What buyers actually read in a data room

A data room can hold thousands of documents. Buyers read only a few of them closely, and those few decide whether the price in the letter of intent survives to signing. As Silverpeak puts it in its H1 2026 report, “AI has become a core diligence lens,” on top of the financial and commercial questions buyers always asked.

1. ARR that reconciles to the billing system

Buyers rebuild ARR from raw billing exports and compare it with your management figure. What typically gets stripped out:

• Paid pilots and trials annualised as if they were committed.
• Customers who have already given termination notice.
• Implementation, services and custom development booked as subscription.
• Usage spikes above the committed minimum.
• Ramped multi-year contracts reported at their final-year rate.

If ARR cannot be rebuilt from the billing system, every other metric loses credibility with it.

2. Cohorts, not averages

A healthy blended net retention can hide a real problem. If 2023 customers retain at 115% and 2025 customers at 88%, the average still looks fine, but the business is getting worse. Buyers ask for monthly retention and expansion by signup cohort, and refusing to produce them is read as a warning sign.

3. Quality of earnings

Accrual accounting, correct revenue recognition and a deferred-revenue schedule are the minimum. Gross margin gets more attention than before, especially for AI products, where inference costs now sit in the cost of revenue.

4. Customer contracts

Buyers read the top contracts in full. They look for concentration (scrutiny rises above about 15% of revenue from one customer and starts to affect structure above 25%), change-of-control clauses that require customer consent, termination rights and the length of commitments.

5. AI defensibility, tested rather than asserted

Some buyers now use AI tools to build working replicas of a target’s product to see how hard it is to copy. The answer they want to find in the data room is what a copy would not have: proprietary data accumulated over years, workflows customers would have to retrain to replace, integrations into systems of record, certifications. They also expect the commercial terms of your own AI dependencies - which models, at what cost, under what data-use terms - to be documented.

6. Technology, IP and security

Signed IP assignments from every employee and contractor who wrote code, an open-source inventory, and security evidence such as SOC 2 reports, penetration tests and incident logs.

The issues that end deals early

• Cash-basis books with no accrual conversion.
• Unsigned IP assignments for contractor code.
• A customer above 25% of revenue without contractual protection.
• No security evidence at all.
• Undocumented commercial terms for the AI services the product relies on.

Build the data room before the process starts, structure it by workstream with access controls, and answer the hard questions in it before a buyer has to ask them.