Build an ICP from your best and worst customers

Here are my customers: {{paste list with size, industry, use case, ACV, tenure, expansion, churn, support load, satisfaction}}.

1. Define "best" explicitly — retention, expansion, margin after support cost, referrals — and rank customers on it. State the tradeoff in whichever definition you choose.
2. Find what the top decile share that the bottom decile doesn't. Test firmographics, but push harder on situational and behavioural traits: what was happening at the company, who bought it, what they had before, what problem triggered the purchase.
3. Produce the ICP as observable, filterable criteria — things a rep can verify before a call, not psychographics.
4. Produce the ANTI-ICP just as specifically, with the cost of serving them.
5. Say how many of my current customers actually match the ICP, and what that implies about the pipeline.
6. Give the two qualifying questions that predict fit best, and the disqualifying signal we currently ignore.
7. Note where my data is too small or biased to conclude — be honest about that.

No "mid-market companies that value innovation".