within seven days - **Execution:** Simple agent - **Hands off to:** Work the warm list; Ask for the referral; Triage an inbound inquiry #### Prepare a discovery call - **Trigger:** A first call is booked. - **Ceiling:** Thirty minutes before the call there is a one page brief: the company, the people on the invite, what they said in email, three hypotheses about their real problem and the questions that would confirm or kill each one. The call goes deeper because the first fifteen minutes are not spent on background. - **Touches:** - Public company and people research - Prior correspondence - The calendar invite - **Never:** - Never states a price or a scope in the brief as if agreed - Never uses a paid data source outside its license - **Agent:** Complex agent that assembles the brief and writes the hypotheses with the evidence behind each one. - **Measured by:** Discovery call to proposal request rate - **Execution:** Complex agent - **Hands off to:** Turn call notes into a scope; Triage an inbound inquiry; Draft the proposal ### Scoping and Proposals Turning a conversation into a scope, a price and a signature. #### Turn call notes into a scope - **Trigger:** A discovery or scoping call ends. - **Ceiling:** Within an hour the messy transcript becomes a structured scope: the problem in the client's words, the outcome they described, deliverables, what is explicitly out, assumptions and the dependencies on their side. The thing people hire consultants for stops being lost in a notebook. - **Touches:** - Call recording or notes - Scope templates - Prior similar engagements - **Never:** - Never invents a deliverable the client did not ask for - Never sets the fee or the timeline - **Agent:** Complex agent that structures the notes into a scope draft and flags what was never actually answered. - **Measured by:** Hours from call end to a scope draft the consultant would send - **Execution:** Complex agent - **Hands off to:** Draft the proposal; Prepare a discovery call; Catch scope creep early #### Draft the proposal - **Trigger:** A scope is agreed in principle. - **Ceiling:** The proposal goes out within one business day, written in the client's language, with the outcome first, the approach second and the fee last. It reuses the structure that has closed before and pulls the two most relevant pieces of past work. Nobody loses a deal to a week of silence. - **Touches:** - Scope draft - Proposal templates - Past engagement summaries - **Never:** - Never sets or discounts the fee - Never sends without the consultant reading every line - Never names a past client the engagement terms keep confidential - **Agent:** Complex agent that writes the full draft from the scope and the template, held for review. - **Measured by:** Business days from scope agreement to proposal sent - **Execution:** Complex agent - **Hands off to:** Follow up on an open proposal; Check the terms against the floor; Turn call notes into a scope #### Follow up on an open proposal - **Trigger:** A proposal is five days old with no answer. - **Ceiling:** Every open proposal is followed up on a schedule the consultant would actually keep, with something added each time rather than a nudge. Proposals stop dying of silence on both sides, and a no arrives as a no instead of as nothing. - **Touches:** - Open proposal list - Client contact record - Prior correspondence - **Never:** - Never lowers the price to prompt a decision - Never follows up more than three times without the consultant deciding - **Agent:** Simple agent that tracks the open list and drafts each follow-up with a reason to reply. - **Measured by:** Percent of proposals that get a definite yes or no - **Execution:** Simple agent - **Hands off to:** Draft the proposal; Run the kickoff; Work the warm list #### Check the terms against the floor - **Trigger:** A proposal or contract draft is ready for review. - **Ceiling:** Before anything goes out, payment terms, expenses, IP ownership, kill fee, liability cap and the rate against the practice floor are all checked against what the consultant decided in a calm moment rather than in a negotiation. Net 90 and unlimited revisions never slip through again. - **Touches:** - Practice terms and rate floor - Proposal or contract draft - Prior signed agreements - **Never:** - Never approves terms below the floor - Never gives a legal opinion on a clause - Never signs anything - **Agent:** Simple agent that compares the draft against the standard terms and lists every deviation. - **Measured by:** Engagements signed below the rate floor or outside standard terms - **Execution:** Simple agent - **Hands off to:** Draft the proposal; Run the invoices; Run the kickoff ### Delivery The work itself, and the client knowing it is happening. #### Run the kickoff - **Trigger:** A proposal is signed. - **Ceiling:** Inside two days the client has the schedule, the working agreements, the list of people to interview, the data requests with owners and dates, and the first meeting on the calendar. The first week is spent working, not waiting for a spreadsheet somebody forgot to send. - **Touches:** - Signed scope - Kickoff checklist - Client contact list - **Never:** - Never requests access to a system before the client's security review clears it - Never adds a workstream that is not in the signed scope - **Agent:** Simple agent that builds the kickoff pack from the scope and chases the data requests. - **Measured by:** Business days from signature to first working session - **Execution:** Simple agent - **Hands off to:** Send the weekly client update; Turn a meeting into actions; Catch scope creep early #### Send the weekly client update - **Trigger:** Friday morning on any active engagement. - **Ceiling:** The client gets the same short page every Friday: what moved, what is next, what is blocked and what you need from them by when. Status meetings shrink because the status is already known, and no engagement ends with a client saying they were not sure what was happening. - **Touches:** - Work log and deliverable status - Open blockers and requests - Client contact record - **Never:** - Never reports a deliverable complete before it is finished - Never sends without the consultant reading it - **Agent:** Simple agent that assembles the update from the week's work and writes it plainly. - **Measured by:** Percent of active weeks with an update delivered - **Execution:** Simple agent - **Hands off to:** Turn a meeting into actions; Catch scope creep early; Check a deliverable before it ships #### Turn a meeting into actions - **Trigger:** Any client working session ends. - **Ceiling:** Within an hour there is a summary with decisions, open questions and actions with a name and a date next to each, sent to everyone who was there. Decisions stop being relitigated three weeks later because nobody wrote down what was agreed. - **Touches:** - Meeting recording or notes - Action list - Attendee list - **Never:** - Never assigns an action to a client person without the consultant confirming it - Never records a decision that was not actually made - **Agent:** Simple agent that summarizes, extracts the actions and drafts the recap. - **Measured by:** Percent of client sessions with a recap sent inside two hours - **Execution:** Simple agent - **Hands off to:** Send the weekly client update; Check a deliverable before it ships; Catch scope creep early #### Check a deliverable before it ships - **Trigger:** A deliverable is marked ready to send. - **Ceiling:** Every number traces to a source, every claim has evidence, the client's own terminology is used consistently, and nothing from another client's file survives in a template. The consultant stops being the only quality control at 11pm the night before. - **Touches:** - The draft deliverable - Source data and citations - Scope commitments - **Never:** - Never sends a deliverable to the client - Never changes a finding or a recommendation - Never leaves another client's data in a reused template - **Agent:** Complex agent that checks sources, consistency and scope coverage, and lists what it could not verify. - **Measured by:** Client-reported errors per deliverable - **Execution:** Complex agent - **Hands off to:** Send the weekly client update; Close an engagement properly; Turn a meeting into actions #### Catch scope creep early - **Trigger:** A request arrives that is not in the signed scope. - **Ceiling:** The moment somebody asks for the extra analysis, it is named against the scope with an estimate of the hours and a draft change order. The consultant has the conversation in week two when it is easy instead of in week nine when the engagement is underwater. - **Touches:** - Signed scope - Work log and hours - Client correspondence - **Never:** - Never refuses a client request directly - Never prices a change order - Never agrees to absorb extra work - **Agent:** Complex agent that compares requests against the scope and drafts the change order for a human call. - **Measured by:** Actual hours against scoped hours per engagement - **Execution:** Complex agent - **Hands off to:** Turn call notes into a scope; Send the weekly client update; Capture the time ### Money and Back Office Hours captured, invoices out, cash actually collected. #### Capture the time - **Trigger:** A working day ends. - **Ceiling:** Hours land against the right engagement the same day, reconstructed from calendar, documents and messages, ready for the consultant to correct in two minutes. No more rebuilding a month from memory on the 30th and quietly losing nine billable hours. - **Touches:** - Calendar - Document and file activity - Engagement codes - **Never:** - Never submits time without the consultant confirming it - Never bills an hour it inferred but cannot show a source for - **Agent:** Simple agent that reconstructs the day and presents it for a one-screen confirmation. - **Measured by:** Billable hours captured against hours actually worked - **Execution:** Simple agent - **Hands off to:** Run the invoices; Catch scope creep early; Close an engagement properly #### Run the invoices - **Trigger:** The first of the month, or a milestone completes. - **Ceiling:** Every invoice goes out on the same day each month with the right hours, the right PO number, the agreed expenses and the correct billing contact. Nothing slips a month because the consultant was in a workshop on the 1st. - **Touches:** - Confirmed time and milestones - Engagement billing terms - Invoice templates - **Never:** - Never sends an invoice without the consultant approving the amount - Never bills expenses without a receipt on file - **Agent:** Script that builds the batch from confirmed time and holds it for one release. - **Measured by:** Percent of invoices issued on the scheduled day - **Execution:** Deterministic script - **Hands off to:** Chase what is owed; Capture the time; Check the terms against the floor #### Chase what is owed - **Trigger:** An invoice passes its due date. - **Ceiling:** Late invoices get a polite, escalating sequence that actually runs: a reminder to the contact, then to accounts payable, then a note to the sponsor. The consultant stops choosing between chasing money and doing the work, and average days to payment falls. - **Touches:** - Open invoice list - Client billing contacts - Payment records - **Never:** - Never threatens to stop work or to take legal action - Never contacts the client sponsor before the consultant approves that step - **Agent:** Simple agent that runs the escalation ladder and drafts each message. - **Measured by:** Average days sales outstanding - **Execution:** Simple agent - **Hands off to:** Run the invoices; Close an engagement properly ### Practice and Reputation What the finished work does for the next engagement. #### Close an engagement properly - **Trigger:** The final deliverable is accepted. - **Ceiling:** Every engagement ends the same way: the final invoice, the archive, a written summary of what was done and what it changed, the client's own words on the outcome, and the next conversation named. The work stops disappearing the day it ends. - **Touches:** - Engagement archive - Deliverable history - Client contact record - **Never:** - Never archives client data outside the terms of the agreement - Never publishes anything about the work - **Agent:** Complex