Ideal Customer Profile for B2B: The Decision That Comes First

By Brian Shelton — Founder of GrowPredictably.com

TL;DR: An ideal customer profile is one decision about which companies you are built to serve, and it comes before any demographic detail. It is the foundation that the more granular customer avatars get built on top of. Use AI to gather and test the evidence. The customer choice belongs to the people responsible for building the business.

Key Takeaways

• The profile answers which companies you are built to serve. The avatar answers who inside those companies you are writing to. Collapsing the two produces a document that is too vague to disqualify anyone.
• The decision that comes first is picking one aspirational client you want to become known for, before a single demographic field gets filled in.
• Your own records carry the constraints real buyers stated, and they matter more each year because a growing share of buyer research now happens inside private AI tools you will never see.
• Use AI to refine and validate attributes the business proposes. Leadership owns the choice of which customers the company should pursue.
• A profile that cannot disqualify a prospect is not finished. The test is a short fit filter you apply before a deal enters the pipeline.

The common failure looks like this. A profile describes a company that resembles everybody you have ever sold to, which means they cannot rule anyone out, and a profile that cannot rule anyone out does no work. It sits in a slide deck while the sales team keeps taking meetings with whoever answers.

This covers the decision that comes before the profile, what evidence actually belongs in it, where AI helps and where it quietly does damage, and how to tell whether the finished thing can disqualify a prospect.

Why does the wrong profile cost more than no profile?

Serving the wrong customers is subtractive. They consume delivery capacity, support hours and roadmap attention that the good accounts paid for, and they rarely renew at a price that covers any of it. The accounting on this has been clear for a long time, and the distribution is steeper than most teams assume.

Robert Kaplan documented the pattern in a study of customer profitability. In the case he describes, the most profitable 40 percent of customers generate 130 percent of annual profits and the middle 55 percent roughly break even.

Then comes the part worth sitting with:

the least profitable 5 percent of customers incur losses equal to 30 percent of annual profits

Robert Kaplan, Marvin Bower Professor of Leadership Development Emeritus, Harvard Business School

He put the general pattern at 15 to 20 percent of customers generating 100 percent or more of the profits.

That work dates from 2005 and predates every tool on your stack, which is rather the point. It is an accounting fact about cost to serve, and nothing since has repealed it. A small group of customers pays for the business and a smaller group quietly takes the money back out.

So the reason to name your ideal customer has nothing to do with tidiness. The bottom of that distribution is expensive, and you choose your way into it or out of it every time you accept a deal.

What is an ideal customer profile, and how is it different from a customer avatar?

These get used interchangeably and they sit at different layers. That confusion is what makes a profile unusable, because it produces one document trying to do two jobs and doing neither. Separating them takes one distinction, and everything downstream depends on getting it the right way round.

The profile describes the company. What kind of business, at what stage, with what problem, running what systems, with what forced them to look. It answers which organizations you are built to serve.

The avatar describes a person inside that company. What they want, what frustrates them, what they are afraid of, and the change they are trying to buy. It answers who you are writing to and what will move them.

The order matters, because the profile is the foundation the avatars are built on. One profile usually supports several avatars, since the person who signs, the person who runs the thing daily, and the person who will block it are three different readers inside one company you have already decided to serve.

Build them the other way round and you get a detailed person floating free of any decision about which businesses you want. That reads convincing and disqualifies nobody, which is exactly the failure this whole exercise is supposed to prevent.

Which decision comes before the profile?

Before a single field gets filled in, there is one choice to make, and it is easy to skip straight past: which client do you actually want to become known for? The answer governs every attribute you write down afterward, which is why making it explicitly beats letting it emerge.

Set aside who you can serve and who you have already served. Ask which one, if you had a stack of them, would make the business you want. That is an aspirational choice. It cannot be derived from your customer list, because that list records who found you, and you are deciding who you are for.

Skip it and the profile becomes an average of your existing accounts. Averages are the enemy here. Average the good-fit customers with the ones who drain your team and you produce a description that admits both.

Make the choice out loud and write it as one sentence naming the kind of company and the problem you solve for it. Everything downstream, the attributes, the avatars, the messaging, inherits from that sentence. If you cannot write it in one sentence, the decision has not been made yet.

Here is the difference the sentence makes. Say you sell a reporting product to software companies.

The average	The decision
B2B SaaS companies, 50 to 500 employees, North America	Series A and B software companies whose finance team is still assembling board reporting by hand every month
Growing fast, data-driven culture	Ten to 30 people, one analyst or none, a founder who still owns the numbers personally
Looking to improve reporting	Just missed a board deadline, or got a question they could not answer in the meeting
Uses modern tools	Already runs a warehouse, so the data exists and the assembly is the bottleneck

The left column describes thousands of companies and turns away none of them. The right column describes far fewer and disqualifies most of the market on purpose, which is what makes it usable. Read the right column and you can tell, within a sentence or two of a discovery call, whether you are talking to one.

Notice what is doing the work in the right column. The situation, the shape of the team, and the system the product has to live alongside. Both columns carry a size band, so the size band is doing none of it.

What evidence actually belongs in the profile?

Your own records, before anything else. Sales call notes. Lost deal reasons. Onboarding conversations. Support tickets from your best accounts and from your worst. Those hold the constraints real buyers stated in their own words, and constraints are what let a profile disqualify anyone.

