Ideal Customer Profile: Where Yours Comes From

What a profile describes

A profile names companies, not people. The confusion between the two is the failure the whole idea exists to prevent, and the cost of it is specific: a perfectly qualified contact sitting inside a company that was never going to buy.

Three layers, three owners: the profile filters companies, the persona describes a role inside one, the score ranks an individual record against both.

Three company facts decide the first row and none of them shows on a person record: too small to afford you, locked into a competitor, or years away from having the problem.

None of this is new. Deciding which customers to serve is fundamental to strategy, and so is deciding not to serve others.

That second half is the part a profile is for. If yours does not rule anybody out, you have written a description of your market.

The count runs one to many in the other direction: many buyer personas, one profile.

The unsettled part is where the profile comes from, and the two answers on this page disagree about it completely.

Mixing them produces a specific and expensive result: a well scored contact at a company that was never going to buy. The score was right about the person and nobody had asked about the company.

The software you keep the profile in blurs the same line. HubSpot ships an ideal customer profile tier property with three grades, tier one a great fit and tier three acceptable but low priority.

The company property a CRM ships for this, as its own documentation describes itWhat the field asks, next to what a profile asks

Drawn from the product documentation for the field. The tier two row carries no label because none is recorded.

A filter answers yes or no and a tier answers how much, so the field that stores your profile is already a score. Nothing warns you about that, and it is the same collapse one layer down.

Which method your customer count allows

Two methods produce a profile. Derivation reads your existing accounts for what they share and writes it down as a filter; discovery states a guess, sells against it and corrects it with what closes. The first has an input requirement: enough existing accounts for a shared pattern to be more than a coincidence.

Derivation is the familiar one, and it is usually sold on its benefits. Its inputs go unmentioned.

Now the other method. Lenny Rachitsky interviewed more than twenty-five named founders, from Gong and Figma to Notion and Databricks, and published what they had in common.

Lenny Rachitsky, 29 August 2023: what the founders did

1. 1Most founders got the first version wrong, so a first version is a starting position and not an answer.
2. 2The signal that corrected it came from outbound selling. Leads arriving through investors and friends did not carry it.
3. 3Everyone landed on at least three attributes. That floor was observed across the group; nobody imposed it.

Read those two together and the disagreement is not about definitions. The evidence runs in opposite directions: one derives the profile from customers you already have, the other produces the customers to derive it from.

Mathilde Collin, who runs Front, is quoted in the same piece calling the failure to think about this early one of her biggest mistakes. She names the omission itself as the cost, not a wrong answer.

Fair question: are the two methods in conflict at all, or does one describe the early stage and the other the later one? That reading holds.

What does not hold is presenting the later method as the only method. A company with eight customers is then handed an exercise its own numbers will struggle to fill.

How many customers derivation needs

Derivation is the method with an input requirement, and the requirement is the part that usually goes unmentioned. A profile lands on at least three attributes. Count the segments three attributes create and the requirement becomes arithmetic, holding only on its own two assumptions.

Three attributes, each with two possible values, divide the world into eight kinds of company.

Say you have eight customers and three attributes. Spread evenly that is one account per cell, and real lists never spread evenly, so the useful question is whether yours clusters.

If most of them sit in one cell, you have something to look at and a sample too small to be sure of. If they scatter, what you have describes your customers; it does not filter for new ones.

Which method your numbers allow

1. Count your closed accounts, then run the arithmetic above on your own list and look at how they land. The fewer accounts your largest segment holds, the less there is to derive from and the more the founder method is what you have.
2. Write the guess down anyway, in three attributes, specific enough that any company can be checked against it in a minute.
3. Sell against it deliberately. Outbound data is what carries the signal; leads arriving through investors and friends do not.
4. Watch for four signs: conversion rate up, enthusiasm up, urgency up, and the nod.
5. Rewrite the guess when the evidence contradicts it, and expect to. Most first versions turn out to be wrong.

Once you do have volume, derivation has something to work with, and its advantage is that it reads evidence instead of memory.

No defensible crossover number exists, and this page will not invent one. Check instead whether your largest segment holds enough accounts to be more than a coincidence, using the same arithmetic on your own list.

What three attributes look like

The word attribute sounds like a field on a form, and the real examples are nothing like that. Eleven companies published their first profile in exactly three conditions each, and the conditions are specific enough to check.

Lenny Rachitsky, 29 August 2023: five of the eleven first profiles

1. 1Linear names a company size, a specific pair of tools, and a leadership pattern. Not one of the three is an industry.
2. 2Ramp names a funding stage, a job the buyer wants done, and a monthly spend band.
3. 3Gong is the only row with a price range in it, and Snyk is the only one that names a developer community.

