Every deal screen makes two kinds of mistakes, and only one of them ever gets reported. False negatives in deal screening, the deals that fit the buy box and got killed anyway, leave no evidence behind. A false positive announces itself: an analyst builds a model, the numbers fall apart, and somebody says out loud that the deal never should have advanced. A false negative produces a quiet inbox and a closing somewhere else. That asymmetry, not the raw error rate, is what makes automated screening dangerous. A team tunes against the errors it can see, and there is only one of those.
Key Takeaways
A false positive costs a bounded number of analyst hours and reports itself. A false negative costs the full economics of a deal the firm would have closed and reports nothing.
Because only one error is observable, every tuning cycle pushes an automated screen tighter. The drift is structural, not a lapse in anyone's judgment.
Under a standard expected-cost threshold, a team whose false negative costs 375 times a false positive should advance any deal with better than a 0.27 percent chance of fitting. Almost no screen sits anywhere near that loose.
Altus Group reported U.S. transaction dollar volume up 9.4 percent year over year in Q2 2026 while property transaction counts stayed roughly flat. The same number of assets is trading at higher prices, so each missed fit costs more than it did a year ago.
A screen that has never reported a false negative is not accurate. It is unaudited.
What is a false negative in automated deal screening?
A false negative is a deal that fit the buy box but got rejected at the initial screen. Unlike a false positive, which announces itself when the model falls apart, a false negative leaves no trace in the pipeline. The deal simply disappears from the firm's record and closes with somebody else.
A deal screen is a classifier whether or not anyone built it as one. It takes an offering memorandum, applies criteria, and outputs advance or reject, producing the four outcomes below. Automation does not change the error types. It applies the same criteria at scale, converting an occasional misjudgment into a systematic one.
Screen output | Deal fit the buy box | Deal did not fit |
|---|---|---|
Advanced | True positive: the pipeline works | False positive: analyst hours burned on a model that dies |
Rejected | True negative: the screen did its job | False negative: a deal the firm would have closed, gone |
Three of those four cells produce evidence inside the firm. The fourth produces nothing. That is the entire problem.
Why do false negatives cost more than false positives?
A false positive costs analyst hours, which are bounded and recoverable. A false negative costs the entire economics of a deal the firm would have closed, which is neither. Measured against each other, the asymmetry usually runs in the hundreds to one, and it runs in the direction nobody tunes for.
Work the numbers from stated inputs. A false positive gets modeled, discussed, and killed: four analyst hours at a fully loaded $125 per hour, so $500. Now price a false negative. Say the total value of one closed deal to the firm, across acquisition fee, asset management fees, and promote over a five-year hold, is $1.5 million, and one in eight advanced deals reaches a closing. The expected value of correctly advancing a real fit is $1.5 million divided by eight, or $187,500.
The ratio is $187,500 to $500, or 375 to 1. Under those inputs the firm should absorb 375 wasted models to avoid killing one deal that fit. The cost of 375 false positives and the cost of one false negative are the same number.
In a Monday pipeline meeting the 375 false positives are visible and attributable. The one false negative is not in the room. Every incentive in that review points toward tightening the screen. The arithmetic points the other way.
Why does an automated screen drift toward rejecting good deals?
Because tuning feeds on visible error. Every false positive produces a complaint, a wasted model, and a name attached to it. Every false negative produces silence. A team that adjusts its criteria in response to the errors it can see will tighten the screen, run after run, in one direction only.
This is a feedback loop problem, not a modeling problem. Each tightening removes false positives and true positives together, and the team observes only the first. Six months of that and the screen is rejecting deals the firm would have bought, with a clean record of improvement to show for it.
Tuning signal | What the team observes | What it does to the screen | What it does to recall |
|---|---|---|---|
A modeled deal dies at underwriting | Wasted hours, named deal | Tightens a criterion | Falls |
A deal dies at investment committee | Wasted hours plus credibility | Tightens harder | Falls further |
A rejected deal fit the buy box | Nothing | Nothing | Already fell |
A rejected deal closes with a competitor | Sometimes, months later, anecdotally | Rarely anything | Already fell |
The fourth row is the only correction mechanism most firms have, and it arrives too late and too rarely to matter. It is the same failure behind treating extraction accuracy as a single number rather than precision and recall: one error type is easy to count, so it becomes the whole measurement.
