A deal screening framework is a written, ordered sequence of gates you run every inbound deal through, killing the ones that fail at the cheapest gate first, so the only deals that reach a meeting are the ones worth a meeting. Most small commercial real estate firms do not have one. They have a principal’s instinct, an overflowing inbox, and a Thursday deal-review call where a team of four spends ninety minutes talking through deals that a two-minute test would have rejected on Monday. The framework below fixes that. It has three gates, it ends most deals in minutes, and it shows exactly where an AI first pass compresses each step without ever taking the decision out of human hands.
What deal screening actually is
Screening is triage. It answers one question — is this deal worth my underwriting time? — and nothing more. It is not underwriting, which models the cash flows so you can price a bid. Confusing the two is the root mistake, because underwriting a deal that should have been killed at a glance is how a lean team burns its scarcest resource on deals that were never going to close.
The distinction matters because the two jobs have different accuracy bars. A screen can be fast and approximate; it only has to sort deals into keep and kill with enough confidence that you are comfortable never looking again at the ones it rejects. Underwriting has to be precise, because a number moves money. Screening protects underwriting: it makes sure the hours you spend on a full model go to the two or three deals a week that deserve them, not the forty that do not.
Every firm screens. The question is whether they do it deliberately, with a written test anyone on the team can run the same way, or by feel, one deal at a time, in a meeting. The deliberate version is a framework. The felt version is a bottleneck.
The real cost: screening in meetings instead of minutes
Here is the pattern a lean firm falls into. Deals arrive all week — broker blasts, offering memorandums, marketplace alerts. Nobody has a fast, trusted way to reject one alone, so they pile up until Thursday, when the team gathers and talks through them together. The meeting becomes the screen. Four people, ninety minutes, and most of that time is spent reaching a “no” that one person could have reached in two minutes with a written test.
Industry observers have a name for the waste on the analyst side of this: the dead deal tax — the hours spent evaluating deals that never close. For a firm seeing eight to twelve serious packages a week and rejecting most of them, that is several hours a week of skilled time spent producing rejections. Add the meeting, and a small firm can lose the better part of a day to deciding what not to work on.
The framework attacks both. A written screen lets any one person kill a deal the day it lands, so it never reaches the meeting. What survives to Thursday is a short, ranked list — and the meeting turns from a triage session into a decision on real candidates. You are not screening faster in the meeting; you are removing screening from the meeting entirely. That is the difference between minutes and meetings.
The framework: three gates, cheapest first
The framework is three gates run in order. The principle that makes it fast is simple: kill at the cheapest gate a deal can fail. Each gate costs more effort than the last, so you never spend Gate 3 judgment on a deal that a Gate 1 fact would have killed for free.
| Gate | Question it answers | Effort | Typical kill rate |
|---|---|---|---|
| 1 — Structural knockouts | Does this deal violate a hard rule? | Seconds | High |
| 2 — Economics screen | Do the numbers clear our floor? | Minutes | Moderate |
| 3 — Human judgment | Is the story worth pursuing? | Real thought | Low |
Most deals die at Gate 1. A meaningful share of the rest die at Gate 2. Only the deals that clear both reach a person for real judgment, and by then the list is short enough that judgment is affordable. This ordering is what the broader deal-analysis playbook frames as the top of the funnel: triage the blast, then underwrite only the survivors.
Gate 1: Structural knockouts
Gate 1 is a checklist of hard rules that require no analysis to apply — only a fact match. These come straight from your buy box, the written statement of what your firm buys and never buys. If you do not have one on paper, this is the first thing to build, because without it there is nothing for the framework to test against.
Structural knockouts are the deal-breakers that no price fixes:
- Geography — outside your target markets.
- Asset class — a product type you do not touch.
- Deal size — below or above the range you can fund or want to hold.
- Position — you buy stabilized, this is ground-up; or you buy value-add, this is a trophy at a premium.
- Hard exclusions — flood zone, ground lease, environmental flag, a market you have blacklisted.
