Most definitions treat ICP as a document: a paragraph describing company size, industry, and pain points, written from intuition about who the product 'should' serve. That approach produces a plausible-sounding profile that frequently doesn't match who actually buys, retains and expands, because it was never checked against real outcome data.
A derived ICP starts from the other direction. Pull your closed-won accounts, especially the ones with strong retention or expansion, and identify what they share: company size range, tech stack, industry, buying trigger, deal velocity. Score prospective accounts against that same fit matrix, not a written narrative, and the profile becomes something you can test and revise as new outcome data arrives.
The reason this matters beyond definitional precision is that a properly derived ICP answers a question a written description can't: how many accounts in your addressable market actually clear the fit bar. That count feeds a real strategic decision: whether your market is enumerable enough to run account-based marketing, or long-tail enough that it needs inbound instead.
Formula
Fit Score = Σ(weight × signal match) across firmographic, technographic, and behavioral signals, scored per account against your closed-won base
Set a minimum fit-score threshold using your closed-won accounts' own score distribution. Accounts scoring below where your best customers cluster don't qualify. Then count how many accounts in your addressable market clear that threshold. A small, nameable set (roughly hundreds) supports account-based targeting. A large or indeterminate set supports inbound and demand generation instead.
Why Ideal Customer Profile matters
A written ICP description can't tell you whether your market is small enough to target by name or too large to list. A scored, derived ICP produces an actual count, which is the input every account-based-vs-inbound resourcing decision actually depends on.
Two companies, two different answers from the same method
Company A scores its closed-won base and finds the qualifying fit threshold produces roughly 400 named accounts across its addressable market. That's small enough to list on a spreadsheet, which supports running account-based marketing against that named list. Company B runs the identical scoring method and finds over 40,000 accounts clear its threshold. That's a truly long-tail market where no sales team could realistically work each account by name, which means inbound and content-led demand generation is the fit, not ABM. Same method, opposite strategic conclusion, because the method produced a real count instead of a description.
Common mistakes
Writing the ICP before checking it against closed-won data
A profile written from intuition about who 'should' buy frequently diverges from who actually buys and retains. Always check a written profile against real account outcomes before treating it as settled.
Confusing ICP with buyer persona
ICP describes which companies to target; buyer persona describes which individuals within those companies to reach and how to message them. Conflating the two produces targeting that's either too broad (persona-level) or messaging that's too generic (ICP-level).
Treating the ICP as static once defined
A fit matrix built from last year's closed-won accounts can drift out of date as the product, market, or ideal buyer changes. Re-score against current outcome data periodically instead of treating an ICP as a one-time exercise.
Where we work on this
The decision this number feeds into