These are not the firmographic filters a data vendor sells. Google Auth is a technographic signal, a founder-driven product company is a judgment, and less waste with more predictability is a job someone is trying to get done.

That mix is the practical difference between the two methods. Derivation returns whatever attributes your data happens to hold, and those categories are firmographic, technographic and behavioural.

Discovery returns whatever separated the deals that closed from the ones that did not, and that is often something no vendor has a column for.

Look at what these rows name: a tool, a spend band, a community. Sector is the easiest attribute to fill in, and not one of the rows above uses it. Two companies in one industry can differ more than two companies across industries.

For instance, turning your finished profile into terms and pages is our B2B SEO service.

A CRM company view with the profile applied as filtersWhat a profile looks like when somebody has to use it

Three conditions become three chips and the list sorts itself. A profile that cannot be turned into chips describes the customers you already have instead of finding the next ones, and the difference shows up the first time somebody other than the author tries to apply it.

Where profiles go wrong

Four failures come up again and again, and not one of them is a mistake about the definition. Each is a step skipped somewhere between having a profile and using it, and each one is cheap to check.

The third is the one with a number attached elsewhere on this site. Once the profile becomes a named list of companies, account-based marketing sets out the condition under which working that list is worth the effort.

The second is subtler. A profile assembled only from the columns a data provider supplies is bounded by that provider, and a founder-driven product company is not a column anyone sells.

Skipping the whole thing is commoner than any of the four. Gary Lilien and Ofer Mintz found 45 percent of B2B startups doing no systematic marketing at all. They read 693 companies from the Equidam platform, with a separate panel of 377 behind the result.

Systematic marketing in their definition is an ongoing process of collecting customer data and using it. A profile you never revise is closer to that 45 percent than to the four habits above.

Having one and using one are also different things. The Content Marketing Institute asked 1,015 B2B marketers in mid-2025 how far their personalisation reaches, and 59 percent called it basic, in one or two channels.

The fourth is the one the founder interviews speak to directly. A profile treated as settled quietly becomes a reason to reject evidence, and a deal that closes outside it gets filed as an exception instead of read as a correction.

A fifth objection comes from outside this set and it disputes the whole exercise. Jenni Romaniuk at the Ehrenberg-Bass Institute calls narrow targeting counter-productive to B2B growth, and recommends targeting all buyers in the category.

Her evidence is 16 US business insurance products, where customer overlap between brands ran from 14 to 24 percent. The reading behind it is the duplication of purchase law: brands share customers in proportion to competitor size, whatever their positioning says.

That does not make a profile useless to you. It narrows what a profile is for. Use it to decide who you serve well and what you build, not to decide who is allowed to see your advertising.

The same three conditions are also the input to B2B keyword research, where they decide which search terms are worth writing for.

Measure the profile against what closes

All four are failures to check the profile against something. A profile is a claim about which companies buy, so the test is whether the companies it names do, and that test needs the profile written down before the quarter.

Tag every opportunity as in profile or out of profile at the moment it is created. For example, if you write your profile after the quarter, you write it knowing which deals closed, and the test stops being a test.

Then compare win rates between the two groups over the same period. If your in profile group does not win more often, separate two explanations before anything else: the profile may be wrong, or your team may not be applying it.

Watch the four qualitative signs as well as the rate. Conversion up, enthusiasm up, urgency up and the nod all arrive before a number does. A win rate needs enough deals behind it before it says anything.

When the profile is stable enough to score against, lead scoring turns the company level filter into a number on individual records. B2B lead generation sets out what reaching those companies costs.

That test has a timing problem under it. 6sense surveyed nearly 4,000 B2B buyers for its 2025 Buyer Experience Report and found they choose from their Day One Shortlist 95 percent of the time.

First contact with a seller now falls around 61 percent of the way through the journey. A profile that decides who reaches that shortlist is doing its work months before your win-rate column moves.

An account can match the profile and still buy nothing this quarter. John Dawes at the Ehrenberg-Bass Institute puts roughly 95 percent of business buyers out of market in any given quarter.

That figure is division, not measurement: a five-year average purchase cycle spread over quarters, and Dawes calls it a heuristic. A quiet in-profile list is a timing question before it is a profile question.

One more comparison is cheap and rarely run: the accounts that closed outside the profile. If several of them share something, that something belongs in the next version, and your profile was too narrow instead of wrong.