The Complete Guide to Startup Screening

Startup screening is the process of turning a large pool of applications or deals into a small, ranked shortlist worth a closer look. Done well, it lets a committee spend its scarce attention on the few companies that matter instead of drowning in the many that do not.

An accelerator cycle or an open call can bring 300 applications. A partner cannot read 300 decks with equal care, so in practice they skim, and skimming is where good companies get missed and polished-but-hollow ones get through. Screening is the discipline that replaces uneven skimming with an even, comparable standard.

This guide covers why gut calls fail at scale, what a common scoring basis is, the three questions every candidate should answer, why verification belongs inside screening, how to weight for different program types, a worked example, and what makes a shortlist defensible.

Why gut calls fail at scale

A single expert reading a single deck can be excellent. The same expert reading the ninetieth deck on a Friday is not, and neither is a panel where each reviewer scores differently and no two candidates are judged on the same axes. The failure is not a lack of expertise. It is a lack of consistency.

Three specific errors dominate unstructured screening:

• The polished-deck bias. A confident narrative and a clean design read as quality. They are not. A deck is a sales document, and its job is to look convincing whether or not the claims hold up.
• The blindness error. A reviewer weighs the three things they happened to notice and misses the fourth that mattered. A total addressable market stated at 45B goes unchallenged because no one had time to check that the reachable figure was closer to 10B.
• Fatigue drift. Scoring gets harsher or more lenient as the pile grows. The rank a company gets depends partly on when it was read.

All three come from the same root: no shared, evidence-backed basis for comparison. Fix the basis and all three shrink.

A common basis is the whole point

Screening is comparison. Comparison only works if every candidate is measured on the same axes, from the same evidence, to the same standard. That is a scoring basis: a fixed set of criteria, applied identically to all, producing one comparable result.

A common basis does three things a pile of opinions cannot. It makes candidates rankable rather than merely likeable. It makes a reviewer's reasoning visible and auditable. And it lets you defend the outcome, because "why this one and not that one" has an answer that is not "it felt stronger."

This is not about removing judgment. It is about giving judgment the same starting line for everyone. The reviewers still bring expertise. They just apply it to the same questions, in the same order, against the same evidence.

The three questions

A useful scoring basis separates three questions that gut calls tend to fuse. The Deckwise Method frames them as:

1. Is it true?

Do the material claims in the application hold up against public evidence? This is the layer most screening skips entirely. It is also the one that changes decisions, because a company that misstated its traction has told you something about how it will report to you later.

2. Is it a good company, objectively?

Traction, management, market, product and moat, financials, team, risk, independent of whether it fits you. A company can be excellent on this axis and still be wrong for your fund.

3. Is it a good company for us?

Alignment with your thesis, stage, geography, values, and your explicit anti-thesis. A perfect-fit company built on an inflated claim is a trap; a strong company that violates your anti-thesis is still a no.

Keeping them apart matters because a company can pass one and fail another. Fused into a single impression, those distinctions disappear. Scored separately, they survive to the committee, which is exactly where they need to be.

Verification is a screen, not a formality

The first question deserves its own emphasis, because almost no screening process runs it. Most treat the deck as the input and evaluate the story. Screening on evidence treats the claims as the input and evaluates whether they are real.

Every material claim gets pulled out and checked, official registers first, then sourced web search for what registers cannot answer. Each one lands as verified, qualified, contradicted, or to-confirm. A contradiction is not a footnote. It changes the score.

One honesty rule keeps this from over-reaching: matching a company name in a register proves an entity exists, not that a specific claim is true. Verification credits the claim only when the evidence supports the claim itself. And a claim that shifts over time, a customer count from three years ago, a valuation from a prior round, is treated with care rather than scored as a contradiction just because a stale source disagrees. (For the trade-off between doing this by hand and automating it, see Manual vs AI Screening.)

Weight for the program, not for everyone

Different programs value different things. An early-stage accelerator weighs team and momentum over financials. A growth-stage fund does the opposite. A grant program may care most about mission fit and eligibility. A single fixed weighting cannot serve all of them.