agent that assembles the closeout package and drafts the summary. - **Measured by:** Percent of engagements closed with a written outcome summary - **Execution:** Complex agent - **Hands off to:** Ask for the referral; Write up the work; Chase what is owed #### Ask for the referral - **Trigger:** A client says the work went well, or an engagement closes clean. - **Ceiling:** The ask happens in the two weeks when the client is happiest, names one or two specific people or kinds of problem, and makes the introduction easy to forward. Referrals become a steady share of new work instead of an accident. - **Touches:** - Client satisfaction signals - Contact record - Engagement history - **Never:** - Never asks a client with an open dispute or an unpaid invoice - Never names another client as a target without permission - **Agent:** Simple agent that watches for the moment and drafts the ask with a forwardable paragraph. - **Measured by:** Percent of new engagements sourced by referral - **Execution:** Simple agent - **Hands off to:** Close an engagement properly; Work the warm list; Catch a network signal #### Write up the work - **Trigger:** An engagement closes with a measurable outcome. - **Ceiling:** Within a week there is a written account of the problem, the approach and the result, anonymized where it has to be, ready for the consultant to edit and use in the next proposal. The practice accumulates evidence instead of starting each pitch from nothing. - **Touches:** - Engagement archive - Outcome measurements - Client confidentiality terms - **Never:** - Never names a client without written permission - Never publishes anything - Never states a result the engagement record cannot support - **Agent:** Complex agent that drafts the write-up from the archive and flags anything that needs client clearance. - **Measured by:** Usable write-ups produced per closed engagement - **Execution:** Complex agent - **Hands off to:** Close an engagement properly; Draft the proposal; Ask for the referral ## Are you an expert operator? If you’ve mastered an industry and want to turn that expertise into its AI Operating System, we should talk. Our operator partners lend their knowledge, we provide the engineering, and together we build the system that the rest of your industry runs on. --- # Custom AI OS > A complete AI operating system built from scratch for one business, starting with a map of every action it takes. Source: https://goldstarlabs.ai/custom Machine view: https://goldstarlabs.ai/machine/custom This is the full build where we come in and act as your fractional AI team for as long as we need to get this built. We don’t just slap AI agents over a bunch of existing broken systems, we work to make sure the foundational layer and all of your systems that write to and read from the AI OS are in good enough shape to be useful. ## It starts with a map Once the foundation is solid, the fun part begins. Create a map of every action taken inside your business. How does each action impact the other parts? What are the data sources? What is the scope? What is the best possible result from each individual action? What about the business as a whole? How much do you want to be in the loop vs letting the agents execute? What are the second and third order effects of each action? Our custom internal agents are designed to build this map quickly in tandem with our expert team members. This map goes deep. We trace your most complicated custom workflows, preferences, the rules that only exist in your head, and the way you handle weird edge cases. The mapping turns your tacit knowledge, company history, and systems into a useful foundational layer that we can build an AI OS on top of. ## Every action gets a ceiling and a fence Once the mapping is complete, we identify the entitlement value of every action and its set of bounds. In plain English this means what is the best possible outcome of a specific action and what each agent can and cannot do. Once this is done, the mapping is complete. We know every action that can possibly be taken in your business, what an agent working on that action can and cannot do, and what the best possible outcome of every single action is and how it relates to the overall goals. This map makes sure that every agent we build marches in the same direction towards your overall goals. ## Gold Star’s Terraforming Engine This is where we can really show off the engineering firepower. We have a suite of custom built internal agents that are specifically designed to turn those detailed maps into a working reality. We call this our terraforming engine. This engine spins up and coordinates thousands of parallel AI agents to turn your map into reality, run tests to make sure it's secure and actually works, and make sure it ultimately matches the map that we built together during the first phase. ## Once the Terraforming Engine is done, we put it in your hands The Terraforming Engine can write perfect code and run a million tests, but without putting it in your hands and getting your feedback it's impossible to build something that is truly useful to you. This iteration process goes for as many rounds as needed until the end result is a suite of AI agents that genuinely deliver the value to you in a way that you expect. We call this the “evals”. ## Who this is for Owners who want the whole thing. If you’d rather start smaller, please look at our individual components or a pre-built industry-specific OS. The custom AI OS process is both a time and financial investment, with both parties acting as partners. We’re selective about who we take on, because the payoff comes from doing it right. Interested in seeing if this would be a good fit? We offer a complimentary custom AI audit that lets you get an idea of the types of things we’d tackle first, the overall scope of investment, and gives us a chance to get to know each other a little bit before embarking on this endeavor. ## A reference map > Sample data. The company below is fictional, built to show what a finished map holds: 512 doors, a team of 11, 69 actions. It is not a client. Manages 512 doors across Spokane and the Valley from an office above a print shop on Monroe. Single family rentals, duplexes, and eleven small apartment buildings. ## The foundation #### Property management system - **Kind:** System - **What it is:** The system of record where every unit, lease, tenant and move-out date lives. - **Used by:** Process a notice to vacate; Launch a listing; Send the move-in packet; Make a renewal offer; Triage a maintenance request; Read tenant sentiment; Forecast capacity before the next door #### Tenant and property ledgers - **Kind:** Data - **What it is:** What was charged, what was paid, and what is still owed, on every unit and every household. - **Used by:** Post and match incoming rent; Work the delinquency list; Prepare a late notice; Handle a payment plan request; Produce the monthly owner statement; Run the owner draw; Bill back shared utilities; Post a tenant chargeback #### Trust account and bank feed - **Kind:** System - **What it is:** The account rent and deposits land in, and the feed that says when they did. - **Used by:** Post and match incoming rent; Collect the deposit and first month; Reconcile the trust account; Run the owner draw; Run vendor payables; Ask for a reserve top-up; Dispose of a security deposit #### Lease files - **Kind:** Data - **What it is:** Signed leases, addenda, and the clauses each one actually carries. - **Used by:** Process a notice to vacate; Generate the lease; Make a renewal offer; Prepare a lease violation notice; Process a pet or occupant request; Check lease terms against current law; Run the habitability clock #### Lease history and comps - **Kind:** Data - **What it is:** What comparable units actually leased for, and what is asking on the market today. - **Used by:** Price a vacant unit; Launch a listing; Check a stalled listing; Make a renewal offer; Survey the market; Work an inbound owner lead #### Unit photo library - **Kind:** Data - **What it is:** Photos from listings, move-ins, move-outs and finished work, filed against the unit. - **Used by:** Order and check listing photos; Launch a listing; Close a work order and ask the tenant; Scope a turn; Run the move-in condition report; Run the move-out inspection; Drive by the exterior #### Vendor roster - **Kind:** Data - **What it is:** Trades, rates, the on-call rotation, and whose insurance certificate is current. - **Used by:** Dispatch an after-hours emergency; Assign a vendor; Chase an open work order; Run preventive maintenance; Review a vendor invoice; Score a vendor; Block an uninsured vendor; Run vendor payables #### Condition reports - **Kind:** Data - **What it is:** Move-in, interior, move-out and drive-by reports, with the photos that back them. - **Used by:** Run the move-in condition report; Run a mid-lease interior inspection; Run the move-out inspection; Drive by the exterior; Certify smoke and CO detectors; Dispose of a security deposit; Scope a turn #### Owner agreements - **Kind:** Rules - **What it is:** Approval limits, reserve minimums, and what each of the 214 owners wants decided without them. - **Used by:** Request an owner approval; Produce the monthly owner statement; Onboard a new owner and property; Build the year-end owner package; Dispatch an after-hours emergency; Scope a turn; Ask for a reserve top-up; Save a leaving owner #### House rules and limits - **Kind:** Rules - **What it is:** Screening criteria, spend authority, quiet hours, pet terms, and the fee schedule. - **Used by:** Screen an application; Deliver a decline; Dispatch an after-hours emergency; Handle a neighbor complaint; Process a pet or occupant request; Post a tenant chargeback; Check a stalled listing #### Washington law and city code - **Kind:** Rules - **What it is:** State landlord-tenant law and Spokane ordinance, and the clocks each one starts. - **Used by:** Deliver a decline; Arrange entry to an occupied unit; Run the habitability clock; Check lease terms against current law; Keep the fair housing record; Assemble an eviction file; Dispose of a security deposit #### Shared inbox and calendars - **Kind:** System - **What it is:** Where inquiries arrive, and where showings, inspections and vendor visits get their time. - **Used by:** Answer a rental inquiry; Schedule a showing; Handle a showing no-show; Run preventive maintenance; Arrange entry to an occupied unit; Drive by the exterior; Answer an owner question #### Call and message history - **Kind:** Data - **What it is:** Every conversation with a tenant, an owner or a vendor, written down where the next person can find it. - **Used by:** Answer a rental inquiry; Triage a maintenance request; Handle a neighbor complaint; Read tenant sentiment; Answer an owner question; Spot an owner about to leave; Keep the fair housing record; Request and answer reviews #### Work order queue - **Kind:** System - **What it is:** Every request, who is holding it, and everything already done to that unit. - **Used by:** Triage a maintenance request; Assign a vendor; Chase an open work order; Close a work order and ask the tenant; Check warranty and home coverage; Review a vendor invoice; Run the habitability clock; Post a tenant chargeback ## The actions ### Leasing From a notice to vacate to a signed lease and a set of keys. #### Process a notice to vacate - **Trigger:** A tenant sends notice that they are leaving. - **Ceiling:** The notice is read the hour it arrives, checked against the lease for the required days, and the move-out date is confirmed in writing the same day. The unit is on the market before the tenant has finished packing. Nobody discovers the vacancy from a rent roll two weeks later. - **Touches:** - The lease document - The tenant's move-out date field - The vacancy calendar - **Never:** - Never negotiates an early release or waives notice days - Never discusses the deposit or what will be withheld - **Agent:** Simple agent that reads the notice, computes the required dates, and writes the confirmation for a human to send. - **Measured by:** Hours from notice received to confirmed move-out date - **Execution:** Simple agent - **Hands off to:** Launch a listing; Scope a turn; Warn an owner about a vacancy #### Price a vacant unit - **Trigger:** A unit is 30 