Any one of these is an instant kill. No cap rate is worth a deal in a market you have decided not to operate in. Gate 1 is deliberately blunt, and that bluntness is the point: it lets you clear most of a week’s inbox in the time it takes to read the header of each package.
Gate 2: The economics screen
Deals that survive Gate 1 fit what you buy. Gate 2 asks whether this one clears your economic floor at a first approximation. This is not underwriting — you are not building a ten-year cash-flow model. You are checking a handful of numbers against thresholds you set in advance.
For most firms that is a short list: in-place or stabilized cap rate, price per unit or per square foot against your market comps, a rough debt-service coverage test at today’s rates, and the going-in yield against your minimum. Pull the numbers, compare to the floor, and sort the deal into three tiers rather than a binary:
- Kill — misses the floor with no realistic path to it.
- Watch — close, or dependent on an assumption you would need to verify.
- Pursue — clears the floor and earns a full underwrite.
The tiering matters because a rigid pass/fail throws away the deals that are wrong today but worth a call if a number moves. “Watch” is a holding pen, not a rejection. What Gate 2 does not do is decide the deal — it decides whether the deal is worth the hours a real model costs. The model itself, and the debate over whether a purpose-built tool beats a spreadsheet for it, is the subject of our look at AI-assisted underwriting against a tool like Argus.
Gate 3: Human judgment on the survivors
By Gate 3 the list is short — the handful of deals a week that fit the box and clear the economics. This is where judgment belongs, and where it is finally affordable because you are spending it on candidates, not on the whole inbox.
Gate 3 is the part no framework and no model should automate: the story. Is the sponsor credible? Is the upside real or a broker’s pro forma? Does this deal fit the portfolio you are trying to build, or is it just a good deal in the abstract? Is there a relationship or a market signal that makes it worth more to you than the numbers show? These are the questions a principal is paid to answer, and they are exactly the questions that got drowned out when the team was using the meeting to reject deals that should have died at Gate 1.
The framework’s whole purpose is to protect this gate. Every deal killed cheaply upstream is time returned to the two or three decisions a week that actually deserve a principal’s full attention.
Where AI compresses each gate
AI does not replace the framework — it accelerates the mechanical parts of each gate so the human runs the same process in a fraction of the time. The division of labor is the one the small-firm manifesto argues lets a lean shop out-operate a bigger one: let the machine clear the busywork, keep the judgment human.
- Gate 1 — an AI first pass reads each inbound package and checks it against your written knockouts, flagging market, asset class, size, and any hard exclusion. A model with a clear buy-box prompt can clear a week of inbox against structural rules in the time it takes to make coffee. You confirm the kills; you do not hunt for them.
- Gate 2 — this is where document extraction earns its keep. The tool pulls the cap rate, price-per-unit, rent, and expense lines out of the OM and T-12 and lays them against your thresholds, tiering each deal Kill / Watch / Pursue. The discipline that makes this trustworthy — every extracted number citing the page and line it came from — is what turns a fast read into one you can act on. A screen that fabricates a missing NOI is worse than no screen, because it looks finished.
- Gate 3 — AI drafts, a person decides. A model can summarize the survivors, surface the questions worth asking, and pre-fill a one-page brief. It cannot judge a sponsor or a story, and it should never be asked to.
Two guardrails hold across all three gates. First, verify every number a model pulls before it informs a decision; current tools are strong on clean documents and weaker on the scanned, typo-ridden files that fill a real inbox. Second, deal packages are confidential and often under NDA, so settle where your data is stored, whether it trains a shared model, and how you delete it before you upload the first OM. What this looks like as a running system is laid out in our walkthrough of deal-screening automation from inbox to ranked pipeline.
The kill-reason log: how the framework sharpens itself
The one habit that separates a framework that decays from one that improves is logging why each deal died. One line per kill — “wrong market,” “cap rate 80 bps under floor,” “sponsor track record too thin” — takes seconds and builds a record no vendor can sell you.
