Pre-Screening at Scale: Scoring Startups Before the First Call
Pre-screening is the layer where an investor decides which companies get an hour. It is the highest-leverage process in the entire dealflow stack and it is almost always the least designed one.
The reason is time. Building a real screening process feels like overhead when you can just read the deck. It stops feeling that way somewhere around forty inbound companies a month, at which point the undesigned process quietly reverts to "whoever came through a warm intro".
This is how to build one that survives volume.
Key takeaways
- Screening has one job: decide what deserves an hour, in under five minutes, consistently.
- Written criteria beat judgement at this layer. Judgement is for the hour, not for the five minutes.
- A score is only comparable if the inputs are comparable. Scoring a PDF measures the PDF.
- Separate thesis fit from quality. They are different decisions and conflating them produces bad passes.
- Track your pass reasons. The pattern in your own passes is the most useful data you will ever have about your process.
The two-gate model
Most screening failures come from asking one question when there are two.
Gate 1: Thesis fit. Is this something I could invest in? Sector, stage, geography, cheque size, structure. Binary, takes thirty seconds, no judgement required.
Gate 2: Quality. Is this something I should invest in? Four dimensions, judgement required, takes four minutes.
Keeping them separate matters because a great company outside your thesis should be passed on fast and referred on, not agonised over. And a mediocre company squarely in your thesis should be passed on for quality reasons, not waved through because it fits.
Investors who run one combined gate systematically make the second error.
Gate 1: thesis fit in thirty seconds
Write your thesis as a filter you could hand to someone else:
| Criterion | Example |
|---|---|
| Sector | B2B SaaS, fintech infrastructure, developer tools |
| Stage | Pre-seed and seed. Priced or SAFE. |
| Geography | Europe, UK, and US companies with a European founder |
| Cheque | $100K - $500K |
| Ownership | Comfortable following; will not lead |
| Hard exclusions | No crypto trading, no regulated lending, no single-founder companies |
If you cannot write this down in ten lines, you do not have a thesis, you have preferences. Preferences cannot be applied consistently, cannot be delegated, and cannot be published so that founders self-select.
That last point is underrated: a published, specific thesis is a screening tool. It pre-filters your inbound before it reaches you.
Gate 2: the four-dimension read
Team, product, market, deal, each scored one to five. This is the same framework founders should be applying to themselves, described in the fundraising readiness score.
Team. Is there a specific reason these people win this problem? Not "impressive backgrounds" - a stated, verifiable link between what they have done and what they are doing.
Product. Does something exist, is its state honestly represented, and can a technical reader sanity-check the approach?
Market. Is the buyer defined specifically enough to count? Is the sizing derived rather than asserted? Are competitors named?
Deal. Is the round structured, priced and explained? Does the use of funds buy a milestone that justifies the next round?
Score, then apply a threshold. A common rule: nothing below 3.0 gets a call; nothing with a 1 in any dimension gets a call regardless of average. The second half matters - a company scoring 5/5/5/1 has a fatal problem, and the average hides it.
The comparability problem
Here is the part most screening processes never solve.
If you score Company A from a twelve-slide deck and Company B from a Notion page and Company C from a forty-minute call, you have not scored three companies. You have scored three artefacts. The company with the best deck scores highest, which selects for design budget and founder writing ability rather than for company quality.
For scores to be comparable, the inputs must be comparable: the same fields, present for every company, from the same source.
This is the whole argument for structured company records, and it is why the industry has not fixed screening despite everyone knowing it is broken. It cannot be fixed on the investor side alone. It requires companies to publish a standard field set, which requires a standard to exist. We go into why this remains unsolved in you cannot compare two startups today.
Where structured records do exist, the screening arithmetic changes completely: in our experience a minute or two per company instead of twenty, applied identically, with a score that means the same thing across the whole pipeline.
Where AI helps, and where it does not
AI is genuinely good at the mechanical parts of screening: extracting claims from a document, normalising metrics into comparable units, flagging internal contradictions, checking whether required information is present at all, and drafting a first-pass structured summary.
It is not good at the part that matters: judging whether this team wins this market. Nor should it be trusted to, since the judgement is the investor's job and the accountability cannot be delegated.
The useful division:
| AI does | You do |
|---|---|
| Extract and normalise the facts | Decide what the facts mean |
| Flag missing or contradictory information | Weigh the risk you are willing to take |
| Produce a comparable first-pass score | Override it, and record why |
| Draft the memo skeleton | Write the thesis and the decision |
An AI pre-score that you cannot override, or whose reasoning you cannot inspect, is worse than no score. The value is in the consistency of the first pass, not in the delegation of the call.
Close the loop on your passes
The single highest-return habit in screening: record the reason for every pass, in a controlled vocabulary.
Not free text. Six to eight categories: out of thesis, team gap, market too small, no evidence of demand, deal structure, valuation, too early, cannot verify claims.
After a hundred passes you can answer questions you currently cannot:
- What fraction of my inbound is out of thesis? If it is 80%, my public thesis is not doing its job.
- How many passes were "cannot verify claims"? That is a process cost, not a company quality signal.
- Are my "too early" passes converting for other investors twelve months later? That is the most expensive category and nobody tracks it.
This is also the only way to audit your own bias. If 90% of your calls come from one channel, you have a sourcing problem dressed up as a screening outcome.
Frequently Asked Questions
Typically a fast thesis-fit check followed by a judgement pass on team, product, market and deal. The rigour with which the criteria are written down varies enormously.
Thirty seconds for thesis fit, four to five minutes for a quality read. If it takes longer, the input format is the problem, not the company.
For extraction, normalisation and a comparable first pass, yes. For the investment judgement, no. Keep the reasoning inspectable and the override in human hands.
Highly variable. As a rough rule of thumb, expect a small minority of screened companies to earn a first call, and only a handful of first calls to end in an investment.
Read next
The Modern Dealflow Stack: From Inbox Chaos to a Scored Pipeline
The Investment Memo, Standardised: What Belongs in a Pre-Seed IC Memo
You Cannot Compare Two Startups Today. Here Is Why That Is a Data Problem
The Fundraising Readiness Score: How Investors Rate You Before the First CallRaiseTalks gives investors a filterable catalogue of startups built on the same structured field set, up to 143 fields once access is granted, plus a one-click AI pre-score across team, product, market and deal for approved data rooms. Comparable by construction. See the investor side.