days from available and has no price set. - **Ceiling:** Rent is set from what comparable units actually leased for last month, not what they were asked. The number comes with the three closest comps and a line on what it means for days vacant. Off by twenty dollars a month at worst, and the owner sees the reasoning before anyone argues about it. - **Touches:** - Closed lease history - Public listing comps - The unit's condition notes - **Never:** - Never publishes a price without the portfolio manager signing off - Never prices below the owner's agreed floor - **Agent:** Complex agent that pulls comps, adjusts for condition and season, and writes a short recommendation with sources. - **Measured by:** Median days on market against the Spokane submarket average - **Execution:** Complex agent - **Hands off to:** Launch a listing; Check a stalled listing; Survey the market #### Launch a listing - **Trigger:** A unit has a price and a confirmed available date. - **Ceiling:** Photos, description, amenities and pet rules go live on every syndicated site within an hour of the price being approved, with the copy written for that specific house and not a template. No listing goes up with last tenant's furniture in the photos. - **Touches:** - Listing syndication accounts - The unit photo library - Availability dates - **Never:** - Never invents amenities the unit does not have - Never uses language that touches a protected class - **Agent:** Simple agent that assembles and posts the listing, with fair housing language checked before publish. - **Measured by:** Hours from price approval to live listing - **Execution:** Simple agent - **Hands off to:** Answer a rental inquiry; Order and check listing photos; Check a stalled listing #### Order and check listing photos - **Trigger:** A turn is scheduled to finish within five days. - **Ceiling:** The photographer is booked the day the turn scope is set, shoots the morning after the clean, and every listing goes up with twenty-two current photos and a floor plan. The old photos are archived, not reused. - **Touches:** - The photographer's calendar - The unit photo library - Turn completion dates - **Never:** - Never approves a photo invoice over $250 - Never publishes photos before the final clean is signed off - **Agent:** Script that books the slot from the turn date and flags listings whose photos are older than the last tenancy. - **Measured by:** Percent of live listings with photos from the current turn - **Execution:** Deterministic script - **Hands off to:** Launch a listing; Scope a turn #### Check a stalled listing - **Trigger:** A vacant unit reaches 10 days with no application. - **Ceiling:** On day ten the listing gets a real diagnosis: how many views, how many inquiries, how many showings, and where it sits against the four units competing with it right now. Either the price moves, the photos change, or the pet rule is the problem, and the owner hears which within the day. A unit never sits at the wrong price for six weeks. - **Touches:** - Listing performance data - Competing active listings - Showing and inquiry counts - **Never:** - Never changes the published price on its own - Never offers a concession the owner has not agreed to - **Agent:** Complex agent diagnosing the funnel stage that is failing and recommending one change. - **Measured by:** Percent of vacancies leased within 21 days of listing - **Execution:** Complex agent - **Hands off to:** Price a vacant unit; Request an owner approval; Order and check listing photos; Survey the market #### Answer a rental inquiry - **Trigger:** An inquiry arrives from a listing site, often at 11pm. - **Ceiling:** Every inquiry gets a real answer in under five minutes, day or night, with the pet policy, the income requirement and two open showing times. Nobody waits until Monday. The ones who do not qualify find out kindly and immediately instead of after an application fee. - **Touches:** - The inquiry inbox - Showing calendar availability - Published screening criteria - **Never:** - Never makes an exception to the income or pet rules - Never answers a question about whether the neighborhood is safe or who lives nearby - **Agent:** Complex agent handling the back and forth in plain language, escalating anything outside the published criteria. - **Measured by:** Median first response time to inquiries - **Execution:** Complex agent - **Hands off to:** Schedule a showing; Screen an application; Launch a listing #### Schedule a showing - **Trigger:** A qualified prospect asks to see the unit. - **Ceiling:** The prospect picks a time from real availability, gets a confirmation with the lockbox process and the address, and a reminder two hours out. Showings cluster so the leasing agent drives one loop instead of four. - **Touches:** - Showing calendar - Lockbox codes for vacant units - Prospect contact record - **Never:** - Never issues an access code for an occupied unit - Never books a showing without an identity check on file - **Agent:** Simple agent that books, confirms, reminds and groups showings by geography. - **Measured by:** Showings per leasing agent hour - **Execution:** Simple agent - **Hands off to:** Handle a showing no-show; Screen an application #### Handle a showing no-show - **Trigger:** A scheduled showing passes with no arrival. - **Ceiling:** The agent knows within ten minutes, not after standing in an empty living room for half an hour. The prospect gets one polite message offering to rebook. Repeat no-shows stop getting confirmed slots and get self-showing access instead. - **Touches:** - Showing calendar - Prospect contact record - No-show count on the prospect - **Never:** - Never charges a fee or blacklists a prospect without a human deciding - Never sends more than one rebooking message - **Agent:** Script that watches for the check-in, marks the no-show and sends the single rebooking note. - **Measured by:** No-show rate per booked showing - **Execution:** Deterministic script - **Hands off to:** Schedule a showing; Answer a rental inquiry #### Screen an application - **Trigger:** An application is submitted with fees paid. - **Ceiling:** Income, credit, rental history and the eviction search are assembled into one file within four hours, every applicant measured against the same written criteria in the same order. The decision is defensible a year later because the reasoning is written down. No applicant waits three days to hear no. - **Touches:** - Screening vendor reports - Published criteria - Prior landlord references - **Never:** - Never issues the approval or denial itself - Never weighs anything outside the published criteria - Never touches the adverse action letter without a human reading it - **Agent:** Complex agent that gathers reports, applies the criteria line by line, and hands a recommendation with evidence to a human. - **Measured by:** Hours from completed application to decision delivered - **Execution:** Complex agent - **Hands off to:** Generate the lease; Deliver a decline; Keep the fair housing record #### Deliver a decline - **Trigger:** A human marks an application denied. - **Ceiling:** The adverse action notice goes out the same day with the correct disclosures and the reason drawn from the written criteria. The applicant knows what happened. The file holds the notice, the report and the criteria version used. - **Touches:** - The adverse action template - The screening report reference - The applicant record - **Never:** - Never sends without a named human approving - Never states a reason that was not in the recorded decision - **Agent:** Script that assembles the compliant notice and holds it for one-click human release. - **Measured by:** Percent of declines with a compliant notice sent within 24 hours - **Execution:** Deterministic script - **Hands off to:** Screen an application; Keep the fair housing record #### Generate the lease - **Trigger:** An application is approved and the term is agreed. - **Ceiling:** The lease is built from the current Washington template with the right addenda for pets, smoke detectors, lead paint and the specific property rules, dated and sent for signature within an hour. Nobody hand-edits a Word file from 2021 again. - **Touches:** - The lease template library - Approved terms and rent - E-signature system - **Never:** - Never changes a clause or a term without counsel-approved language - Never countersigns on behalf of the owner - **Agent:** Simple agent assembling the document set from the approved terms and routing for signature. - **Measured by:** Hours from approval to lease sent - **Execution:** Simple agent - **Hands off to:** Send the move-in packet; Collect the deposit and first month; Run the move-in condition report #### Send the move-in packet - **Trigger:** A lease is fully signed. - **Ceiling:** The tenant gets keys, utility transfer deadlines, the portal login, the maintenance number and the renters insurance requirement in one message that they actually read. Nothing is discovered on move-in day. - **Touches:** - Tenant portal accounts - Utility provider contact list - The signed lease - **Never:** - Never releases keys before the deposit and first month clear - Never sets up the utility account in the tenant's name for them - **Agent:** Script that fires the packet and tracks which items are confirmed back. - **Measured by:** Percent of move-ins with utilities and insurance confirmed before key release - **Execution:** Deterministic script - **Hands off to:** Track required insurance; Run the move-in condition report; Collect the deposit and first month #### Make a renewal offer - **Trigger:** A lease reaches 90 days from expiry. - **Ceiling:** Every tenant hears about renewal at day 90, with a rate set from the market and the tenant's payment and care record, and the owner's position already known. Good tenants get a reason to stay. The company never finds out about a move-out at day 20. - **Touches:** - Lease expiry dates - Payment history - Market rent comps - Owner renewal preferences - **Never:** - Never sets an increase above the owner-agreed cap - Never sends the offer before a human reviews the number - Never contacts a tenant who is in an active eviction - **Agent:** Complex agent that builds the offer, the rationale and the owner note, then waits for approval. - **Measured by:** Renewal rate and percent of expiries addressed by day 90 - **Execution:** Complex agent - **Hands off to:** Check lease terms against current law; Read tenant sentiment; Process a notice to vacate; Survey the market ### Maintenance Every request, every vendor, every invoice, from the first call to the closed job. #### Triage a maintenance request - **Trigger:** A tenant reports a leak under the kitchen sink. - **Ceiling:** Within minutes the request has a category, an urgency, the unit's appliance and plumbing history attached, three clarifying questions answered and two photos. A dispatcher never has to call the tenant back to ask what color the water was. - **Touches:** - The request inbox - Unit maintenance history - Tenant contact record - **Never:** - Never promises a visit time - Never tells a tenant a repair is their financial responsibility - **Agent:** Complex agent that interviews the tenant briefly, classifies, and attaches history. - **Measured by:** Percent of work orders dispatched without a clarifying callback - **Execution:** Complex agent - **Hands off to:** Dispatch an after-hours emergency; Assign a vendor; Check warranty and home coverage; Run the habitability clock #### Dispatch an after-hours emergency - **Trigger:** A call comes in at 1am with water actively running. - **Ceiling:** The tenant is told how to shut off the water in under two minutes, the on-call plumber is reached and confirmed, and the owner has a message waiting when they wake up. Nobody sits in a voicemail queue while a ceiling comes down. - **Touches:** - On-call vendor rotation - Emergency spend authority up to $1,500 - Owner notification list - **Never:** - Never authorizes above $1,500 without the portfolio manager - Never dispatches to a unit with an active safety flag without a human - **Agent:** Complex agent handling the live call, the shutoff instructions and the vendor confirmation. - **Measured by:** Median minutes from emergency call to confirmed vendor en route - **Execution:** Complex