Your ICP Is an Elimination Rule

If your ideal customer profile has never cost you a deal, it is not a profile. It is a description of who you were already talking to.

Personas fail in a particular way, and it is not that they are inaccurate. A persona can be entirely true and still be useless, because a true description of a buyer excludes nobody. Head of operations at a mid-market company, worn down by manual processes, wants to look competent in front of her board. All of that can be correct, and none of it tells the team which of this week’s twelve inbound leads to refuse.

The test is not whether the definition is accurate. It is whether it can say no to money.

A description cannot refuse

Christensen’s argument about jobs to be done begins in the same place. Firms study who the customer is, when the thing that predicts a purchase is the circumstance the customer is in. Attributes correlate. Circumstances cause. A persona records the things a buyer carries into every situation, including all the situations in which they will never buy anything.

A constraint is different in kind, not in detail. It is not a richer description. It is a condition that some real accounts fail. “Head of operations” fails nobody. “No trigger event in the last ninety days” fails a great many, including some that would have taken the meeting and some that would have signed.

The same five accounts, put to each definition. A column that answers every row the same way is not filtering anything.Dr. Hafiz Muhammad Ali

Which is where the discomfort lives. The rule only works if it costs something, and it is paid for in deals you could have chased.

Why the boundary leaks

A porous boundary does not announce itself as poor performance. It announces itself as variance.

One segment converts from qualified call to next step several times better than another. Median stage-to-stage time differs by multiples. One cohort renews and expands while another churns inside two quarters. A subset of accounts consumes implementation and support attention far out of proportion to what it pays. Each of these arrives looking like a separate execution problem, and each gets an execution answer: better enablement, tighter follow-up, more onboarding.

They are not five problems. They are one problem reported five ways. The pipeline holds segments that were never going to behave alike, and the average hides it. Dispersion is the signature of a boundary that exists on a slide and not in the intake process.

It stays porous for a reason that has nothing to do with ignorance. Enforcing it means refusing revenue at the moment revenue is scarcest. A founder is being asked to turn down a real deal in order to protect an argument, and the deal is concrete while the argument is not. That trade is made honestly, in good faith, and it is made again every week until there is no boundary left. A venture at this stage is still searching for its business, and a search with no boundary is not a search. It is a survey.

What enforcement looks like

A constraint that lives in a document is a description with better formatting. The only evidence a boundary exists is a record of what it has kept out.

So disqualification has to happen where the time is actually spent, which is before a demo rather than after one. The constraints have to be answerable in a single sentence, because any condition that needs a paragraph to adjudicate will be adjudicated in favour of the deal. And somebody has to be able to point at a name and say that the rule refused this one.

This is also why the ICP question cannot be settled by spending. A budget line can generate more leads at the top, and more leads through a porous boundary produce more dispersion, not more revenue. That is the same failure as buying advertising in place of a decision, arriving one step earlier in the funnel.

The number worth watching is not how many leads were disqualified. It is whether any of them hurt.

THE RULE

Three questions, one for each way a boundary fails.

1. Name the last deal your ICP refused.

   A name, a date, and roughly what it was worth. If nothing comes to mind, the boundary has never been load-bearing, and everything downstream of it has been running without one.

2. Which of your constraints can be answered in one sentence?

   Go through them and mark the ones a salesperson could settle on a first call without asking you. The rest are not constraints. They are preferences that will lose to any deal with momentum behind it.

3. What is the variance telling you that the average is not?

   Split conversion, cycle time, retention and support load by segment rather than reading the blended number. The spread is the diagnosis. A healthy average across two incompatible segments is not a healthy business, it is two businesses being run as one.

An ideal customer profile is not a picture of who you want. It is the list of people you agreed in advance to turn away, and until it has turned one away, you do not have one.