Every Definition of Ideal Customer Profile Describes a Guess

I spent ten minutes this morning reading page one of Google for "ideal customer profile software".

Not skimming. Reading. Every result, every definition, the AI Overview, the People Also Ask boxes, the lot. I was doing it for a boring reason, checking whether a search term was worth targeting, and I found something I wasn't looking for.

Almost every definition on that page describes a guess.

What is an ideal customer profile?

An ideal customer profile is a description of the type of company most likely to buy from you and stay. That is how the industry defines it. The problem is the word "description", because a description is something you write down, and writing something down does not make it true.

A useful ICP is not a description at all. It is a measurement of which characteristics actually correlated with your closed-won deals, weighted by how much each one mattered.

Almost nobody sells that. Here is the evidence.

What page one actually says

I went through the organic results one by one and wrote down the noun each source used. This is what page one looked like in August 2026.

Qualtrics: a description of a company you believe to be a perfect fit. SuperOffice: a document. New Breed: a hypothetical description. Pipedrive: a detailed description. DealHub: describes a business's ideal customer. Beauhurst: a representation. Aexus: a detailed description. Adobe: a unified view. IBM: a file containing relevant data.

Description. Document. Hypothetical. Representation. View. File.

Not one of them says model. Not one says score. Not one mentions weighting.

That is not a criticism of any of them. Given the tools most people have, it is honest advice. You cannot compute something without the machinery to compute it, so you write down your best judgement instead, and then the judgement becomes the slide.

But look at what that means in practice. The entire industry agrees that an ICP is the single most important input into who your sales team talks to, and the entire industry also agrees that the way you produce one is to think hard and write it down.

Why "best guess" is not good enough

Here is the bit that should bother you.

Nobody guesses one characteristic. They guess a list. Industry, company size, region, tech stack, job title, funding stage. Six or seven attributes, sometimes more.

Then they treat them as equal, because a document has no way of saying that one thing matters four times more than another.

I have sat in the rooms where these get made. Twenty years of them. The VP of Sales says mid-market. Marketing says SaaS. The CEO remembers a competitor winning something in healthcare and now healthcare is on the list. Everyone compromises, someone types it up, and by Friday it is a slide.

What comes out of that room is not wrong exactly. It is usually directionally sensible. But it is a political settlement rather than a finding, and a rep with forty open deals on a Tuesday morning cannot do anything with it.

Ask that rep which of their deals to work first. The document has no answer. It describes a type. It cannot rank the actual companies sitting in the pipeline right now.

Written once, and never again

Here is the other thing every one of those definitions has in common. They are all static.

A description gets written, agreed, and put in a deck. Nobody sets a reminder to check whether it is still true. Meanwhile the market moves, the product changes, and the deals you are actually winning drift away from the ones you wrote down two years ago.

Nobody notices, because there is nothing to notice with. A document cannot disagree with you.

If your profile is computed from your closed-won data instead of written in a room, this stops being something you have to remember. Run it on last year's wins, run it on this year's, and the difference is right there in front of you.

The thing your ICP should be able to do

A real ideal customer profile answers three questions the document cannot.

Which characteristics correlated with winning, rather than which ones you believe correlated with winning. How much each one is worth relative to the others. And how confident you should be in any of it, given how many deals you actually have.

That third one is the one everybody skips. If you have closed forty deals, some of your patterns are real and some are coincidence, and no amount of confident formatting changes that. A profile built from forty deals should say so out loud.

The uncomfortable version of this: your won deals already contain the answer. Industry, size, deal value, cycle length, source, seniority. Every closed-won record is a data point saying this type of company converts for us. That data has been sitting in your CRM for years while people in a meeting room guessed at what it says.

Search for software, get a template

The other thing I noticed on that page is what was missing.

I searched for software. Google returned definitions, glossaries and blank templates. Two of the top-ranking results had "Missing: software" printed underneath them, which is Google's way of admitting it could not find enough relevant pages to fill the slot.

The tools that did appear were mostly template generators. Fill in the industry. Fill in the company size. Fill in the pain points. Which is the same guessing exercise as the meeting room, except now it has a progress bar.

A blank template cannot tell you that headcount matters more than sector in your data. It cannot tell you that you have two winning patterns rather than one. It certainly cannot tell you when there is not enough evidence to say anything at all.

That is not a gap in the content. It is a gap in the category.

What we do differently

Telepath reads your closed-won deals and derives the profile from them. The weights are computed, not chosen. If headcount turns out to matter more than industry in your data, that is what the model says, whether or not it matches the slide.

Every load-bearing number is calculated in code. The AI narrates figures it has been handed. It never works one out for itself.

And when there is not enough data to support a claim, the product says so rather than producing a number anyway. That is the part I am most proud of and the part that is hardest to sell, because "we cannot tell you that" has never won a feature comparison in the history of software.

But it is the only honest answer when it is true. And if a tool will not tell you when it does not know, you have no way of telling the difference between its findings and its guesses.

Which puts you right back where you started. In a room. Writing something down.

Your won deals already know who you should be selling to. It takes three minutes to ask them.

Free ICP report, no signup: telepath.pro