AI for commercial real estate
We speak the languageCommercial brokers working multifamily, retail, and hospitality, whose day still runs on a cold-call list.
A commercial broker already has the data. Buildout holds the pipeline, CoStar holds the comps, and the county holds the ownership records. What is missing is something that reads all of it and answers a plain question: who should I call today, and what do I say when they pick up.
Plain English
Ask the pipeline a question, get a list
Argues back
It challenges the model, not just fills it
Your data
Built on the sources you already pay for
Where the time goes
Brokers do not have a data problem. They pay for more data than they can read. The problem is that nothing turns it into the next phone call.
The day runs on cold calls, and the gold is accurate owner contact data: a cell number that works and an email that reaches the decision-maker. Assembling that list means cross-checking county records against a platform against a spreadsheet, and it happens before the selling starts, every morning.
The list is built before the day begins
County records are genuinely useful and genuinely imperfect. An LLC owns the asset, the mailing address is a registered agent, and the contact on file has not been the decision-maker for six years. No single source solves this, and the broker who works out which source to believe is doing that work with judgment nobody has written down.
No single source is clean
Owned it eight years, financing is coming due, last contact was nineteen months ago, it is a retail strip in a submarket that just repriced. Every one of those is a reason to call, and together they are a ranking. In practice they live in four systems and get combined in somebody's head, on a good day.
Timing signals live in four systems
A rent roll, a trailing twelve, an offering memorandum, and a broker's opinion of value. All the answers are in there. Reading them properly takes an hour per deal, and there are more deals than hours, so they get skimmed and the model gets built on the skim.
An hour per deal file, if you read it
The sharpest critique of AI we have heard from a broker: it is a yes-man. Ask it about your underwriting and it will find a way to like it. That is worse than useless in a business where the entire skill is knowing which assumption is soft, because it launders your own optimism back to you with a confident tone.
A model that agrees is not a check
You met the owner at a conference, had a good call, and made a note. Eighteen months later the asset is trading and somebody else is representing it, because their follow-up was a system and yours was a memory attached to a name you half-recall.
The relationship decays quietly
The machine
The machine underneath a brokerage practice, before anybody points AI at it.
01
02
03
04
The split
The design goal a broker gave us, and it is the right one: build AI that argues back. Not a system that fills in your model, but one that reads the rent roll, reads your assumptions, and tells you which one is not supported. A model that agrees with your underwriting is not a check on it. It is a mirror with a confident voice.
What we build
01
The daily engine. Who to call, with contact data that works, ranked by the signals that mean timing: years owned, financing situation, last contact, and how complete the contact record actually is. The output is a list a broker can work, not a report.
02
Ask for what you want the way you would ask an analyst. Build me a call list with these filters. Show me the top ten owners most likely to sell. Rank them by how long they have owned it and their financing situation. The value is not the language model. It is that you stop building the list by hand.
03
It reads the rent roll and the model side by side and tells you where they disagree. Not a valuation, and not a recommendation. A list of the assumptions your model is carrying that the source documents do not support, which is the conversation a lender is going to have with you later anyway.
04
The trailing twelve, the rent roll, the offering memorandum, and the broker's opinion of value, read and reconciled into what actually matters and what contradicts. The hour per deal file is the bottleneck. This is the system that gives it back.
05
The book, watched for reasons to reach out: an ownership anniversary, a financing event, a comparable trade in the submarket, or simply too long since the last real conversation. The follow-up becomes a system instead of a note you meant to act on.
06
Your submarket knowledge, your relationships, and how your shop actually underwrites, written down in your own cloud so the systems above read from it. Otherwise every one of them is a generic tool that happens to have your logo on it.
Invented numbers, not a client result
This is a walkthrough we built to show what an underwriting check looks like in practice. The numbers are made up. There is no client, no deal, and no real property behind any of it. A model projects net operating income of $467,000. The rent roll, read line by line, supports $420,000. At a 5 percent cap rate that gap is roughly $940,000 of valuation, sitting inside an assumption nobody flagged. That is the shape of the problem. It is not a result we produced for anybody, and if we ever have a real one, it will say so and it will name the client.