So the weighting between the three questions, and between the criteria inside them, should be set to sensible defaults and then adjusted per program type. The important discipline is that the weighting is decided once, in advance, and applied to every candidate the same way. Weights tuned per candidate are not a scoring basis. They are rationalization with extra steps.

Two safeguards keep a weighted score honest:

• The wildcard. A candidate that misses the top tier overall but spikes on a single dimension is surfaced, so a standout is never buried by an average total.
• The red flag. A hard negative signal pulls the overall score down on purpose, so a serious negative cannot hide behind a strong average.

A worked example

Consider three candidates in a seed accelerator call:

• Company A has a beautiful deck, a 45B market claim, and strong-sounding traction. Verification finds the market figure counts three adjacent categories it does not serve, and one named partnership cannot be confirmed. On a gut read it topped the pile. On the evidence it drops, not because it is disqualified, but because its headline claims are softer than they looked.
• Company B is unglamorous, with a modest deck and honest, verifiable numbers. It scored middle on a skim. On a common basis, with every claim checked, it rises, because nothing about it falls apart on inspection.
• Company C spikes on one dimension, an unusually strong technical moat, but is average overall. A pure ranking buries it. The wildcard surfaces it for a human to look at.

Same three companies, same reviewers. The difference is that the second pass judged them on one basis, with the claims checked. That is screening.

A defensible shortlist

The output of screening is not a number. It is a ranked shortlist you can defend line by line, to a committee, a boss, or an LP. Defensible means each ranking arrives with the evidence and the reasoning attached: which claims held up, where the company scored and why, how it fit the thesis, what got flagged.

That defensibility is the real deliverable. The time saved is welcome, but the deeper benefit is that every yes and every no can be explained. In selection, being able to justify the decision is not bureaucracy. It is the job.

Common mistakes

• Evaluating the deck instead of the claims. The story is the sales pitch, not the evidence.
• Per-reviewer axes. If two candidates are judged on different criteria, they were never compared.
• Tuning weights per candidate. Decide the basis first, then apply it to all.
• No verification layer. An unchecked claim scored as fact is a scored fiction.
• Confusing a good company with a good fit. Score them separately or lose both.
• Treating a stale number as a lie. A metric that moved over time is not a contradiction. Check the date before you flag it.

Frequently asked questions

What is startup screening? Startup screening is filtering a large set of applications down to a ranked shortlist worth a closer look, by scoring each company on a common basis, ideally including whether its claims hold up, so a committee focuses on the few that matter.

How is screening different from due diligence? Screening ranks many companies quickly to produce a shortlist. Due diligence goes deep on the few finalists that shortlist produces. Screening is breadth; due diligence is depth. (See The Complete Guide to Startup Due Diligence.)

What makes a shortlist "defensible"? Each ranking comes with the evidence and reasoning behind it: which claims were verified, how the company scored on each question, how it fit the thesis, and what was flagged. Defensible means you can justify every yes and no to a committee or an LP.

Can screening be automated end to end? The first pass can: extract, verify, score, rank. The decision should not be. For evaluating people, human oversight is both the responsible design and what regulators expect. A good tool is a co-pilot, not a judge.

How many applications justify a real screening process? Roughly, once you can no longer read every application deeply, usually past 50 per cycle, unstructured screening degrades into skimming. That is the point to add a common basis and a verification layer.

The bottom line

Screening is where a pile of impressions becomes a defensible decision, but only if every candidate is judged on one basis, with the claims actually checked. Keep the three questions apart, verify what is verifiable, weight for the program and not per candidate, and let the humans decide on a shortlist they can defend.

How this shows up in Deckwise

Deckwise's Selection pillar takes a pool of applications and scores every one on the same basis: whether the claims hold up, the company's quality, and the fit with your thesis. Weights are set by default and adjustable per program type, versioned so a past decision can be reproduced. The result is a ranked shortlist with the scores and sources attached, exportable for the committee. The human still decides, which is the right posture under the EU AI Act for evaluating people.

Next: The Complete Guide to Startup Due Diligence, going deep on the finalists a shortlist produces.