Most traction claims in a pitch deck are true in some sense. The trouble is the sense. "20 enterprise customers" can mean 20 paying contracts, or 8 paying customers and 12 unpaid pilots. "$50k MRR" can include a one-time setup fee. "40% month-on-month growth" can be one good month from a small base. None of that is necessarily dishonest; founders summarise, and decks are short.
Knowing how to verify startup traction is mostly about turning a summary back into specifics. What follows is a claim-by-claim method, a table of common claims and the evidence that checks each one, how to run customer reference calls, and a simple labelling scheme that keeps your team honest about what has actually been checked. It explains process, not whether any company is a good investment.
What counts as traction at pre-seed and seed
Traction is evidence that customers want what the company makes. At pre-seed it might be a waiting list, design partners or a handful of paying pilots. At seed it is more often recurring revenue, retention and a repeatable way to win customers. The stage changes what you expect to see, but not how you check it.
Every claim worth relying on has five parts. If one is missing, you do not yet know what the claim means:
- The assertion: what is being claimed, in the founder's words.
- The metric definition: what is counted and what is excluded.
- The period: which month, quarter or date range.
- The amount and units: currency, customers, users, percentage.
- The source: the slide, form answer, interview moment or document it came from.
Andreessen Horowitz's widely read list of 16 startup metrics is a good reference for definitions. It warns that bookings and revenue are often used interchangeably but are not the same, that annual recurring revenue should exclude one-time and professional service fees, and that companies have almost unlimited definitions of what an active user is. Each of those is a place where a reasonable-sounding number can mean something other than you assume.
How to verify startup traction, claim by claim
Work through the deck and any application answers once, slowly, and build a short list of the claims that would change your view if they were wrong. Usually that is five to ten claims, not fifty.
- 1List each material claim with its source: the slide number, the form answer or the call where you heard it.
- 2Ask the founder to define it. A short written question works well: how do you define an active customer, and which period is this figure for?
- 3Request evidence that matches the definition. Ask for exports, not screenshots, where you can, and only for what you need.
- 4Compare the evidence with the claim. Does the period match? Do the totals add up? Are pilots counted as customers?
- 5Corroborate from a second, independent source where the claim matters most: a customer call, a public filing or a document from a third party.
- 6Label the result and record what you checked, so the next person does not redo the work.
Keep requests proportionate to the stage. A pre-seed founder with three pilots should not receive a 60-item questionnaire. Our diligence information request list is a starting point you can cut down, and the guide to the VC due diligence process places traction checks within the wider diligence order.
Common traction claims and the evidence that checks them
The table pairs the claims you will see most often with the definition question to ask, the evidence that usually settles it, and the gap that most often turns up. Adapt it to the business model: a marketplace, a hardware company and a developer tool will each have their own variants.
| Claim | Ask the founder to define | Evidence to request | Common gap |
|---|---|---|---|
| Monthly recurring revenue (MRR) | What is included, and for which month | Billing system export by customer for the last 6 to 12 months | One-time or setup fees counted as recurring |
| Annual recurring revenue (ARR) | Contracted recurring revenue or the latest month times 12 | Contracts and billing export; how ARR was calculated | A run rate from one strong month presented as ARR |
| Revenue | Recognised revenue or bookings | Bank statements or accounting export for the period | Signed but unbilled contracts counted as revenue |
| Paying customers | What makes a customer paying | List of customers with contract start dates and amounts | Free pilots or trials counted as customers |
| Pilots and letters of intent | Paid or unpaid, and what converts them | Signed pilot agreements or letters, with terms | Non-binding interest described as a commitment |
| Active users | What action counts as active, and over what window | Product analytics export with the definition shown | Sign-ups or downloads presented as active users |
| Growth rate | Which months are compared | The monthly series behind the percentage | One good month from a small base |
| Retention or churn | Gross or net, logo or revenue | Cohort table by start month | Net figures that hide lost customers behind upsells |
| Customer acquisition cost | Paid or blended, and which costs are included | Marketing spend and new customers by month | Blended figures that include organic growth |
Two of those gaps come straight from the a16z definitions: gross churn shows the actual loss to the business while net revenue churn understates it, and blended acquisition cost does not show whether paid acquisition works. Glossary entries for MRR, ARR, net revenue retention and CAC payback give short definitions you can share with founders.
Customer and reference calls
Documents show that money moved. Customers tell you why, and whether it will keep moving. In the Gompers, Gornall, Kaplan and Strebulaev survey of 885 venture investors, the average firm called 10 references during due diligence; early-stage firms called 8 on average and late-stage firms 13. The same survey found the average deal took 83 days to close, with 118 hours spent on due diligence.
At pre-seed and seed you will usually call fewer, and that is fine. What matters is that the calls test the claims on your list.
- Agree the calls with the founder and ask for a mix: their happiest customer, a recent one and, if they will share it, one who left or did not convert.
- Ask the customer to describe the problem in their own words before you mention the product.
- Check the specifics: how long they have paid, roughly how much, who uses it, and what they used before.
- Ask what would make them stop using it. The answer often tells you more than the praise.
- Write notes the same day and link each note to the claim it supports or contradicts.
Respect the founder's relationships. Calling a customer without telling the founder can damage a sale they are in the middle of. Our reference check questions template has questions for customers, former colleagues and co-investors.
Label what you know, and keep what changed
A simple set of labels stops a team from treating a number as checked because it has been repeated in three documents. Use four, and show the label every time the claim appears:
- Founder reported: stated by the founder in the deck, a form or a call, not yet checked.
- Corroborated: backed by a document, a data room file or your own check.
- Disputed: two sources disagree. Keep both versions and cite each.
- Unresolved: not enough evidence either way yet. It becomes a question or a task.
A worked example
A deck says the company has 20 enterprise customers. In the interview, the founder explains that 12 of them are unpaid pilots. Do not overwrite the first claim with the second. Keep both, mark the deck figure as disputed, record the clarified figure as founder reported, and create a task to see the 8 paid contracts and the pilot terms. When the contracts arrive, the paid figure becomes corroborated, and anyone reading the record later can see how the number evolved and why.
Common mistakes
- Asking for everything at once instead of the evidence for the claims that matter.
- Accepting a screenshot of a dashboard when an export would show the definition and the period.
- Overwriting a claim when the number changes, which loses the history and the reason.
- Treating a difference between deck and data as dishonesty before asking. Most differences are definitions.
- Letting a disputed claim travel into a committee memo without its label.
Where Delsivo fits
Delsivo is an AI investment workspace for venture funds, at the design partner stage; nothing is generally available yet. Its claim model follows the method above.
In the Pilot, Delsivo is designed to extract claims from a PDF deck with a citation to the slide, and to keep each claim with its definition, period, amount, units, sources, owner and one of the four evidence labels in the claim tracker. Earlier statements are kept when facts change. Basic follow-up forms and scoped file requests let you ask the founder for exactly the evidence a claim needs, through your branded portal, with reminders only for outstanding items.
In the first release, live AI founder interviews are planned to ask follow-up questions out loud, drawn from the pitch and open claims, and data rooms with permissions and activity are planned for the documents. AI suggests claim updates; an analyst accepts or dismisses each one. See claims and evidence for how the pieces connect.
Sources
Delsivo publishes this guide and makes software for venture funds, so we have a stake in the topic. Facts about other companies, research and regulations come from the sources listed, on the dates shown, and can change. Corrections are welcome through our contact page.