The situation
Leadership at a subscription software company had a growth target and a budget to put behind it. What they did not have was an agreed answer to the question the budget depended on: which channels were producing customers.
Three sources were available and all three disagreed. The ad platforms reported one thing, the CRM reported another, and the sales team, asked directly, described a third picture that matched neither.
What was in the way
The disagreement was not a reporting nuance. The platforms and the CRM were both wrong, in different directions.
- Paid social claimed 215 leads at roughly $41 each
- The CRM showed 24 contacts from that channel, closer to $715 each, and one customer
- The demo request form, the single most important conversion point on the site, had no conversion tracking on it at all
Scaling spend against those numbers would have multiplied a mistake with real money.
What we did
We traced every demo individually, from first touch to closed deal, rather than trusting any platform’s own account of itself.
- 132 demos coded against the sales team’s own notes
- 218 chat threads pulled from the CRM and read for origin signals
- 45 downloads checked against a public registry by agents running in parallel, to confirm who the accounts actually were
Parallel agents made the volume tractable. The coding decisions stayed human, because attribution is a judgment call about ambiguous evidence, and treating it as arithmetic is how the original numbers got wrong.
What came out of it
- Referrals turned out to be the strongest channel, and the CRM could not see them at all
- Three of the four headline numbers leadership had been working from were corrected
- The funnel was fixed so demos could be counted properly going forward
The specific finding worth naming: referred deals closed at 56%, against 31% for everything else. The most productive channel in the business was the one nobody was measuring or investing in.
What this means for a growing business
Attribution gets skipped because it feels like accounting rather than growth work. Then the growth plan is built on the platform’s self-reported numbers, which are the numbers the platform has the most incentive to flatter.
Before a plan multiplies a number, someone has to prove the number. That is usually a few weeks of unglamorous tracing, and it is the cheapest work in the entire plan.
Before a growth plan can multiply a number, someone has to prove the number. Referred deals closed at 56% against 31%, and the CRM had no idea.