A data room is not a folder. It is a set of measurements.

A claim and a measurement look identical on a slide

Diligence is a translation exercise. The company arrives with a set of claims — net revenue retention of a hundred and twelve percent, payback inside fourteen months, gross margin at seventy-one — and the reader’s job is to turn each one into a measurement: a defined quantity, computed a stated way, from a named system, reproducible by somebody who was not in the room when it was written. Almost everything that slows a process down happens inside that translation.

The difficulty is that the two are indistinguishable at the point of presentation. Both are a number in a large font with a label underneath it. The difference only appears at the second question — what counts as a customer here, and whether that same definition holds in the row above — and a company that has not answered it in advance answers it live, under time pressure, in front of the people setting its valuation. That is the ordinary route by which a correct number becomes a discount.

The fix is not more documents. Most data rooms are already too large, and the common failure is two hundred exports that nobody can tie to one another. What converts the contents into evidence is a much smaller artifact: about a dozen definitions, one query behind each, and a person’s name against both.

The definitions that have to be written down

Write these once, date them, and keep them identical from quarter to quarter. Where one has genuinely changed, say when and why in the same document rather than quietly restating history around it.

• What counts as a customer: a signing entity, a billing account or an individual site. Retention moves several points depending on which one you picked.
• When revenue is recognized, and the distance in any given month between contracted value, invoiced value and recognized revenue.
• What sits inside acquisition cost and what sits outside it — which salaries, which share of marketing, and whether customer success is acquisition or delivery.
• What makes a contract renewed rather than replaced, and how a downgrade and an upgrade in the same quarter are counted.
• What makes pipeline qualified, and which person applies the test.
• Headcount as of a date, on a stated basis: full-time equivalents, with contractors either in or out and the same answer every time.
• The period boundary — calendar or fiscal, and what happens to a contract signed on its last day.

None of these is right in the abstract, and nobody is going to argue you out of a defensible answer. What is not survivable is two answers to the same question, which is what appears when finance, revenue operations and whoever assembled the deck were each left to decide separately and none of them knew the others had.

A number somebody else can run

A definition with nothing behind it is a policy statement. The useful form is the pairing: this metric, defined this way, produced by this query, against this system of record, returning this answer for this period. Written down in that shape, a disagreement stops being an argument and becomes a re-run.

Three properties make a query hold up under examination. It reads from the system that is authoritative for the fact — billing for revenue, payroll for headcount, the contract record for terms — rather than from an export that was correct on the day somebody took it. It takes the period as a parameter, so the same code produces every quarter instead of each quarter being its own artifact. And it returns the same answer when run again six months later.

That third property is the one most companies fail, and it fails silently. A start date gets corrected in the customer relationship tool. A contract value is updated on renewal rather than versioned. Neither is a mistake, and both mean last year’s figure can no longer be reproduced — at which point it has quietly reverted to being a claim.

Where a number lives only in a spreadsheet one person maintains, treat that as a finding rather than a tooling preference. The spreadsheet is usually correct. It is also unreviewable, its logic is undocumented, and it leaves the company when its author does.

Every definition needs a name against it

Ownership is the part almost always left implicit, and it is the cheapest of the three to fix. Each definition and each query needs one named owner: the person who decides what the definition is, and who gets asked when the number looks wrong. Not a function — a person. A metric owned by the finance team is owned by whoever happens to be free that week, which is how the definition moves without anyone deciding that it should.

A workable split at this size: finance owns recognition, margin, and anything that ties back to the accounts. Revenue operations owns pipeline, conversion, and the customer definitions underneath retention. Whoever runs the product owns usage and adoption. One person — commonly a chief financial officer or a head of revenue operations — owns the register itself and holds the authority to refuse a second definition of a word already defined.

Put the owner’s name beside each entry and the date the definition last changed next to that. Those two columns do more work in diligence than any volume of supporting detail, because together they say which numbers are maintained and which have been sitting untouched since somebody built a deck around them.

What to assemble, and when

Start roughly two quarters ahead, and start with the definitions rather than the exports. The exports take a week at any point; the definitions take two quarters because that is how long it takes to discover which two teams disagree and to get the disagreement settled by someone with the standing to settle it.

• One page per metric: the definition, the query, the system it reads, the owner, and the date it last changed.
• Eight quarters of each, produced by re-running the query rather than by collecting the reports that were circulated at the time.
• A short note on every definition that moved inside that window, with the restated history shown alongside the original rather than in place of it.
• The two or three numbers you expect to be challenged, named by you first, each with what you are doing about it and by when.
• One operational change from the last year and the number that moved because of it.

Two quarters is the window because it is long enough for the disagreements to surface and short enough that nobody loses interest. Doing it during the process has a specific cost: the first thing an investor watches is a company discovering its own numbers in public, and that impression is not reversible later with a better spreadsheet.