agent - **Hands off to:** Assign a vendor; Request an owner approval; Run the habitability clock; Open a property insurance claim #### Assign a vendor - **Trigger:** A triaged work order needs a trade. - **Ceiling:** The job goes to the vendor with the right license, current insurance, capacity this week and the best record on that trade in that zip code. Scheduling happens directly with the tenant. No work order sits three days waiting for someone to remember who does drywall on the South Hill. - **Touches:** - Vendor roster and insurance status - Work order details - Tenant availability - **Never:** - Never assigns a vendor with an expired certificate of insurance - Never assigns work above the owner's standing approval limit - **Agent:** Simple agent matching trade, coverage, insurance status and load. - **Measured by:** Median hours from triage to vendor accepted - **Execution:** Simple agent - **Hands off to:** Chase an open work order; Block an uninsured vendor; Review a vendor invoice; Arrange entry to an occupied unit #### Arrange entry to an occupied unit - **Trigger:** A vendor needs inside access to a lived-in home. - **Ceiling:** Proper written notice goes out with the hours, the tenant confirms, and the vendor knows about the dog. The entry is logged. No tenant comes home to a stranger in their kitchen and no vendor drives across town to a locked door. - **Touches:** - Notice of entry templates - Tenant confirmations - Pet and access notes - **Never:** - Never shortens the state-required notice period - Never releases a key or code to an unverified vendor - **Agent:** Script that issues compliant notice, collects confirmation and logs the entry window. - **Measured by:** Percent of entries with documented compliant notice - **Execution:** Deterministic script - **Hands off to:** Assign a vendor; Chase an open work order #### Chase an open work order - **Trigger:** A work order has no update for five days. - **Ceiling:** Nothing goes quiet. The vendor gets a specific question about a specific job, the tenant gets an honest status even when the answer is that the part is backordered, and the manager sees the five oldest jobs every morning. Average age of open work orders stops drifting. - **Touches:** - Open work order queue - Vendor contacts - Tenant status updates - **Never:** - Never gives a completion date the vendor has not confirmed - Never reassigns a job without a human decision - **Agent:** Simple agent that nudges, records replies and escalates by age. - **Measured by:** Average age of open work orders - **Execution:** Simple agent - **Hands off to:** Assign a vendor; Close a work order and ask the tenant; Score a vendor #### Check warranty and home coverage - **Trigger:** An appliance or system fails. - **Ceiling:** Before anyone pays a vendor, the file is checked for the install date, the manufacturer warranty and the owner's home warranty policy. Covered repairs get filed the same day. The company stops spending owner money on a four-year-old water heater that was still under warranty. - **Touches:** - Appliance install records - Owner warranty policies - Vendor invoices pending - **Never:** - Never files a claim in the owner's name without their standing authorization on file - Never delays an emergency repair waiting for a warranty answer - **Agent:** Simple agent that matches the failure to coverage records and flags a claim path. - **Measured by:** Dollars of repair cost recovered through warranty per quarter - **Execution:** Simple agent - **Hands off to:** Triage a maintenance request; Review a vendor invoice; Request an owner approval #### Run preventive maintenance - **Trigger:** The season turns, or a unit hits its service interval. - **Ceiling:** Furnace service before October, gutters after leaf fall, filters quarterly, sump pumps tested before the spring melt, all batched by neighborhood so one vendor does eleven houses in a day. Emergency calls in January drop because the furnace was looked at in September. - **Touches:** - Property service intervals - Vendor batch scheduling - Owner preventive budgets - **Never:** - Never schedules preventive work the owner has declined in writing - Never exceeds the standing annual preventive budget per door - **Agent:** Script driven by service intervals and season, batching by geography. - **Measured by:** Emergency work orders per door per winter - **Execution:** Deterministic script - **Hands off to:** Assign a vendor; Request an owner approval; Drive by the exterior #### Review a vendor invoice - **Trigger:** A vendor invoice arrives against a work order. - **Ceiling:** Every invoice is matched to the scope, the approval, the photos and what that vendor charged for the same job last time. Line items that drifted get questioned before payment, not at the annual review. Owners never see a charge that nobody checked. - **Touches:** - Work order scope and approvals - Vendor rate history - Completion photos - **Never:** - Never releases payment - Never approves an overage above the authorized amount - Never disputes a charge with a vendor directly - **Agent:** Complex agent comparing invoice to scope and history, writing a short exception note. - **Measured by:** Percent of invoices with a scope and rate match before payment - **Execution:** Complex agent - **Hands off to:** Run vendor payables; Score a vendor; Produce the monthly owner statement #### Close a work order and ask the tenant - **Trigger:** A vendor marks a job complete. - **Ceiling:** The tenant confirms the fix actually worked before the job closes, in one question they can answer with a tap. Photos are filed to the unit history. Callbacks on the same problem within thirty days are visible by vendor and by trade. - **Touches:** - Work order status - Tenant confirmation replies - Unit maintenance history - **Never:** - Never closes a job the tenant says is unresolved - Never asks a satisfaction question during an active dispute or eviction - **Agent:** Script that requests confirmation, files photos and reopens on a negative reply. - **Measured by:** Thirty-day callback rate per completed work order - **Execution:** Deterministic script - **Hands off to:** Score a vendor; Read tenant sentiment; Chase an open work order #### Score a vendor - **Trigger:** The month closes. - **Ceiling:** Every vendor has a live record of response time, callback rate, price against the trade median and what tenants said. The three best get more work. The worst one gets a conversation with evidence instead of a vague feeling. - **Touches:** - Work order timings - Invoice history - Tenant confirmations - **Never:** - Never removes a vendor from the roster - Never shares one vendor's pricing with another - **Agent:** Script computing the monthly table, with outliers called out. - **Measured by:** Share of work orders going to top-quartile vendors - **Execution:** Deterministic script - **Hands off to:** Assign a vendor; Review a vendor invoice; Block an uninsured vendor #### Scope a turn - **Trigger:** Keys come back on a vacated unit. - **Ceiling:** Within 24 hours the unit has a walkthrough, a photo set, a line-item scope split into owner cost and tenant damage, and a vendor schedule that puts it rent-ready in nine days. The owner sees the scope and the cost before work starts, with the move-in photos next to the move-out ones. - **Touches:** - Move-in and move-out condition reports - Vendor scheduling - Owner turn budget - **Never:** - Never assigns damage cost to a tenant without a human reviewing the photos - Never starts work above the owner's turn budget without approval - **Agent:** Complex agent building the scope from paired condition reports and prior turn costs. - **Measured by:** Median days from keys returned to rent-ready - **Execution:** Complex agent - **Hands off to:** Dispose of a security deposit; Order and check listing photos; Request an owner approval; Run the move-out inspection ### Tenant Relations The daily contact with 512 households, money and complaints included. #### Send the rent reminder - **Trigger:** The 28th of the month arrives. - **Ceiling:** Every tenant gets a reminder in the channel they answer, with their exact balance and a payment link that works on a phone. Tenants who always pay on time get one note, not four. On-time payment rises without a single phone call. - **Touches:** - Ledger balances - Tenant contact preferences - Payment links - **Never:** - Never contacts a tenant on a payment plan outside its agreed schedule - Never threatens a fee or a filing - **Agent:** Script that sends balance-accurate reminders on the schedule. - **Measured by:** Percent of rent collected by the 3rd - **Execution:** Deterministic script - **Hands off to:** Prepare a late notice; Work the delinquency list; Handle a payment plan request #### Prepare a late notice - **Trigger:** Rent is unpaid after the grace period ends. - **Ceiling:** The correct Washington notice is prepared with the exact amount owed, the right dates and the right service method, ready for a human to send the same morning the grace period closes. Nothing is ever posted with last month's balance on it. - **Touches:** - Ledger balances - Notice templates - Service records - **Never:** - Never serves a notice without a named human releasing it - Never contacts a tenant about money owed outside permitted hours - Never adds a fee not written in the lease - **Agent:** Simple agent that builds the compliant notice and holds it for release. - **Measured by:** Percent of late notices served on the first permitted day with a correct balance - **Execution:** Simple agent - **Hands off to:** Handle a payment plan request; Assemble an eviction file; Work the delinquency list #### Handle a payment plan request - **Trigger:** A tenant writes to say they will be short this month. - **Ceiling:** The tenant gets a same-day human reply, their history is already summarized for whoever answers, and a plan inside policy is drafted in minutes. Tenants who were going to pay eventually do not become filings because nobody wrote back. - **Touches:** - Payment history - Plan policy limits - Ledger balance - **Never:** - Never agrees to a plan; a human signs every one - Never waives rent, fees or the notice timeline - **Agent:** Simple agent that assembles the history and drafts a policy-compliant plan for a person to offer. - **Measured by:** Percent of payment plans completed as agreed - **Execution:** Simple agent - **Hands off to:** Prepare a late notice; Work the delinquency list; Read tenant sentiment #### Handle a neighbor complaint - **Trigger:** A tenant reports noise from the unit upstairs at midnight. - **Ceiling:** The complaint is logged with time, date and specifics while it is fresh, the pattern across the building is visible, and the lease clause that applies is already pulled. When a warning is finally needed, it is backed by six dated entries instead of one angry phone call. - **Touches:** - Complaint log - Lease quiet-hours clauses - Building complaint history - **Never:** - Never tells one tenant who complained about them - Never issues a lease violation notice - Never mediates a dispute between neighbors - **Agent:** Simple agent that records, pattern-matches and escalates to a human with the file assembled. - **Measured by:** Repeat complaints per building per quarter - **Execution:** Simple agent - **Hands off to:** Prepare a lease violation notice; Read tenant sentiment #### Prepare a lease violation notice - **Trigger:** A documented pattern crosses the lease threshold. - **Ceiling:** The notice cites the exact clause, lists the dated incidents, states what must change and by when, and is ready for review within an hour of the threshold being crossed. It reads like a record, not a scolding. - **Touches:** - Complaint and inspection logs - The lease clauses - Notice templates - **Never:** - Never sends without a human approving - Never asserts a violation without dated evidence in the file - **Agent:** Simple agent that drafts from the evidence log and holds for approval. - **Measured by:** Percent of violations cured without a second notice - **Execution:** Simple agent - **Hands off to:** Handle a neighbor complaint; Assemble an eviction file; Run a mid-lease interior inspection #### Process a pet or occupant request - **Trigger:** A tenant asks to add a dog, or