Buy deal-screening AI to rank an inbox faster, never to make the buy decision for you, and hold every tool to the same ten rules before you sign. A screening tool earns its keep when a principal with a written buy box can triage a week of broker blasts in an afternoon and spend the recovered hours underwriting the two or three deals that deserve it. It becomes a liability the moment a confident score, built on a number the model misread off a T-12, moves money. Most vendors sell you the score and stay quiet about the second half. These ten rules are the buyer’s code that keeps the first outcome and forecloses the second, and they end in a one-page scorecard you can take into any demo.
What deal-screening AI actually is
Deal-screening AI reads the inbound deal — an offering memorandum, a T-12, a rent roll, a broker’s one-pager — and does two jobs: it pulls the key terms into structured fields, and it ranks the deal against your criteria so you know what to open first. That is triage, not a verdict.
The category blurs on purpose. Some products are pure extraction-and-rank engines; some layer a first-pass model on top; some are full underwriting platforms that happen to ingest a PDF. The word “AI” covers all of it, which is why two firms can buy “deal-screening AI” and get two different products.
The screening layer sits at the top of the funnel described in our deal-analysis playbook: triage the blast, pick the deals worth the hours, then underwrite the survivors properly. Buy for that job, and the rules below tell you how to buy well.
Rule 1: Write your buy box before you shop
A screening tool ranks deals against criteria. If you have not written yours down — target markets, asset classes, deal size, minimum yield, hold period, deal-breakers — the tool ranks against generic defaults and hands you a confident, useless order.
The buy box is the input that makes screening work, and most small firms carry it in a principal’s head rather than on paper. Get it out of the head first: a one-page statement of what you buy and what you never buy. That page is worth more than any subscription, because it is what you will test every vendor’s ranking against.
Rule 2: Know whether you are buying screening or underwriting
Screening ranks deals so you know what to open. Underwriting models the cash flows so you can price a bid. They are different jobs, and buying one when you need the other is the most common and most expensive mistake in this category.
Ask every vendor directly: does this rank my pipeline, or does it produce a model I would defend to an investment committee? A ranking tool that markets itself as “AI underwriting” will over-promise on the model; a modeling platform sold as “screening” will be too heavy for triage and priced accordingly.
Small firms almost always need the screening layer first — the thing that turns forty broker blasts into a ranked short list. The underwriting model can stay in the spreadsheet you already trust, at least until volume justifies more.
Rule 3: Test on your real inbox, not the demo deal
Every demo runs on a clean, well-formatted OM the vendor chose. Your inbox is scanned PDFs, a rent roll pasted into an email body, a T-12 with a typo in the tax line, and a “call for offers” with no numbers at all. The tool has to survive that, not the demo.
Run the trial on ten of your own recent deals, including the two ugliest. Watch what the tool does with the messy ones: does it extract cleanly, flag what it could not read, or silently guess? Silent guessing is the disqualifier — a screen that fabricates a missing NOI is worse than no screen, because it looks finished.
Rule 4: Demand a ranking you can audit
A score is only useful if you can see why the deal earned it. “This deal scored 82” is a number you cannot defend; “this scored high on market and size, low on in-place yield, and here are the three inputs that drove each” is a screen you can stand behind in front of a partner.
Ask to see the reasoning behind a rank, in a form a principal can read in ten seconds. If the tool cannot show its work, you are trading your judgment for a black box, and the first time it ranks an obvious dog above a strong deal, the whole team stops trusting the output.
Rule 5: Make every extracted number cite its source
The dangerous errors in screening are not the deals the tool ranks wrong — those get caught. They are the numbers it extracts wrong and presents cleanly: an in-place rent read off the wrong column, a T-12 line item mislabeled, an expense that quietly went missing.
Require grounding: every extracted figure should link back to the page and line it came from, so a person can verify in one click. This is the same discipline that matters when buying document AI, covered in our companion piece on buying document AI when your firm runs on PDFs — a number with no source is a rumor, not data.
For screening, grounding is what makes the triage trustworthy enough to act on quickly. Without it, you re-check everything by hand and the tool has saved you nothing.
Rule 6: Keep the go/no-go human by design
The tool ranks and drafts. A person decides, and money never moves on a number no human has verified. Build the workflow that way on purpose, because the failure mode of a good screening tool is that it works well enough that people stop checking.
Draw the line explicitly: the AI produces a ranked list and a first-pass summary; a principal reads the survivors, verifies the numbers that matter, and makes the call. That is not a lack of trust in the tool — it is the correct division of labor, and it is the same principle the small-firm manifesto argues makes a lean shop faster than a bloated one: AI clears the busywork so human judgment goes where it counts.