This matters more every year. Forrester’s Buyers’ Journey Survey, reported in January 2026, found that 94 percent of business buyers now use AI in their purchasing process, up from 89 percent the year before, and that 61 percent use private AI tools provided by their organization. That second figure is the one that should change how you work.

A growing share of the research that decides whether you get shortlisted happens inside systems you have no visibility into at all.

So the analytics you own see less of the buying process every year, and your own recorded conversations become a larger share of the evidence you actually have. Mine them before you buy another data source.

Hunt for three things in particular, because they are what genuinely rules companies in or out: the shape and scale of the business, the specific situation that made them look, and the systems the product has to live alongside.

What role should AI play in building the profile?

AI handles evidence gathering, pattern testing and profile validation. Leadership owns customer selection, because that choice defines the business you are trying to build. Inside that split the model earns its keep quickly: it reads a pile of call transcripts faster than you can, surfaces the patterns across them, and exposes the attribute you keep assuming and have never checked.

Ask it to choose instead and watch what happens. Hand a model your customer list, ask who your ideal customer is, and it returns a confident, reasonable, average answer, because averaging is the operation it is performing. It holds no view on which business you are trying to build, and that view is the entire input.

So the aspirational decision stays with you, the evidence stays yours, and the model becomes the fastest way to test both against what your records actually say.

That is the arrangement the framework describes, and it is the one that produces a profile the company believes, because somebody in the company decided something.

How do you know whether a prospect belongs?

A profile that cannot turn a deal away is decoration. The working version ends in a short filter you apply before a prospect enters the pipeline, and it is deliberately blunt, because a filter with nuance in it becomes a filter you can talk your way around.

Does this company match the situation the profile names, meaning the actual problem and circumstance behind the industry label? Can they act, meaning is the person you are talking to able to decide and fund it? And is the change they want the exact change you are built to produce, as opposed to an adjacent one you could probably manage?

Three honest answers. Anything short of three is a conversation, not a deal, and treating it as a deal is how the bottom of Kaplan’s distribution gets populated.

The uncomfortable part is applying it to prospects you want. A filter you only apply to obvious mismatches has become a formality, and the deals that hurt are rarely the obvious mismatches.

How do you check the profile against reality?

Treat the finished profile as a claim, and go test it. Take your last 20 closed deals, sort them into fits and misfits using the filter, then check what actually happened to each: how long they took, what they cost to serve, whether they renewed, whether they expanded.

If your fits and your misfits produced similar outcomes, the profile is not describing anything real and needs redoing rather than defending.

There is reason to expect the check to matter. In a 2022 study in Frontiers in Psychology, Jin, Jiang and Liu analyzed 19,953 firm-year observations of listed Chinese companies between 2010 and 2019 and found customer concentration had a significant negative effect on corporate performance, with managerial ability mitigating that effect through better customer selection.

The sample consists of listed Chinese firms. Its finding is therefore directional for software companies, and benchmarking your own business would require evidence from a comparable software sample.

The directional finding is the useful part: who gets selected shows up in performance, and the skill of selecting is what separates outcomes.

Then re-run the check on a schedule. The profile describes a market, and markets move, so a profile written once and never tested becomes a description of the company you used to sell to.

Where do you start?

Write the one sentence. Name the kind of company you want to become known for and the problem you solve for them, then hold it up against your last 20 deals and see how many of them it would have admitted.

If it admits nearly all of them, it is an average rather than a decision, and the work is to narrow it until it starts turning things away. That narrowing is uncomfortable and it is the whole job. Everything after it, the avatars, the messaging, the campaigns, inherits whatever discipline you managed in that sentence.

Find the constraint capping your pipeline with the Growth Gap Scan

Frequently Asked Questions

What is the difference between an ICP and a buyer persona?

The profile describes the company you are built to serve: its stage, its situation, and the systems it runs. The persona or avatar describes a person inside that company and what will move them. The profile is the foundation and one profile usually supports several personas, because the person who signs, the person who uses it daily and the person who blocks it are different readers in the same account.

How narrow should an ideal customer profile be?

Narrow enough to turn deals away. The working test is to hold it against your last 20 closed deals and count how many it would have admitted. If it admits nearly all of them it is an average of your customer list rather than a decision, and the work is to keep narrowing until it starts disqualifying companies you could have served but should not.

Can AI build my ideal customer profile for me?

It should not. Ask a model who your ideal customer is and it will average your existing accounts into a confident, reasonable answer that decides nothing, because it has no view on which business you are trying to build. AI is genuinely useful one step later, testing attributes you propose against your call transcripts and surfacing the assumption you have never checked.

How often should you update your ideal customer profile?

Re-run the check on a schedule rather than rewriting on instinct. Sort your recent closed deals into fits and misfits using your filter, then compare what actually happened to each: cycle length, cost to serve, renewal, expansion. A profile that has stopped predicting those differences is describing the company you used to sell to.

What evidence should an ideal customer profile be built from?

Your own records first. Sales call notes, lost deal reasons, onboarding conversations and support tickets from your best and worst accounts. Those carry the constraints buyers stated in their own words, and constraints are what let a profile disqualify anyone. This matters more each year as a growing share of buyer research happens inside private AI tools you cannot see.

What makes an ideal customer profile useless?

Being unable to rule anyone out. A profile assembled by averaging every account you have ever won describes a company that resembles all of them and disqualifies none, so it never changes a decision. The fix is upstream: choose one aspirational client you want to become known for, then build the profile from that choice rather than from the customer list.