Seed vs Series A: how the data room changes

Seed: there is little history to verify, so an investor is assessing judgement, market and team. The room supports a conversation.

Series A: there is now a track record, and the job is checking whether what you said matches what happened. The room is evidence.

This is why a Series A room cannot be a seed room with more files in it. The documents are selected to be reconcilable against each other.

Lean is correct here. Over-preparing signals the wrong thing.

• Incorporation documents, MoA and AoA
• Cap table, fully diluted, including any SAFEs or notes
• Whatever financials exist, often management accounts only
• A model with visible assumptions
• Founder agreements and IP assignments
• Early traction: whatever metrics are real
• The deck

Two or three investors, a handful of documents. Google Drive frequently handles this, and saying otherwise would be dishonest.

Do fix the IP assignments at seed. It is the cheapest moment in the company’s life to do it, and it is found missing in a large share of first institutional rounds, at which point it is a closing condition instead of an afternoon.

Everything above, plus the reconcilable layer:

• Audited financials for every completed year
• 12 months of bank statements. Reconciled to the model
• Monthly cohort and retention data, in aggregate
• Customer contracts, at least the largest by revenue
• The full statutory filing history. ROC, GST, TDS, PF, ESI
• FEMA/FDI filings if there is foreign money
• The ESOP scheme document, not only the grant list
• Employment contracts for key people
• Statutory registers reconciled to issued share certificates
• Any litigation, notice or dispute, including trivial ones

The complete list, with the India-specific items marked, is in the checklist.

Rows five and six are the operational break. Two or three readers who all see the same thing is a folder. Ten readers with different grants, including people you have not met, is access control. See who should see what.

Rarely a missing document. Usually one of four:

1. Register does not reconcile to certificates. The most common. Resolutions exist, certificates do not, or the two disagree.
2. IP not assigned. A founder or early contractor wrote code before any assignment was signed.
3. Model does not reconcile to bank statements. Not fraud, usually a definitional difference nobody wrote down. Write it down.
4. An undisclosed related-party transaction surfaces. The amount is almost never the problem. Discovering it undisclosed is.

All four are fixable in advance and expensive during. If you are 6–12 months from a Series A, this is the list to work through now.

At seed, a partner reads a handful of documents and asks you directly. At Series A, associates work through a long structured list, and most of their questions have answers already sitting in your documents.

That repetitive half is what an AI data room removes. Investors ask in plain language and get answers assembled from the documents they were granted, cited to the source document and page so the answer can be checked rather than trusted. Because the access limit applies when the system searches, an associate who was not granted the employee contracts is told so rather than receiving a summary of them.

The questions that decide the round are still the ones about judgement, and nothing answers those for you. The point is arriving at them sooner, with the reconciliation already done.

How the retrieval boundary works, and the four ways it can fail: can you trust an AI with your data room. India-specific context: data rooms for Indian startups.