That log does two things over a quarter. It tells you whether your buy box is calibrated: if you keep killing deals for a reason that is not on your written knockout list, add it. And it tells you whether your floor is right: if every deal in a market is missing your cap-rate threshold by the same margin, either the market has moved or your number has. The kill-reason log turns screening from a static filter into a feedback loop that gets more accurate the more deals you run through it — the compounding advantage a bigger firm’s static process never captures.
Run it on what you own before you buy a platform
The framework is tool-agnostic by design. It works on a printed checklist and a spreadsheet. It works better with the AI model your firm already pays for — a well-written buy-box prompt over ChatGPT, Claude, Gemini, or Microsoft Copilot will run Gate 1 and a first-pass Gate 2 on most deals with tools you already own. And only once you have run it by hand long enough to know your real volume and your real bottleneck does a dedicated screening platform earn the conversation.
That order protects you from the most common mistake in this category: buying a platform to solve a problem you have not defined. Get fluent first, and two things happen. You find out whether your volume even justifies a platform — many small firms discover the prompt-plus-spreadsheet workflow covers them — and if you do buy, you are a far sharper buyer, because you know exactly which gate the software is meant to accelerate.
On cost, keep the ranges honest. A hands-on fluency workshop to get a team prompting the framework runs in the low-single-digit to low-five-figure range; off-the-shelf screening subscriptions are a monthly per-seat or per-firm fee; a custom-built workflow is a project, generally in the tens of thousands to low six figures depending on scope. Price the twelve-month total with your own hours in it, not the sticker — the maintenance tail is where the real cost lives, and a tool nobody maintains decays into noise within a quarter.
The one-page screening scorecard
Turn the framework into a single sheet the whole team runs the same way. Every deal gets scored top to bottom; the first failed gate ends the scoring.
| Step | Test | Outcome |
|---|---|---|
| Gate 1 | Market, asset class, size, position, hard exclusions | Any fail → Kill, log reason |
| Gate 2 | Cap rate, price/unit vs comps, DSCR at today’s rate, going-in yield vs floor | Miss floor → Kill · Close → Watch · Clear → Pursue |
| Gate 3 | Sponsor, story, portfolio fit, upside credibility | Principal decides: underwrite or pass |
| Every kill | One-line reason | Logged for the quarterly buy-box review |
A deal that reaches Gate 3 has already earned real attention; a deal killed at Gate 1 or 2 never touches the meeting. That is the entire win: the Thursday call sees a ranked shortlist of survivors, not a week of unscreened inbox.
Frequently asked questions
What is a deal screening framework?
A deal screening framework is a written, ordered sequence of tests you run every inbound commercial real estate deal through to decide, quickly, whether it is worth underwriting. The version most useful to a lean firm has three gates run cheapest-first: structural knockouts from your buy box, an economics screen against your return floor, and human judgment on the few survivors. Its purpose is triage, not valuation — it sorts a flooded inbox into kill and keep so your underwriting hours go only to deals that fit and clear your numbers.
How is deal screening different from underwriting?
Screening answers “is this worth my time”; underwriting answers “what is it worth.” Screening is fast and approximate — it only has to reject deals confidently enough that you never look at them again. Underwriting is precise, because its output prices a bid and moves money. The two have different accuracy bars, and confusing them is the classic small-firm mistake: building a full cash-flow model on a deal that a two-minute screen would have killed. Screening protects underwriting by making sure the expensive work goes only to the deals that earned it.
What should be in a CRE buy box?
A buy box is the written statement of what your firm buys and never buys, and it is the input the whole framework tests against. At minimum: target markets, asset classes you touch, deal-size range, position (stabilized, value-add, development), a minimum return or yield floor, and hard exclusions (flood zone, ground lease, environmental flags, blacklisted markets). Most small firms carry this in a principal’s head; getting it onto one page is the single highest-impact step, because a screen can only rank deals against criteria that have been written down.