a partner moves in. - **Ceiling:** The request is checked against the lease, the owner's pet policy, breed rules the insurer cares about and the occupancy limit, and answered within a day with the fee, the deposit and the addendum ready. Assistance animal requests are routed to a person immediately and never treated as pets. - **Touches:** - Lease occupancy and pet terms - Owner pet policy - Addendum templates - **Never:** - Never decides an assistance or service animal request - Never approves an addition; a human signs - Never asks about a disability or requests medical records - **Agent:** Simple agent that checks policy and prepares the addendum and fee schedule. - **Measured by:** Percent of unauthorized pets or occupants found at inspection - **Execution:** Simple agent - **Hands off to:** Run a mid-lease interior inspection; Track required insurance; Prepare a lease violation notice #### Handle a lockout - **Trigger:** A tenant is locked out at 9pm on a Sunday. - **Ceiling:** Identity is verified against the lease, the tenant is told the after-hours locksmith fee before anything is dispatched, and someone is on the way in fifteen minutes. The charge lands on the ledger the same night with the tenant's agreement recorded. - **Touches:** - Tenant identity verification - Locksmith rotation - Tenant chargeable fee schedule - **Never:** - Never gives access without verifying identity against the lease - Never dispatches without the tenant agreeing to the fee - **Agent:** Simple agent verifying identity, quoting the fee and dispatching. - **Measured by:** Median minutes from lockout call to locksmith dispatched - **Execution:** Simple agent - **Hands off to:** Post a tenant chargeback; Arrange entry to an occupied unit #### Read tenant sentiment - **Trigger:** A lease crosses month six, or a tenant sends a third unhappy message. - **Ceiling:** Long before a notice to vacate arrives, the company knows which twenty households are unhappy and why, drawn from work order tone, response times and payment patterns. A property manager calls the top five each month. Turnover becomes something you see coming. - **Touches:** - Work order history and tone - Payment record - Complaint log - Renewal status - **Never:** - Never contacts a tenant directly - Never surfaces sentiment about a tenant to an owner by name - Never scores a household on anything outside its own tenancy record - **Agent:** Complex agent producing a weekly short list with the reason and the evidence. - **Measured by:** Voluntary turnover rate against prior year - **Execution:** Complex agent - **Hands off to:** Make a renewal offer; Close a work order and ask the tenant; Handle a payment plan request; Request and answer reviews #### Post a tenant chargeback - **Trigger:** A cost is incurred that the lease makes the tenant's responsibility. - **Ceiling:** The charge is posted with the invoice, the photo and the lease clause attached, and the tenant sees all three at once. Disputes drop because the reason is visible before anyone asks. - **Touches:** - Tenant ledger - Vendor invoices - Lease responsibility clauses - **Never:** - Never posts a charge above $300 without a human approving - Never assigns normal wear and tear to a tenant - **Agent:** Script that posts the charge with its evidence bundle attached. - **Measured by:** Percent of tenant charges disputed - **Execution:** Deterministic script - **Hands off to:** Handle a lockout; Review a vendor invoice; Dispose of a security deposit ### Owner Relations Reporting, approvals and trust with the 214 people who own the doors. #### Produce the monthly owner statement - **Trigger:** The books close on the last business day of the month. - **Ceiling:** Every owner gets a statement on the same day with income, expenses, each invoice attached and a two-line plain English note about what happened at their property. The owners with three questions a month have zero, because the answers are already on the page. - **Touches:** - Property ledgers - Vendor invoices - Owner contact preferences - **Never:** - Never sends a statement before the trust reconciliation passes - Never adjusts a ledger entry to make a statement balance - **Agent:** Complex agent that assembles the statement and writes the narrative note per property. - **Measured by:** Percent of statements delivered by the 5th with zero corrections after - **Execution:** Complex agent - **Hands off to:** Run the owner draw; Reconcile the trust account; Answer an owner question; Bill back shared utilities #### Request an owner approval - **Trigger:** A repair estimate exceeds the owner's standing limit. - **Ceiling:** The owner gets one message with the photo, the estimate, a second bid where it matters, what happens if they wait, and a yes or no button. Median approval time goes from four days to four hours, and no tenant sits with a broken furnace because an email went unanswered. - **Touches:** - Estimates and photos - Owner approval limits - Owner contact record - **Never:** - Never proceeds on silence - Never approves spend on the owner's behalf - Never delays a habitability repair waiting for an answer - **Agent:** Simple agent that packages the decision and chases it on a schedule. - **Measured by:** Median hours to owner decision on approval requests - **Execution:** Simple agent - **Hands off to:** Dispatch an after-hours emergency; Scope a turn; Run the habitability clock; Produce the monthly owner statement #### Warn an owner about a vacancy - **Trigger:** A notice to vacate is confirmed on their property. - **Ceiling:** The owner hears it from Tamarack within the hour, with the expected vacant days, the likely turn cost from the last two turns on that house, and the new asking rent. They never learn about a vacancy from a smaller deposit hitting their account. - **Touches:** - Notice records - Historical turn costs - Market rent estimate - **Never:** - Never shares the tenant's stated reason for leaving - Never commits to a rent number before pricing is approved - **Agent:** Simple agent that drafts the heads-up with numbers drawn from that property's history. - **Measured by:** Percent of vacancies where the owner was told within 24 hours - **Execution:** Simple agent - **Hands off to:** Process a notice to vacate; Price a vacant unit; Scope a turn #### Answer an owner question - **Trigger:** An owner emails asking why the plumbing charge was $740. - **Ceiling:** The answer comes back the same day with the invoice, the work order, the photos and the approval trail in one reply. The owner stops feeling like they have to audit their own property manager. - **Touches:** - Ledgers and invoices - Work order history - Approval records - **Never:** - Never sends a reply about a dispute or a legal matter without a human - Never speculates about a charge it cannot source - **Agent:** Complex agent that finds the evidence and drafts the reply for review. - **Measured by:** Median hours to first substantive answer on owner questions - **Execution:** Complex agent - **Hands off to:** Produce the monthly owner statement; Review a vendor invoice; Spot an owner about to leave #### Onboard a new owner and property - **Trigger:** A management agreement is signed. - **Ceiling:** Within five business days the property has W-9 and insurance on file, banking set up, keys collected, a condition report, existing lease and tenant records imported, and the tenant introduced to the new manager. Nothing is discovered three months in. - **Touches:** - Onboarding checklist - Banking and tax forms - Existing lease documents - **Never:** - Never sets up a payout account without verified banking details - Never contacts an inherited tenant before the transfer letter is approved - **Agent:** Complex agent running the checklist, chasing documents and importing records. - **Measured by:** Business days from signed agreement to fully onboarded property - **Execution:** Complex agent - **Hands off to:** Produce the monthly owner statement; Run the move-in condition report; Track required insurance; Forecast capacity before the next door #### Spot an owner about to leave - **Trigger:** An owner asks about their termination clause, or a property loses money two quarters running. - **Ceiling:** The company knows which fifteen owners are at risk and why, weeks before a cancellation letter. The cause is on the page: three slow approvals, a long vacancy, two unanswered emails. Someone calls them before they call a competitor. - **Touches:** - Owner communication history - Property performance - Approval response times - **Never:** - Never contacts an owner directly - Never offers a fee discount - **Agent:** Complex agent scoring risk from behavior and performance, with the reason stated. - **Measured by:** Owner retention rate by quarter - **Execution:** Complex agent - **Hands off to:** Save a leaving owner; Answer an owner question; Produce the monthly owner statement; Ask for a referral #### Build the year-end owner package - **Trigger:** December 31 closes. - **Ceiling:** Every owner has their 1099, the annual income and expense summary and every invoice as one downloadable file by January 20, and vendor 1099s go out correctly the same week. The accountant's questions arrive answered. - **Touches:** - Annual ledgers - Vendor payment totals - Owner and vendor tax forms on file - **Never:** - Never files with the IRS without the controller's sign-off - Never issues a form against an unverified tax ID - **Agent:** Script that assembles the packages and flags missing or mismatched tax forms. - **Measured by:** Percent of owner tax packages delivered by January 20 with no corrections - **Execution:** Deterministic script - **Hands off to:** Produce the monthly owner statement; Reconcile the trust account; Run vendor payables ### Accounting Rent in, owners paid, deposits handled, trust account clean. #### Post and match incoming rent - **Trigger:** Payments land in the trust account. - **Ceiling:** Every payment is matched to the right ledger within the hour, including the money order with the unit number written wrong and the tenant who paid from their mother's account. Unapplied cash never sits for a week. - **Touches:** - Bank feed - Tenant ledgers - Payment references - **Never:** - Never moves money between ledgers - Never applies a payment it cannot match with confidence; those go to a human queue - **Agent:** Simple agent matching deposits to ledgers with a review queue for ambiguity. - **Measured by:** Dollars of unapplied cash older than 48 hours - **Execution:** Simple agent - **Hands off to:** Work the delinquency list; Reconcile the trust account; Run the owner draw; Send the rent reminder #### Collect the deposit and first month - **Trigger:** A lease is signed and the move-in date is set. - **Ceiling:** Certified funds clear before keys move, the deposit lands in the trust account coded to the right unit the same hour, and the tenant knows the exact amount and deadline from the moment they sign. No tenant moves in owing a partial deposit that takes four months to collect. - **Touches:** - Trust account deposits - Tenant ledger - Lease payment terms - **Never:** - Never releases keys or codes - Never accepts a partial amount without a human approving it in writing - Never commingles a deposit with operating funds - **Agent:** Script tracking the required funds, confirming clearance and blocking key release until it passes. - **Measured by:** Percent of move-ins with funds fully cleared before keys - **Execution:** Deterministic script - **Hands off to:** Send the move-in packet; Reconcile the trust account; Post and match incoming rent #### Work the delinquency list - **Trigger:** Every morning at 7am. - **Ceiling:** One list, sorted by risk rather than by dollars, with each household's history, plan status and next required action already attached. The collections person starts making calls at 7:05 instead of building a spreadsheet until 9. - **Touches:** - Tenant ledgers - Payment plan status - Notice history - **Never:** - Never contacts a tenant - Never recommends a filing; that is a human judgment - **Agent:** Script that produces the ranked morning list with the file attached to each row. - **Measured by:** Delinquency over 30 days as a percent of monthly rent roll - **Execution:** Deterministic script - **Hands off to:** Prepare a