Rule 7: Settle the data terms before you upload a deal package
Deal packages are confidential, often under NDA, and frequently carry a seller’s financials that are not yours to feed into someone else’s model training. Before you upload the first OM, get three answers in writing: where the data is stored, whether it is used to train shared models, and how you delete it and export your history when you leave.
Small firms handle genuinely sensitive information without an IT department to vet vendors, which makes the contract the only safeguard. “Your data is secure” is marketing; a data-processing addendum that says the vendor will not train on your uploads and will delete on request is a commitment.
If a vendor is vague on any of the three, treat that as an answer. The same data-ownership questions apply whether you buy a platform or build your own pipeline, as our build-versus-buy breakdown on CRE data lays out in detail.
Rule 8: Match the tool to your volume and asset mix
A dedicated screening platform pays off at volume. If forty real deals cross your desk a week, a tool that ranks them in minutes is worth a subscription. If you see six a month, an annual platform is an expensive login, and the honest answer is a lighter workflow.
Asset mix matters as much as count. A tool tuned for multifamily rent rolls will underperform on an industrial NNN deal or a land play; a general extraction engine handles variety but ranks less precisely. Buy for the deals you actually screen, not the ones in the vendor’s case study. Buying always-on capacity for an occasional need is the same trap that catches firms overbuying location analytics, as we argued in the case for and against Placer.ai; count your realistic monthly volume before you sign.
Rule 9: Price the maintenance tail, not the sticker
The subscription is the smallest number in the total. The real cost is the tail: the time to load your buy box, connect your inbox and pipeline, train the team, correct the tool’s early misreads, and keep it current as your criteria shift.
Ask who does that work and what it costs. For an off-the-shelf platform, market subscriptions for small-firm deal tooling run a modest monthly fee; a custom-built screening workflow is a project, generally in the tens of thousands to low six figures depending on scope. Both carry an ongoing tax in someone’s time, and a tool nobody maintains decays into noise within a quarter.
Get the twelve-month number, not the monthly one, and put your own hours in it. That is the figure to compare against hiring or against doing nothing.
Rule 10: Get fluent before you buy the platform
Much of first-pass screening is a well-built prompt over ChatGPT, Claude, or Gemini plus the spreadsheet you already run. A principal who can write a clear buy-box prompt and read an OM against it will get most of the triage value with tools the firm already pays for.
Getting fluent first does two things. It tells you whether you need a platform at all — many small firms find the prompt-plus-spreadsheet workflow covers their volume — and if you do buy, it makes you a far sharper buyer, because you know exactly what the automation is replacing.
This is where a short, hands-on fluency workshop earns out: teach the team to prompt against LOIs, deal summaries, and market write-ups, then decide whether a dedicated tool is worth it. Buy fluency before you buy software; the order matters.
The rules as a buying scorecard
Turn the ten rules into a one-page test and score every tool the same way. Bring it to the demo.
| # | Rule | Pass condition |
|---|---|---|
| 1 | Buy box | You have a written buy box the tool can rank against |
| 2 | Screening vs. underwriting | Vendor states clearly which job the tool does |
| 3 | Real-inbox test | Extracts cleanly on your ten deals, flags what it can’t read |
| 4 | Auditable rank | Shows why each deal scored as it did |
| 5 | Grounded numbers | Every figure links to its source page and line |
| 6 | Human go/no-go | Workflow keeps the decision with a person |
| 7 | Data terms | Storage, training, and deletion answered in writing |
| 8 | Volume + asset fit | Matches your real monthly count and asset classes |
| 9 | Total cost | Twelve-month number includes setup and your time |
| 10 | Fluency first | Team can already prompt the workflow by hand |
A tool that fails rules 5, 6, or 7 is out regardless of how it scores elsewhere — those are the ones that lose money or leak data. A tool that passes all ten is worth a paid pilot on live deals before you commit to a year.
Frequently asked questions
What is deal-screening AI for commercial real estate?
Deal-screening AI reads inbound deals — offering memorandums, T-12s, rent rolls, broker one-pagers — extracts the key terms into structured fields, and ranks each deal against your investment criteria so you know what to underwrite first. It is a triage layer that turns a flooded inbox into a ranked short list, not a decision engine and not underwriting software. Used well, it recovers the hours a lean team burns opening deals that never had a chance, and redirects them to the two or three deals worth a full model.
What is the difference between AI deal screening and AI underwriting?
