Due diligence

Everything a buyer has read so far — the teaser, the CIM, the numbers behind
an LOI — is the seller's own account of
the business. Due diligence is where that account gets checked against reality: the actual
contracts, the actual accounts, the actual code, the actual customers. It is rarely a single
pass/fail test. Far more often, what diligence turns up changes the price, changes the structure,
or changes what the seller has to promise in the eventual contract, rather than killing the deal
outright.

For a founder, diligence is the point where the story you told has to hold up under someone else's
scrutiny, and where gaps you knew about but hoped would not come up, come up. For a buyer, it is
the last real chance to find out what they are actually buying before they are legally committed
to it.

In AI companies specifically, diligence increasingly reaches into territory that older playbooks
were not built for: where training data actually came from and whether the company had the right
to use it, whether model performance claims hold up outside a curated demo, and how much of the
product's value depends on a foundation model the company does not control and could lose access
to or see repriced.

Diligence is usually organized into workstreams, run partly in parallel: financial (do the
numbers reconcile, is revenue recognized honestly, what do the underlying contracts actually
say), legal (corporate structure, material contracts, litigation, IP ownership),
commercial (customer concentration, churn, competitive position), and increasingly
technical (architecture, security, code quality, and for an AI company, data provenance and
model dependency). Each workstream has its own specialists, often external ones the buyer brings
in specifically for the deal.

The buyer's team works primarily from the data room — the controlled
repository of documents the seller has assembled — supplemented by management interviews and,
often, follow-up requests for things that were not anticipated. What diligence finds gets carried
into the SPA in one of two ways: as an adjustment to price, or as a specific
promise or protection written into the contract, frequently landing in the
disclosure schedule as a named exception to a general promise.

Diligence takes as long as it takes to cover the workstreams the buyer considers material — there
is no fixed period that applies generally, and estimating one in advance tends to mislead more
than it informs.

A finding is not automatically a deal-killer, and treating it like one wastes leverage. Most
findings result in a price adjustment or a new contractual protection, not a collapsed deal. A
seller who panics at the first flagged issue negotiates worse than one who understands this is
the normal, expected texture of the process.

Prepare the data room before diligence starts, not during it. A buyer who has to keep asking
for documents that should already be organized starts forming an impression of how the company is
run generally — and that impression bleeds into how they read everything else they find.

Know your own weak points before the buyer finds them. Customer concentration, a messy cap
table, an unresolved dispute, an over-reliance on one engineer or one model vendor — surfacing
these yourself, with an explanation, reads very differently from having them discovered.

Technical diligence on an AI product goes further than a code review. Expect real questions
about where training data came from and whether it was properly licensed, how dependent the
product is on a specific foundation model, and whether performance numbers were measured on
representative data rather than a favorable sample.

Diligence findings should map to specific contract language, not vague assurances. If
something is disclosed and resolved during diligence, make sure it is reflected precisely in the
disclosure schedule — a verbal explanation given to the buyer's team is not a substitute for what
is actually written down.

Due Diligence is a named stage in GetDeal's investment track (and the equivalent document
exchange sits inside the Data Room stage on the M&A track), so both sides always know whether the
deal is currently in this phase rather than inferring it from an email thread going quiet. The
data room itself — per-document access, watermarked downloads, logged
reads — is the actual mechanism this stage runs on.

The Playbook explains what a founder needs to have ready before this stage opens, which is worth
reading well before a buyer starts asking.