late notice; Handle a payment plan request; Assemble an eviction file; Post and match incoming rent #### Run the owner draw - **Trigger:** The 10th of the month, after reconciliation passes. - **Ceiling:** Every owner is paid the right amount on the same day every month, with reserves held to the agreed minimum and any property short of cash flagged before the run rather than after a bounced payment. - **Touches:** - Property ledgers - Owner reserve minimums - Payout accounts - **Never:** - Never initiates a transfer; the controller releases the batch - Never draws a property below its reserve minimum - **Agent:** Script that computes the batch, holds exceptions and waits for one human release. - **Measured by:** Percent of owner draws paid on the 10th with no reversals - **Execution:** Deterministic script - **Hands off to:** Reconcile the trust account; Produce the monthly owner statement; Ask for a reserve top-up #### Reconcile the trust account - **Trigger:** Daily at close, and at month end. - **Ceiling:** Three-way reconciliation runs every day, not every month, so a $14 variance is found the morning it happens with the transaction already identified. The state audit is a morning, not a month. - **Touches:** - Bank statements - Property ledgers - Trust liability schedule - **Never:** - Never makes an adjusting entry - Never closes a period with an unexplained variance - **Agent:** Script running the three-way match and reporting variances with candidate causes. - **Measured by:** Days with an unexplained trust variance - **Execution:** Deterministic script - **Hands off to:** Run the owner draw; Produce the monthly owner statement; Post and match incoming rent; Dispose of a security deposit #### Run vendor payables - **Trigger:** Thursday, for invoices approved by Wednesday night. - **Ceiling:** Approved invoices are paid on terms every week, coded to the right property and category, with no duplicate payments and no vendor calling about a check from August. Vendors quote Tamarack better prices because they get paid on time. - **Touches:** - Approved invoice queue - Vendor payment records - Property expense coding - **Never:** - Never pays an invoice lacking a matching approval - Never releases funds; the controller signs the batch - **Agent:** Script that builds the batch, checks for duplicates and holds for release. - **Measured by:** Percent of vendor invoices paid within terms - **Execution:** Deterministic script - **Hands off to:** Review a vendor invoice; Produce the monthly owner statement; Build the year-end owner package #### Dispose of a security deposit - **Trigger:** A tenant moves out and keys are returned. - **Ceiling:** The statement goes out inside the state deadline every time, itemized against paired move-in and move-out photos, with normal wear and tear separated from damage and an invoice behind every dollar. The deposit dispute rate approaches zero because the reasoning is visible. - **Touches:** - Move-in and move-out condition reports - Turn invoices - Deposit ledger - **Never:** - Never sends the statement without a human reviewing every deduction - Never charges for wear and tear or for work with no invoice - **Agent:** Complex agent building the itemized statement from paired evidence, held for review. - **Measured by:** Percent of dispositions sent within the statutory deadline, and dispute rate - **Execution:** Complex agent - **Hands off to:** Run the move-out inspection; Scope a turn; Reconcile the trust account; Post a tenant chargeback #### Bill back shared utilities - **Trigger:** A master-metered building's water bill arrives. - **Ceiling:** Each unit's share is computed on the agreed formula the day the bill arrives, posted to the ledgers, and shown to tenants with the building total and the method. No month is skipped and nothing is estimated twice. - **Touches:** - Utility invoices - Building allocation formulas - Tenant ledgers - **Never:** - Never changes an allocation formula - Never bills back to a unit whose lease does not allow it - **Agent:** Script applying the per-building formula and posting the charges with the method shown. - **Measured by:** Percent of utility cost recovered against the lease allocation - **Execution:** Deterministic script - **Hands off to:** Post a tenant chargeback; Produce the monthly owner statement #### Ask for a reserve top-up - **Trigger:** A property's reserve falls below its minimum. - **Ceiling:** The owner hears before the account goes negative, with the number, the reason and what is coming up on that property in the next sixty days. Nobody ever gets a surprise request the day a roofer wants a deposit. - **Touches:** - Property cash position - Reserve minimums - Scheduled upcoming work - **Never:** - Never draws from another property to cover a shortfall - Never sends the request without a human reviewing the forecast - **Agent:** Simple agent forecasting the shortfall and drafting the request. - **Measured by:** Number of properties that go cash-negative per quarter - **Execution:** Simple agent - **Hands off to:** Run the owner draw; Request an owner approval; Run preventive maintenance ### Inspections & Compliance Proof that each property is in the condition the file says it is. #### Run the move-in condition report - **Trigger:** Keys are released to a new tenant. - **Ceiling:** Every room, appliance and wall is photographed and described before the tenant's first night, the tenant signs the same report, and both copies are filed where a deposit disposition can reach them in one click three years later. - **Touches:** - Condition report template - Unit photo library - Tenant signature record - **Never:** - Never accepts a report without the tenant's acknowledgement - Never edits a filed condition report - **Agent:** Script that issues the report, chases the signature and files the photo set. - **Measured by:** Percent of tenancies with a signed move-in report inside 5 days - **Execution:** Deterministic script - **Hands off to:** Dispose of a security deposit; Run the move-out inspection; Send the move-in packet #### Run a mid-lease interior inspection - **Trigger:** A tenancy reaches its six-month mark. - **Ceiling:** Every occupied unit is looked at twice a year with proper notice, and the report flags the slow leak, the missing smoke detector, the unauthorized cat and the furnace filter nobody changed. Small problems are found at $80 instead of $8,000. - **Touches:** - Inspection schedule - Notice of entry - Unit condition history - **Never:** - Never enters without compliant written notice - Never photographs a tenant's possessions beyond a documented condition issue - Never issues a violation from an inspection finding - **Agent:** Complex agent scheduling, issuing notice, and turning the inspector's findings into work orders and flags. - **Measured by:** Deferred maintenance items found per inspection, and inspection coverage rate - **Execution:** Complex agent - **Hands off to:** Triage a maintenance request; Prepare a lease violation notice; Certify smoke and CO detectors; Process a pet or occupant request #### Run the move-out inspection - **Trigger:** A move-out date arrives. - **Ceiling:** The unit is walked within 24 hours of keys, photographed to match the move-in set frame for frame, and each difference is marked as wear or damage with a cost estimate attached. The deposit statement writes itself from it. - **Touches:** - Move-in condition report - Unit photo library - Turn cost history - **Never:** - Never decides a deduction; it proposes and a human decides - Never inspects before the tenant's possession period ends - **Agent:** Complex agent pairing photo sets and proposing a wear-versus-damage split with costs. - **Measured by:** Percent of move-outs inspected within 24 hours of key return - **Execution:** Complex agent - **Hands off to:** Dispose of a security deposit; Scope a turn; Run the move-in condition report #### Drive by the exterior - **Trigger:** A property goes 120 days without any visit. - **Ceiling:** No property goes half a year unseen. A quarterly photo of every roof, gutter, fence and lawn, routed so one person covers forty houses in a day. Owners get a picture of their house twice a year without asking. - **Touches:** - Property visit log - Route scheduling - Exterior photo history - **Never:** - Never enters the property or speaks to the tenant - Never raises a work order over $500 from a drive-by alone - **Agent:** Script building the route from last-visit dates and filing the photo set. - **Measured by:** Percent of properties with an exterior photo in the last 120 days - **Execution:** Deterministic script - **Hands off to:** Run preventive maintenance; Handle an HOA violation notice; Produce the monthly owner statement #### Certify smoke and CO detectors - **Trigger:** A tenancy starts, or an annual certification comes due. - **Ceiling:** Every unit has a dated, signed detector certification on file and no unit is more than twelve months from its last one. If a fire happens, the file answers the question before the lawyer asks it. - **Touches:** - Certification records - Inspection schedule - Unit device inventory - **Never:** - Never marks a unit certified without a signed record - Never lets a failed certification sit more than 48 hours without escalation - **Agent:** Script tracking expiry, scheduling the check and escalating failures. - **Measured by:** Percent of units with a current detector certification - **Execution:** Deterministic script - **Hands off to:** Run a mid-lease interior inspection; Run the move-in condition report; Run the habitability clock #### Track required insurance - **Trigger:** A policy on file is 30 days from expiry. - **Ceiling:** Renters policies, owner landlord policies and vendor certificates of insurance are all current, all the time, with the chase starting thirty days out. No vendor works uninsured and no claim is denied because a policy lapsed in April. - **Touches:** - Policy and COI records - Vendor roster status - Tenant and owner contacts - **Never:** - Never suspends a vendor or a tenancy - Never accepts a certificate that does not name the required additional insured - **Agent:** Script that watches expiries, chases documents and marks lapsed parties. - **Measured by:** Percent of active vendors and tenancies with current coverage on file - **Execution:** Deterministic script - **Hands off to:** Block an uninsured vendor; Send the move-in packet; Open a property insurance claim; Onboard a new owner and property #### Block an uninsured vendor - **Trigger:** A vendor's certificate lapses. - **Ceiling:** The moment coverage lapses the vendor stops receiving new assignments, their open jobs are listed for a human to decide, and they get a clear request for the new certificate. Nobody discovers the gap after a ladder goes through a window. - **Touches:** - Vendor roster status - Open work orders for that vendor - COI records - **Never:** - Never removes a vendor from the roster permanently - Never cancels a scheduled job without a human - **Agent:** Script that flips assignment eligibility and lists the exposure. - **Measured by:** Work orders assigned to vendors without current insurance - **Execution:** Deterministic script - **Hands off to:** Assign a vendor; Track required insurance; Score a vendor #### Run the habitability clock - **Trigger:** A request is classified as affecting heat, water, sewer or safety. - **Ceiling:** The statutory repair clock starts the minute the report lands, everyone touching the job sees the deadline, and the remedy is documented with dates and photos. Tamarack never misses a habitability window, and the record proves it. - **Touches:** - Work order urgency flags - Statutory timelines - Repair documentation - **Never:** - Never downgrades a habitability classification - Never waits on owner approval to start a habitability repair - **Agent:** Simple agent that starts the clock, escalates against it and assembles the record. - **Measured by:** Habitability repairs completed inside the statutory window - **Execution:** Simple agent - **Hands off to:** Dispatch an after-hours emergency; Triage a maintenance request; Request an owner approval; Certify smoke and CO detectors #### Check lease terms against current law - **Trigger:** A renewal is drafted, or a city ordinance