372 closed won deals, and 13 that could answer the question
The job was to check our ICP document against what really closes, and the first number looked like plenty of data.
372 deals had closed in 90 days. 13 of them were new logos. The other 359 closed on customers we already had, and every one of those describes a choice made earlier, under criteria that may no longer apply.
| Closed won in 90 days | Deals |
|---|---|
| All closed won | 372 |
| Net new logos | 13 |
13 of 372. For validating who to sell to, the other 359 cannot answer the question.
They describe customers the company already had, which is a different question with a different answer.
From six months inside a B2B GTM organization with 36,000 accounts in its CRM.
The starting pool was larger still. 8,438 calls had been recorded in the same 90 days, and narrowing that to something readable was most of the work. My evidence in the end came from six won accounts plus about 34 discovery and demo calls, which is a small sample, and I am putting that before the findings.
Four things in the document did not survive the deals
| What the document said | What was closing |
|---|---|
| A size band well above what the deals show | Deals clustered near the bottom of it and below |
| A criterion marked as a must-have | Missing on won deals, continuously |
| Two segments we had ruled out entirely | Winnable, on the evidence of this sample |
| The primary job was replacing the core system | A layer over the system they already ran |
The size band is the one that shows how a document drifts, and I am deliberately not publishing the numbers on it. That band was written in a trading volume figure, which is one of the quantities we do not communicate externally, so the finding travels and the figures stay inside. What matters for a reader is the shape: the band had been drawn above almost everything that actually closed, so anybody using it to build a target list was filtering out the customers.
One limit on that row is mine to state. The field holding the size figure was unmaintained or empty on new accounts, so the reading came from a second field or from what was said on the call. That is a proxy, and it is the reason the row says the band was too high rather than by how much.
They were buying a layer over the system they already ran
The last row is the expensive one, because the messaging rests on it.
Our document described a company replacing its core system. The people buying described a layer sitting over the system they already had and were not going to remove. Those are different purchases with different risk, different buyers and different objections, and our value proposition was written for the first one, and it gets polite agreement from somebody with the second.
That is also the finding a pipeline report cannot produce. It only exists in what people said on calls, and it only became visible because the sample was narrowed to the deals that could answer the question.
My own correction
One revenue figure in my analysis was out by a factor of 20, because the number belonged to a different company than the one it was attached to.
Verification caught it before anything shipped, and verification is too late for a figure that carries a conclusion. It is why every figure in the final report carried its source line. When the sample is six accounts, one misattributed number moves the answer.
One other limit shaped what the report could claim. The quotes were extracted by a model and not checked by hand one at a time, which is marked in the report rather than smoothed over.
What to do instead
- Filter to net new logos before you count anything. Deals on existing customers close, they look identical in the report, and they describe an earlier ICP.
- Put the sample size in the same sentence as the finding. Six accounts is a small number and a reader is entitled to see it before the conclusion.
- Read the calls, not only the fields. The job people are buying for lives in what they said. No CRM field holds it.
- Check the size band against the deals, not against the plan. A band drawn above what closes turns a target list into an exclusion list.
- Verify any figure that carries a conclusion against its source record, by hand. On a sample this size, one number attached to the wrong company changes the answer.
The same discipline applied to a market sizing model found a coverage claim resting on a field nobody had validated. And once an ICP does hold up, the next question is whether there is anybody to contact at the companies it selects, which is where most of the best scoring companies had nobody on the record.