An ICP Definition That Actually Filters: Rejection as the Test

The slide says: mid-market B2B companies in DACH, 200 to 2000 employees, in manufacturing, software or professional services, with a digital transformation agenda. It has been in the deck two years and nobody disputes it, because nobody could: every account the team has worked passes it, including the twelve that wasted a quarter.

That is not an ideal customer profile, it is a description of the market. A profile that excludes nothing has no opinion, and a definition with no opinion cannot decide anything on your behalf.

A filter is something that rejects

Here is the ten-minute test. Pull last quarter's closed-won and closed-lost, and score each deal against the ICP as written, mechanically. Then look at two numbers: the share of wins the definition accepts, and the share of losses it rejects.

A definition doing real work accepts most of what you won and rejects much of what you lost. If it accepts everything in both columns, it is a description, and the reps work whatever they like, which is what they already do. If it rejects half your wins, it is a fantasy about the customers you wish you had and the team is quietly ignoring it. Both failures look identical on the slide.

Firmographics are the starting point and the weakest part

Industry, headcount, revenue band, geography. They are in the document because they are the fields every list tool sells, which is the problem: your competitors filter on the same four fields and buy the same list.

They are also less reliable than they look. Industry classifications describe what a company was registered as, not what it does now. Headcount figures count contractors, international entities and dormant profiles inconsistently, so a 200-employee threshold moves companies in and out by source. Revenue is modelled rather than observed for privately held firms, which is most of the Mittelstand.

Keep them anyway. They define the addressable set and are cheap to apply at the top of a list build. They just do not say which companies have the problem.

The criteria that separate

Four kinds of signal do the separating, and all are observable from outside the account.

A trigger event: a funding round, a regulatory deadline, a hire into the role that owns your problem, a leadership change in the function you sell to. A trigger does not make a company a fit, but it changes the odds that the problem is painful now.

An existing tool in the stack. A particular CRM, marketing automation platform or ERP implies the next problem, and the tool is checkable through job postings, integration directories, DNS records and vendor customer pages.

An org structure that implies the problem. Six sales reps and no operations person means the routing is manual. A data team with no analytics engineer means the pipelines are ad hoc. Structure is a better proxy for pain than size.

A budget owner who exists, named by title. If no role has a budget line this comes out of, the deal takes two quarters and loses to doing nothing.

Write the negative criteria down

Almost nobody does this, and it is the half that makes the document a filter. Negative criteria disqualify an account regardless of how it scores elsewhere: a procurement model you cannot survive, such as public tender when your contract is annual and self-serve; a stack you cannot integrate with; a works council process that adds six months to deployment; a company mid-merger, where every non-essential project stops.

The test for a good negative criterion is that it has already cost you something. Pull the closed-lost reasons for two quarters and read the free-text field, not the picklist. The patterns appearing three times are your disqualifiers, and they were paid for.

The buying committee belongs in the definition

An ICP that stops at the company is half a definition, because you do not sell to companies. Name the roles that must exist: the person with the problem, the person with the budget, and the person who can block, which in Germany is often IT, the data protection officer or the works council rather than finance.

If the definition does not say who must be present for a deal, the list build returns companies and outbound goes to whichever contact the enrichment tool ranked first. Committee criteria also disqualify faster: an account with a blocking role and no budget role is not an early opportunity, it is a no with a longer sales cycle.

Encode it so systems can execute it

A slide filters nothing. The definition has to exist in four places before it changes behaviour: a scoring rule in the CRM with fields, weights and a threshold; a saved filter in the list build with the negative criteria as hard exclusions; a routing condition separating what reaches a rep from what goes to nurture; and a suppression rule so disqualified accounts stop reappearing in six weeks.

That constraint improves the definition. Every criterion needs a field populated on most records, by enrichment or by an event in your signal ledger. A criterion that requires a person to research each account is not a criterion, it is a wish, and it will be skipped by the second week. Write the definition against the fields you have, then buy enrichment for the one or two criteria worth it.

Review it against won and lost every quarter

Score last quarter's closed deals against the current definition and sort them into four cells: passed and won, passed and lost, failed and won, failed and lost.

Passed and lost is a qualification or product problem. Failed and won is the interesting cell, the one people explain away. One of those is luck. Three in a quarter means the definition is missing a segment that is buying from you, and the honest response is to change the criteria, not call the deals exceptions. The review takes an hour, and it is the only thing keeping an ICP from becoming a description again.

Where a tight ICP goes wrong

Too narrow starves the pipeline. A definition that leaves 60 addressable accounts is not an ICP, it is a target account list, and it needs a different motion: named accounts, multi-threaded, measured over a year not a quarter. If the team cannot build a quarter of coverage from accounts that pass, widen the filter.

The second failure is subtler: an ICP derived from won deals describes who bought, which is partly who you happened to reach, so a single dominant channel gets encoded as a market truth.

And early-stage companies genuinely do not know theirs. With eleven customers, a pattern is noise. The honest artefact then is a written hypothesis with an expiry date and a list of who you said no to and why. The diagnostic asks where you sit on this.

Frequently asked questions

What is an ideal customer profile?

An ideal customer profile is the set of criteria deciding which accounts your go-to-market motion will and will not work, encoded so systems can apply it. It combines firmographic screens such as industry, size and geography with the stronger separating signals: a trigger event, an existing tool in the stack, an org structure implying the problem, and an identifiable budget owner. It also states negative criteria and the roles that must exist on the buying committee. Unlike a persona, it is meant to reject accounts.

How do you test whether your ICP is actually filtering?

Score last quarter's closed-won and closed-lost deals against the definition as written, then check two numbers: the share of wins it accepts and the share of losses it rejects. A working definition accepts most of your wins while rejecting a meaningful share of the losses. If it accepts almost everything in both columns it is a description; if it rejects many of your wins it is a fantasy the team already ignores. The check takes ten minutes with a CRM export.

Should the ICP include the buying committee?

Yes, in the same document rather than as a separate exercise. Name the role that owns the problem, the role that owns the budget, and the roles that can block, which in German organisations is frequently IT, the data protection officer or the works council. Without them the list build returns companies instead of people, and accounts with no budget owner sit in the pipeline for two quarters before losing to no decision.

What should an early-stage company use instead of an ICP?

A written hypothesis with an expiry date. With a handful of customers there is not enough data to separate a pattern from coincidence, and a narrow profile at that stage mostly encodes which accounts the founders happened to reach. Record the criteria you believe in and every account you said no to and why, then revisit both after the next ten deals. That record is what a real definition gets built from.