Preventing an unnecessary AI project

The founder wanted to build an AI lead-scoring system before the sales team had a shared definition of what made a qualified lead.

The stated problem was scoring. The real question was what the model would have been trained to agree with.

Asked three members of the sales team to describe a qualified lead and received three materially different answers — different triggers, different signals, different thresholds.

Any model built on that history would have learned the average of three opinions and returned it with the confidence of a number.

Fix the definition first, in the open, with the people who use it. Automation was not ruled out — it was sequenced behind the thing that had to be true first.

Recommended against the build for now, with a clear condition for revisiting it: a qualification standard the team applies consistently for a full quarter.

Facilitated alignment across the sales team and built a weighted lead qualification framework from what they actually agreed on.

Validated it through a one-month pilot before recommending any automation, so the framework had to prove itself against real pipeline rather than argument.

AI does not create clarity. It scales whatever clarity already exists, including the absence of it.

The most valuable recommendation in this engagement was the one that cost nothing to implement.