Screening ranks deals so you know what to open; underwriting models the cash flows so you can price an offer. Screening answers “is this worth my time,” underwriting answers “what is it worth.” Many vendors market the two interchangeably, which leads firms to buy a ranking tool when they need a model, or a heavy modeling platform when they only need triage. A small firm almost always needs the screening layer first, because triage is the binding constraint; the model can stay in the spreadsheet the team already trusts until volume justifies more.
How much does deal-screening AI cost for a small CRE firm?
It depends on whether you buy or build. Off-the-shelf screening and deal-management subscriptions for small firms typically run a monthly per-seat or per-firm fee, while a custom-built screening workflow is a project that generally lands in the tens of thousands to low six figures depending on scope. The subscription is rarely the real cost — setup, integration, training, and the time to maintain the tool make up most of the twelve-month total. Price the full year with your own hours included, not the monthly sticker.
Can I just use ChatGPT or Claude to screen deals instead of buying a tool?
Often, yes, for the first pass. A well-written buy-box prompt over ChatGPT, Claude, or Gemini plus a spreadsheet will extract and rank many deals with tools your firm already pays for, and at low volume that may be all you need. The limits: general models will confidently misread a number off a messy T-12, and they do not connect to your inbox or keep a pipeline, so you paste and manage by hand. Use them to get fluent and to prove whether you need a platform at all — verify every number they pull, and never let a model’s figure move money unchecked.
Is my confidential deal data safe in a deal-screening AI tool?
Only if the contract says so. Deal packages are confidential and often under NDA, so before uploading anything, get three answers in writing: where the data is stored, whether your uploads are used to train shared models, and how you delete and export your history. “Your data is secure” is a marketing line; a data-processing addendum that commits the vendor not to train on your data and to delete on request is an enforceable one. A small firm without an IT department has the contract as its main safeguard, so treat vagueness as a reason to walk.
How accurate is AI at reading an OM or a T-12?
Modern tools are strong on clean, well-formatted documents and much weaker on the scanned, inconsistent, typo-ridden files that fill a real inbox. The accuracy that matters is on your worst documents, not the vendor’s demo deal. The failure mode to fear is not a visible error — those get caught — but a clean-looking number pulled from the wrong column or line. That is why grounding matters: every figure should link to its source page so a person verifies the ones a decision hinges on. Test on ten of your own deals, including the two ugliest, before you trust any of it.
How many deals a month justify buying a dedicated screening tool?
A dedicated platform pays off when volume is high enough that manual triage is a real bottleneck — dozens of genuine screens a week rather than a handful a month. If you see six deals a month, an annual platform is an expensive login, and a prompt-plus-spreadsheet workflow will cover you. Count your realistic monthly volume of deals you would actually screen, not your total inbound, and weigh it against the twelve-month cost. Volume, not the vendor’s pitch, decides whether the subscription earns its keep.
Should a small firm build its own deal-screening workflow or buy a product?
Buy first if an off-the-shelf tool fits your asset classes and volume, because a subscription is faster and cheaper to start than a build. Build only when your criteria or workflow are specific enough that no product ranks them well, or when integration with systems you already run justifies the project cost. The math turns on how standard your process is and how much a tool would have to be bent to fit it. Many small firms land in the middle: a general tool plus prompt workflows for the parts unique to how they screen.
What should disqualify a deal-screening tool outright?
Three failures. First, silent guessing — a tool that fabricates a missing number instead of flagging it is worse than no tool, because it looks finished. Second, a black-box rank you cannot audit — a score with no visible reasoning is a liability the first time it ranks a bad deal high. Third, vague data terms — no clear answer on storage, training, and deletion means confidential deal data is at risk. Any one of these is a walk-away regardless of how the tool performs elsewhere; they are the failures that lose money or leak data rather than merely waste time.
Do we still need a human to underwrite if the AI screens?
Yes, and by design. The tool ranks deals and drafts first-pass summaries; a person verifies the numbers that matter and makes the go/no-go call. The correct division of labor is AI for the busywork of reading and ranking, human judgment for the decision and for underwriting the survivors. The risk with a good screening tool is that it works well enough that people stop checking, so build the human review in deliberately. A vendor who claims the tool replaces the analyst’s judgment is describing a risk, not a feature.
Where to start
The first question is not which deal-screening tool to buy. It is whether your buy box is on paper, whether your team can already prompt a first-pass screen by hand, and how many deals a month would genuinely use the automation. Answer those and the tool decision mostly makes itself. A free AI-readiness assessment gives you that read: a short working session that looks at your deal flow, your asset mix, and where your screening hours go, then returns an honest recommendation on whether a platform, a custom workflow, or a fluency-first approach with the tools you already own is the right next spend. Book a free AI-readiness assessment before you sign an annual contract for a tool your volume may not need.
Arthur Wandzel