$467K
Net operating income the model assumed
$420K
What the rent roll actually supported
~$940K
Valuation gap at a 5% cap rate
Under the hood
Deal pipeline and marketing
Market data and comps
Property financials
You already pay for these, and they are good at what they do. Buildout runs the pipeline, CoStar holds the comps, and a Yardi export is the unit-level truth about rent. None of them will read all three together and tell you who to call. That is the gap, and it is the only thing worth building here.
Principles
A broker's real skill is knowing which assumption is soft. A system that validates whatever you hand it removes the one thing you needed from it. So we build the check to disagree by default and make it show its source, because a disagreement you can trace is useful and one you cannot is noise.
Every elegant ranking model in the world is worthless if the cell number is wrong. In this industry, data quality is not a prerequisite to the interesting work. It is the interesting work, and any system that treats it as a solved input is going to disappoint you in week two.
Not a dashboard, not a score, not an insight. A list of people to call today, in order, with the reason attached. If a broker has to interpret the output before they can act on it, the system has moved work rather than removed it.
County records, platform data, and the rent roll disagree with each other constantly. A system that silently picks one is making a judgment call and hiding it. Ours shows which source it took each number from, so the broker can overrule it with the thing they know that the data does not.
Same approach, different language
Construction
General contractors, electrical and specialty trades, and renovation companies billing progress against a contract.
Live system →Financial services
Advisory firms, lenders, and finance teams where senior people spend their day gathering what they need to think about.
Patterns, no client yet →Professional services
Agencies, consultancies, and expert firms where the product is judgment and the constraint is the founder's calendar.
We speak the language →Questions
Yes, and it is the highest-value thing to build first in this industry. The list is assembled from ownership records, years owned, financing situation, last contact date, and how complete the contact data is, then ranked by what those signals say about timing. The output is a list of people to call today with the reason attached, not a dashboard to interpret. The hard part is not the ranking, it is the contact data underneath it.
No, and you should be suspicious of anybody who says they will. Buildout runs your pipeline and CoStar holds the comps, and both are better at those jobs than anything we would build. The gap is that nothing reads across your pipeline, the market data, the county records, and the deal files at once and turns it into a call list. That reading layer is what we build, on top of what you already pay for.
That is one of the most useful things it can do, with a boundary: it checks, it does not value. It reads the rent roll line by line against what your model assumes and tells you which assumptions the source documents do not support. It will not tell you what the asset is worth. That is your read on a submarket, and a language model has no business having an opinion about it.
Most of them summarize what is already in the record, which you could have read. The difference is a brain: something that holds how your shop actually underwrites, which sources you trust for what, and what your relationships are worth, then reads across all of it. Without that layer you have a generic tool with your logo on it, which is why the AI features in your CRM have not changed your day.
Not if it is built right, and this is the most common failure. The sharpest critique of AI we have heard from a broker is that it is a yes-man, and he was correct. A check that agrees with your model is a mirror. So the underwriting challenger is built to disagree by default and to cite the source line it disagreed from, which makes the disagreement something you can act on or dismiss on evidence.
No, and we are not going to pretend otherwise. The language on this page came from a real discovery conversation with a commercial broker, and no data was ever touched and no system was ever built. The construction and insurance pages on this site are where the shipped systems and the real numbers are. If a demo scenario appears anywhere on this page, it is labeled as invented, because it is.
The call list, almost certainly, because it is the only system on this page that changes what a broker does tomorrow morning. Everything else is downstream of whether the day starts with a good list. But that is a guess until we have sat with the actual workflow, which is what the readiness workshop or the roadmap is for.
You do, from day one. The code is in your repository, the knowledge base with your submarket read and your underwriting standards is in your cloud, and the accounts are in your name. In a relationship business, handing your relationship intelligence to a vendor who holds the keys is a strange trade.
Read next
Explainer
What an AI-legible business actually looks like, the five-point readiness checklist to score your own operation, why one person is usually compensating for every broken handoff, and the 90-day path from messy to legible.
Strategy
Why businesses should start thinking in workflows instead of tools when evaluating AI consulting, automation, and operations redesign.
Strategy
Why business operators should build portable markdown knowledge bases before relying too heavily on any one AI model, platform, or vendor memory layer.
Where to start
Find the bottleneck, price the fix, build the system, then keep compounding it. The four engagements are how you buy it, and you can start on any rung.
See the four engagements →Get started
Book a 30 minute call. We find the one place your work waits on you, and you get a straight answer about whether AI is the fix. No pitch, no deck.