How do you kill a deal quickly without missing a good one?
Kill at the cheapest gate a deal can fail, and use three tiers instead of a binary. Structural knockouts reject deals that break a hard rule — no price fixes a wrong market, so those are safe instant kills. The economics screen sorts the rest into Kill, Watch, and Pursue, where “Watch” is a holding pen for deals that are close or hinge on one assumption worth verifying. That middle tier is what keeps you from discarding a deal that is wrong today but right if a number moves. Logging why you killed each deal is the backstop: over a quarter it shows whether your criteria are too tight.
Can AI screen commercial real estate deals accurately?
For the mechanical parts of a screen, yes — with verification. AI can read a package, check it against written knockouts, and extract the numbers for an economics screen far faster than a person, and modern tools are strong on clean, well-formatted documents. The limits are real: they are weaker on scanned, inconsistent, typo-ridden files, and a confident misread of a number is more dangerous than a visible error because it looks finished. The rule is grounding — every extracted figure should cite the page and line it came from — and human verification of any number that informs a decision. AI accelerates the screen; it does not make the call.
How many deals does a small firm actually reject at screening?
Most of them. A lean firm serious about its buy box kills the large majority of inbound at or near screening — many deals fail a structural knockout the moment they arrive, and a further share miss the economics floor. That high rejection rate is exactly why the framework matters: if you are going to say no to most of what lands, the cheapest and fastest possible “no” is worth engineering. The waste is not in rejecting deals; it is in rejecting them slowly, in a meeting.
Do we still need a weekly deal-review meeting?
You still need the meeting, but the framework changes what it is for. Without a written screen, the meeting becomes the screen — a team talking its way to a “no” that one person could have reached alone. With the framework, screening happens the day each deal lands, and the meeting sees only the ranked shortlist of survivors. The call turns from triage into a real decision on real candidates, which is a far better use of getting the whole team in a room.
Should we buy a screening tool or build one?
Run the framework on what you already own first. A buy-box prompt over an AI model you already pay for, plus your spreadsheet, will run the first two gates on most deals and tell you your real volume and your real bottleneck. Only then does a purchase make sense — and the decision turns on how standard your process is. Buy off-the-shelf if a product fits your asset classes and volume; build custom only when your workflow is specific enough that no product ranks it well. Getting fluent first makes you a sharper buyer either way, because you know exactly which gate the software is meant to accelerate.
How do we keep confidential deal data safe when screening with AI?
Settle the data terms before you upload the first package. Deal packages are confidential and often under NDA, and a small firm without an IT department has the contract as its main safeguard. Get three answers in writing: where the data is stored, whether your uploads train a shared model, and how you delete and export your history. “Your data is secure” is marketing; a data-processing addendum that commits the vendor not to train on your uploads and to delete on request is enforceable. Treat vagueness on any of the three as a reason to walk.
How do we know the framework is working?
Track two things. First, where deals die: if the mix of kill reasons is stable and matches your buy box, the framework is calibrated; if you keep killing deals for a reason that is not on your written list, the box needs updating. Second, what reaches the meeting: a healthy framework delivers a short, ranked shortlist of genuine candidates, not a pile of unscreened inbox. If Thursday still feels like triage, screening is not happening upstream. The kill-reason log is the instrument for both — it turns the framework from a static filter into a process that sharpens every quarter.
Where to start
The first move is not choosing a tool. It is writing your buy box on one page and drawing the three gates around it — then running a single week’s inbox through them by hand to see how much of the meeting disappears. Most firms are surprised by how many deals die at Gate 1 the moment the rule is written down instead of remembered. A free AI-readiness assessment gives you that read: a short working session that looks at your deal flow, your buy box, and where your screening hours actually go, then returns an honest recommendation on whether the tools you already own, an off-the-shelf platform, or a custom workflow is the right next step. Book a free AI-readiness assessment before you spend a dollar on software your framework may not need.
Arthur Wandzel
