Everyone has an ICP slide. Almost nobody has an ICP model. The difference is that a slide describes your best customer in prose; a model turns that description into a score you can rank thousands of accounts against — and then act on.
The five signals that matter
Not every attribute predicts revenue. After scoring tens of thousands of accounts, five signals do most of the work:
- Revenue & funding — sets budget and realistic deal size.
- Tech stack — reveals fit, integration surface, and displacement angles.
- Authority — are you reaching economic buyers and champions with real power?
- Need — observable pain or initiatives your product directly resolves.
- Intent — live signals: hiring, funding, tech changes, content engagement.
Weight, don't average
A naive model averages every attribute equally. That buries the signals that actually predict a close. Weight each signal by how strongly it correlated with won deals in your own history — for most B2B SaaS, authority and intent deserve heavier weights than raw firmographics.
Set a minimum composite threshold below which an account simply isn't worked. It feels counterintuitive to discard reachable prospects, but restraint is the point: fewer, sharper messages to perfect-fit accounts beat volume to marginal ones every time.
“Precision beats volume. Booking the right forty meetings takes more discipline than blasting ten thousand emails — and produces far more revenue.”
Keep the model alive
An ICP model is a hypothesis, not a monument. Feed closed-won and closed-lost data back into the weights every quarter. When a segment starts converting, raise its priority; when it stalls, cut it. The model should get sharper the longer it runs.