changes. - **Ceiling:** Every renewal goes out on language that is current with Spokane and Washington rules, including notice periods and fee caps, and when an ordinance changes the affected leases are listed the same week. No portfolio-wide clause is wrong for eight months. - **Touches:** - Lease template versions - Tracked ordinance sources - Affected lease list - **Never:** - Never changes template language without counsel approval - Never gives a legal opinion to a tenant or an owner - **Agent:** Complex agent tracking rule changes and mapping them to affected leases for counsel review. - **Measured by:** Days from an ordinance change to updated templates in use - **Execution:** Complex agent - **Hands off to:** Make a renewal offer; Generate the lease; Keep the fair housing record #### Keep the fair housing record - **Trigger:** Any applicant decision or leasing message is recorded. - **Ceiling:** Every inquiry, showing, application and decision is logged with the criteria version used, so the company can show equal treatment across any protected class on any month of any year within minutes. Leasing copy is screened for problem language before it publishes. - **Touches:** - Leasing communication log - Application decisions - Published criteria versions - **Never:** - Never alters or deletes a logged record - Never reports on an individual employee without HR involved - **Agent:** Script that logs immutably and runs the periodic disparity report. - **Measured by:** Time to produce a complete audit trail for a given applicant - **Execution:** Deterministic script - **Hands off to:** Screen an application; Deliver a decline; Launch a listing; Check lease terms against current law #### Assemble an eviction file - **Trigger:** A delinquency passes the policy threshold and a human asks for the file. - **Ceiling:** The ledger, every notice with its proof of service, the lease, the communication history and the payment plan record are assembled into one packet for counsel in under an hour. Cases are never lost on a service defect. - **Touches:** - Tenant ledger - Notice and service records - Communication history - **Never:** - Never files anything with a court - Never contacts the tenant - Never recommends whether to proceed; a person decides that - **Agent:** Simple agent that assembles and checks the packet for gaps. - **Measured by:** Cases dismissed or delayed on a procedural defect - **Execution:** Simple agent - **Hands off to:** Prepare a late notice; Work the delinquency list; Prepare a lease violation notice #### Handle an HOA violation notice - **Trigger:** An HOA letter arrives about a trash can left at the curb. - **Ceiling:** The notice is matched to the property and the responsible party the day it arrives, the tenant or vendor is asked to fix it, and the HOA gets a reply before the fine window closes. Tamarack stops paying $75 fines because mail sat in a tray. - **Touches:** - HOA correspondence - Property and tenant records - Lease rule clauses - **Never:** - Never pays a fine - Never disputes a violation with the HOA without a human - **Agent:** Simple agent that routes the notice, assigns the fix and tracks the cure deadline. - **Measured by:** HOA fines paid per quarter - **Execution:** Simple agent - **Hands off to:** Drive by the exterior; Prepare a lease violation notice; Produce the monthly owner statement #### Open a property insurance claim - **Trigger:** Damage exceeds the deductible after a loss. - **Ceiling:** Photos, the repair estimate, the police or fire report and the policy details are in the carrier's hands within 48 hours of the loss, and the claim has an owner and a next date until it closes. No claim goes stale for a month. - **Touches:** - Loss documentation - Owner policy details - Repair estimates - **Never:** - Never files a claim without the owner's written go-ahead - Never accepts a settlement offer - **Agent:** Complex agent assembling the claim package and tracking the carrier back and forth. - **Measured by:** Days from loss to claim filed, and claim cycle time - **Execution:** Complex agent - **Hands off to:** Dispatch an after-hours emergency; Track required insurance; Request an owner approval ### Growth New doors on, good doors kept, the market watched. #### Work an inbound owner lead - **Trigger:** A landlord fills in the form asking what management costs. - **Ceiling:** Within fifteen minutes they have a real answer: a rent estimate for their address, what the fee covers, and a link to a time on a calendar. The proposal is built from public data on their actual house, not a brochure. Nobody waits two days and signs with the competitor who called back first. - **Touches:** - Lead inbox - Public property data - Rent comps - The calendar - **Never:** - Never quotes a fee outside the published schedule - Never promises a rent figure as a guarantee - **Agent:** Complex agent that researches the property, drafts the estimate and books the call. - **Measured by:** Lead to signed agreement rate, and first response time - **Execution:** Complex agent - **Hands off to:** Onboard a new owner and property; Survey the market; Ask for a referral #### Survey the market - **Trigger:** Every Monday. - **Ceiling:** A weekly picture of asking rents, days on market and concessions across the Spokane submarkets Tamarack works in, by bedroom count and property type. Pricing and renewal decisions run on last week's data, not last year's feel. - **Touches:** - Public listing data - Internal closed lease history - Submarket definitions - **Never:** - Never sets a price - Never scrapes a source the terms of use forbid - **Agent:** Script pulling listings and publishing the weekly submarket table. - **Measured by:** Portfolio rent against submarket median by bedroom count - **Execution:** Deterministic script - **Hands off to:** Price a vacant unit; Make a renewal offer; Work an inbound owner lead #### Ask for a referral - **Trigger:** An owner has two good quarters, or a tenant closes a work order happy. - **Ceiling:** The ask lands at the one moment somebody feels good about Tamarack, with a specific reason and an easy way to pass a name along. Referrals become a predictable share of new doors instead of an accident. - **Touches:** - Owner and tenant satisfaction signals - Referral program terms - Contact preferences - **Never:** - Never asks an owner or tenant with an open complaint or dispute - Never offers an incentive outside the published program - **Agent:** Simple agent that watches for the moment and drafts the ask. - **Measured by:** New doors sourced from referral per quarter - **Execution:** Simple agent - **Hands off to:** Request and answer reviews; Read tenant sentiment; Work an inbound owner lead #### Request and answer reviews - **Trigger:** A happy work order closes, or a new review posts anywhere. - **Ceiling:** The review ask goes to people who just had a good experience, and every review that lands gets a human-sounding reply within a day, including the angry ones. The public rating tells the truth about the company and the bad reviews get answered with facts. - **Touches:** - Review platforms - Work order outcomes - Tenant and owner contacts - **Never:** - Never posts a reply to a negative review without a human approving - Never discloses tenancy or account details in a public reply - Never offers anything in exchange for a review - **Agent:** Simple agent drafting asks and replies, with negatives held for approval. - **Measured by:** Review volume and average rating over rolling 90 days - **Execution:** Simple agent - **Hands off to:** Ask for a referral; Close a work order and ask the tenant; Read tenant sentiment #### Save a leaving owner - **Trigger:** An owner sends a cancellation notice. - **Ceiling:** Within an hour a person has the whole relationship on one page: every approval delay, every vacancy, every unanswered message, what the property actually earned, and the two or three things that would have to change. The conversation is honest and specific, and the ones worth keeping get kept. - **Touches:** - Owner history and performance - Communication record - Management agreement terms - **Never:** - Never contacts the owner - Never offers a discount or a term change - Never delays the offboarding steps the agreement requires - **Agent:** Complex agent building the one-page account history and the candid diagnosis for a human call. - **Measured by:** Percent of cancellation notices retained - **Execution:** Complex agent - **Hands off to:** Spot an owner about to leave; Offboard an owner cleanly; Answer an owner question #### Offboard an owner cleanly - **Trigger:** A termination is confirmed. - **Ceiling:** Final statement, deposit transfer, keys, leases, tenant notification and the new manager's handoff packet are all done inside the notice period, correctly, without drama. They leave well, and about one in six comes back. - **Touches:** - Final ledgers and deposits - Document archive - Tenant notification templates - **Never:** - Never transfers a security deposit without written instructions and a receipt - Never notifies tenants before the owner has confirmed the date - **Agent:** Complex agent running the offboarding checklist and assembling the handoff package. - **Measured by:** Offboardings completed inside the notice period with no post-exit disputes - **Execution:** Complex agent - **Hands off to:** Save a leaving owner; Dispose of a security deposit; Reconcile the trust account #### Forecast capacity before the next door - **Trigger:** The door count moves, or a hiring question comes up. - **Ceiling:** Before Tamarack signs the next forty doors, it knows what that does to work orders per manager, showings per week and hours in the collections seat, and what breaks first. Growth stops being the thing that wrecks service quality. - **Touches:** - Door count and pipeline - Work order and leasing volumes - Staff hours by role - **Never:** - Never makes a hiring decision - Never models an individual employee's performance - **Agent:** Complex agent modeling load per role against the pipeline and naming the first constraint. - **Measured by:** Doors per property manager against the service quality target - **Execution:** Complex agent - **Hands off to:** Work an inbound owner lead; Onboard a new owner and property; Score a vendor --- # Labs > What Gold Star is thinking about, working on, and shipping. Source: https://goldstarlabs.ai/labs Machine view: https://goldstarlabs.ai/machine/labs Before Gold Star, the two of us spent our careers on private engineering and skunk works style teams for three of the four richest people in the world. We’ve built the operational plumbing to scale startups from zero to a billion, and we’ve spent years bridging the gap between sci-fi sounding deep tech concepts and the commercial real world. Robotics, rockets, foundational AI models, energy, fully autonomous companies, and many more. Old habits die hard. Labs is where we keep working on the problems that pull at us, whether or not anyone is paying for them yet. We’re using our expertise, experience, and resources to skate to where the puck is going. Gold Star Labs is where we publish the things we’re thinking about, working on, and shipping. ## What we’re thinking about right now ### Digital and physical AI Operating Systems We’ve built dozens of AI Operating Systems but they all stop at the computer screen. With a majority of the world’s economic value existing in the physical world, it seems to us like the next natural progression is how to integrate these digital AI agents with their real world physical counterparts to capture a larger share of the economic pie. This doesn’t necessarily mean we’re building robots, but rather seeing how our existing mapping and terraforming engine can touch the real world through hardware that already exists where the economic value is created. How do we program a Raspberry Pi with a cheap local model to turn a CCTV camera into a safety asset in a factory? How do we integrate biometric data from a wearable with emergency dispatch to launch a drone carrying antivenom to a spider bite victim who is hiking in a remote canyon? How do we integrate sensors, drones, cameras, and other machines to sense and actuate within a physical bounded environment the same way Claude Code does on a computer? How do