How do you measure an error you never observe?
You have to manufacture the observation, because the pipeline will not produce it. Three methods work: sample rejected deals and underwrite them anyway, match rejections against recorded sales, and log the specific criterion that killed each deal. All three cost time. None of them is optional if the screen is automated.
The hold-out sample is the cleanest. Take a fixed share of rejected deals, five percent is enough to start, and route them to underwriting as though they had passed. Whatever share turns out to fit is a direct estimate of the false negative rate. It is the only method that produces the counterfactual the pipeline destroys.
Recorded-sale matching is slower and cheaper. Pull a quarter of rejections and check which traded, and at what price. A rejected deal that closed at a basis the firm would have paid is a false negative with a public record attached.
Criterion logging should be table stakes. Every rejection records which criterion triggered it, so when a hold-out sample or a recorded sale surfaces a miss, the log says which rule was wrong. Without it a firm knows its screen is too tight but not where. A deal screening memo that captures the killing criterion turns each rejection into a testable claim rather than a disappearance.
Where should the screening threshold actually sit?
At the point where the probability a deal fits exceeds the cost of a false positive divided by the combined cost of both errors. For most acquisitions teams that threshold lands under one percent, far looser than instinct allows. The binding constraint is not the threshold. It is underwriting capacity.
The rule comes from expected-cost minimization. Advancing a deal costs the false positive price when it does not fit; rejecting it costs the false negative price when it does. The two are equal at a probability of C_FP divided by the sum of C_FP and C_FN. With the inputs above that is $500 divided by $188,000, or 0.27 percent. Any deal with better than a one-in-375 chance of fitting is worth advancing on arithmetic alone.
No firm can act on that literally, and that is the honest limit on the argument. Analyst hours are finite. Past some volume the marginal false positive displaces a real fit that now waits a week for attention. So the response to the asymmetry is not to loosen the screen until everything passes. It is to make rejection cheap to reverse and move the expensive judgment later in the funnel, where more information exists. Screen only on the criteria that are genuinely disqualifying and verifiable from the offering memorandum. This is why a buy box has to be written as machine-readable criteria instead of adjectives: a queryable filter can be re-run against yesterday's rejections when a criterion changes, and a discarded pile cannot.
Market conditions raise the stakes. MSCI Real Capital Analytics recorded $113.7 billion in U.S. investment volume in Q2 2026, up 9 percent year over year. Altus Group reported dollar volume up 9.4 percent in the same quarter while property transaction counts stayed roughly flat. More capital is moving through a stable number of assets, and when available deals are not growing, rejecting one that fit costs more.
Frequently Asked Questions
What is a false negative in deal screening? An offering memorandum that met the firm's buy box but was rejected at the initial screen. It never enters the pipeline, so the firm has no record it happened and no mechanism for discovering the mistake.
Why are false negatives worse than false positives in acquisitions? A false positive costs a fixed number of analyst hours and reveals itself. A false negative costs the full expected value of a deal the firm would have closed and reveals nothing. The ratio typically runs into the hundreds to one.
How can a firm detect false negatives in its screening process? Route a fixed sample of rejected deals through underwriting anyway and count how many fit. Add a quarterly check of which rejections traded and at what basis. Both require logging the criterion behind each rejection.
Does tightening the buy box reduce screening errors? It reduces one kind. Tightening cuts false positives and raises false negatives at the same time. Whether that trade helps depends on the relative cost of the two errors, which most firms have never calculated.
Conclusion
Automated screening does not create the false negative problem. It scales it, and removes the human inconsistency that occasionally let a good deal through a bad rule. A screen applying the same criteria to every offering memorandum applies the same mistake to every one, silently, until someone builds the measurement that makes it visible.
The work is not choosing better criteria. It is deciding to observe the error the pipeline is designed to erase, accepting that the observation costs underwriting hours, and setting the threshold from the cost ratio rather than from how many wasted models the team can remember. As precision in deal screening becomes the constraint a busier market tests, the firms with an edge will be the ones that can state their false negative rate out loud. Most cannot, and that silence is not evidence the number is small.