we make the system of digital and physical agents share the same underlying foundation and language to keep everything running smoothly and safely? ### What use cases do the new AI fueled advances in mathematics, physics, computer science, and more open up? Foundational AI models are solving mathematical problems that have been outstanding for hundreds of years, which rewrites the possible solution space for countless industries overnight. This is like rewriting the laws of physics overnight and it’s happening all the time. To us, that means things that were once not possible might now be worth exploring. Whether that’s new technologies, business models, markets, and more. ### How does a team of AI agents that act like employees change the economics of what’s reasonable? When you can run a business that has the output of 1000 with a team of 5 plus a ton of AI agents, what things become feasible that were not before? This is the same concept as the one person billion dollar company, but what if an individual’s profit was not the entire end goal? What if the end goal was creating a system to fight fires effectively that actually fits within a tight local county budget? What about building armies of AI agents to run millions of parallel experiments that would have previously taken researchers a decade? Or how do we create something amazing and use agents to bring to market quickly and cheaply enough that it allows the owner to avoid predatory outside capital? ### Making cheap models and hardware punch way above their weight The smartest models are expensive and it’s not clear what happens to your data when you submit a query. Also in many cases they are way overpowered for the particular use case, like driving a leased Ferrari through the McDonald’s drive-thru. As hardware improves, quantization of small models improves so you can do way more with less, hardware improves so you can actually run a smart local model on a Raspberry Pi, and fine tuning techniques improve we think you can get small models to do one or two things really well. If you are intentional about those things, you can spend a few thousand dollars one time and have a fully local AI system that perfectly fits your use case. ### Local AI models To play off the leased Ferrari to McDonald’s example, how do we build the fully paid off Toyota Camry of AI systems? It’s the daily driver that is reliable, paid off, and perfectly suited for your everyday use case. And most importantly you own it and as you drive it more often, you start to get a sense for what it can and can’t do. Besides just cost, data sovereignty is an important thing to consider. For law firms, businesses working on top secret work, and medical companies working with sensitive data. You can’t just not use AI in today’s day and age. Just like we’re building physical and digital AI operating systems, how do you build a fully local AI operating system that actually provides value and doesn’t need millions of dollars of computing hardware to be effective. Between setting up the hardware correctly, fine tuning smaller and dumber models to execute tasks specifically, orchestrating agents, and keeping everything on a set of servers so that your data never leaves the building, there is a lot to get right. That is exactly the kind of problem we like. ### How do you trust an AI that is sending commands to a machine that will do things in the real world? All signs are pointing to AI being able to actuate machines without a human in the loop, but with LLMs being probabilistic in nature, it’s important that there is a layer of trust between the LLM and the machine, so both sides verify that the command is correct and the actuation has been executed by the machine. And just as importantly, how do we create a log so humans can verify everything that happened and have detailed trace logs whenever they need it. And how does that layer sit in coordination with the rest of a digital and physical AI operating system? ### Agent to agent commerce What does the economy look like when agents are executing financial transactions, negotiating on behalf of their human with other agents representing their human? We’re already seeing this with search and shortlisting things on behalf of a human. But what happens when AI agents are handling the entire commerce stack? How does the go to market motion change when agents are making the full decision just based on the constraints of a human? What are the downstream effects on business models that rely on advertising as their sole source of revenue when humans are no longer the ones that need to be influenced to make a purchase? What happens to negotiations and complex sales cycles when AI agents are the ones doing all the back and forth? How do we trust the AI agents and create systems where an AI agent can access the right amount of information to effectively negotiate and not hallucinate? ### AI safety and alignment How do we make sure that within AI operating systems of all shapes and sizes, that these agents are strictly bounded and tight governance is enforced? Some of what we’re working on here we can’t talk about yet. If you’re working on something at these edges, or you have a problem that needs a team like this, reach out. --- # Philosophy > How Gold Star thinks and why it builds the way it does. Source: https://goldstarlabs.ai/philosophy Machine view: https://goldstarlabs.ai/machine/philosophy Follow one branch all the way down. Every individual action has a purpose, a set of inputs, and a boundary. - [The business map](https://goldstarlabs.ai/index.md#business-map): One branch followed all the way down, on the home page. ## A business is a set of actions Strip any business down to what it actually is and you find the same thing every time. It’s a finite set of actions repeated in a particular way based on specific triggers. A lead submits a form and someone sends an email to follow up with them, a job wraps up and someone sends an invoice, a client has gone dormant for 6 months and needs a check-in, a discovery call wraps up and a proposal needs to be drafted, a listing is won and needs to be matched to a buyer. Every action has a trigger, a set of inputs, some rules, and an output. Historically most businesses don’t have this mapped out to this level even if they know it would be valuable because they don’t have the resources or a map of the business is no good if there is no capability to execute on it. Now with AI agents being able to execute on specific tasks, and handle entire department level work when chained together properly, having this map is tremendously valuable for a small business. If you can map every possible action within your business, what great looks like in that action, how all of those actions fit together, connect them to the right tools so they can execute, and then “hire” a bunch of AI agents, each perfectly trained to be amazing at one particular action inside your business, all of a sudden that map plus a little AI development elbow grease is the difference between tripling your revenue and flatlining. It’s like hiring a massive coordinated team of fractional experts and pointing them directly at your business. ## Every action has a ceiling For every action on the map, we ask a question. If the most talented person alive worked on nothing but this one action, all day, every day, in the unique context of this particular environment, what is the best possible result that could be achieved? We refer to that as the entitlement value of the action, which is essentially its ceiling. In reality, no business or particular action runs at this ceiling, because there are very real time and resource constraints that everyone has to deal with. But by understanding the entitlement value of an action as well as every possible input and output that goes into it, we can make tradeoffs to maximize return on investment for every particular action. However, in almost every business we work with we find a massive gap between current operations and 90% of the entitlement value of the business. That is the exact value that Gold Star and a well executed AI Operating System can bring. It helps you do infinitely more with less in a way that hasn’t been possible for small businesses in the past. ## One agent per action, built to get as close to the ceiling as possible Every action on the map has boundaries that cannot be crossed, metrics that show whether it’s working, and a level of complexity. The level of complexity is key and we architect systems to layer the agents with different levels of complexity to minimize cost and maximize impact. Each action’s execution ranges from a single deterministic script to a complex AI agent that can make complex decisions. The execution of each action is carefully chosen to optimize cost and reliability without sacrificing quality. ## Built to amplify the human in the loop, not replace them AI is great at execution, but it cannot replace human judgement, tacit knowledge, relationships, or the other things that make people so valuable. Gold Star’s AI Operating Systems are built around the principle of amplifying the human that sits at the center of all the agents, not replacing them. The goal is to have the human in the loop have more time and energy to do more of whatever it is they signed up to do in the first place and less time on tedious things like email and document formatting. ## Think slow, act fast There is one book that we come back to constantly when building these systems called How Big Things Get Done by Bent Flyvbjerg. He studied thousands of projects and found that successful ones share two habits. First, they plan carefully and truly understand the arena before committing to anything. Second, they build large complex systems by thoughtfully stitching together smaller battle tested components. That’s exactly how we think about everything we build. We understand every underlying piece that could possibly go into it before writing a single line of code. Then when it is time to execute, we lean on our heavily tested component suite to construct the major building blocks. From there, we rely on our expertise of integrating everything together, training agents on a specific environment, and building the connective tissue between everything we construct. The result is a robust, reliable, traceable, and powerful system that lets any small business reach their entitlement value without breaking the bank. --- # About us > Thomas Baker and Michael Krebs, and how Gold Star started. Source: https://goldstarlabs.ai/about Machine view: https://goldstarlabs.ai/machine/about Thomas Baker & Michael Krebs ## Our story We met in Blue Origin's innovation lab where we were working on the next generation of deep tech and space technology. After Blue Origin we went our separate ways, Thomas going on to work on frontier AI model development and teach AI at Harvard Business School, Michael launching a SaaS company that ran entirely with AI agents as the employees. We both realized just how impactful AI could be for small teams when applied correctly, so we started Gold Star. Since then we've worked with dozens of companies across real estate, healthcare, tech, media, local SMBs, agencies, deep tech companies, solopreneurs, and more. We both now teach classes on AI Agents at Harvard Business School and work with many of the cutting edge AI and deep tech companies. Our mission is to build AI that's actually useful, not implement AI for the sake of implementing AI. That spectrum ranges from building a suite of AI agents around a human to handle all the execution across an entire business, to a simple AI system that eliminates a tedious task that used to take 5 hours per week. ## Companies we’ve helped Commercial Real Estate. Deep Tech. Financial Services. Tech Startups. Local SMBs. Want a free AI audit of your business to see where agents could help you? --- # Contact > How to reach Gold Star AI. Source: https://goldstarlabs.ai/contact Machine view: https://goldstarlabs.ai/machine/contact What are you working on? A small problem, a whole business, or something at the edges. We’d like to hear about it. - **Email:** michael@goldstarworkflow.com - **Form:** https://goldstarlabs.ai/contact - Tell us how things work today. - Tell us what keeps getting in the way. - Tell us what great would look like.