Choosing an ideal customer profile feels like giving up most of the market, which is why founders put it off and describe their customer as anyone who has the problem. The choice is not about who could buy. It is about who you will win first, because every later decision, what to build, what to charge, what to say and where to say it, is easier and better once it is made for one kind of customer and not five. And a good profile is not a paragraph. It is a short list of measured attributes you can score a real company against.

Who this is for

Founders whose honest answer to who is this for is still a list. It is written for the first choice, before there are customers to learn from, and it holds up for the founder with a handful of customers who cannot yet say what they have in common.

The process, in order

1
Accept that you have to choose
A product for everyone is sold to no one in particular, and the cost shows up everywhere at once: a homepage that speaks to nobody, a roadmap pulled five ways, and interviews that cannot be compared because no two people shared a situation. Narrowing is what makes the rest of the work possible.
The choice is a starting point, not a ceiling. A beachhead market is the first small group you can win completely, chosen because winning it gives you the references and the product to reach the next one. Crossing the chasm is the long version of the same argument: later customers buy on the evidence of people like them, so you need a group small enough that everyone in it can hear about you from someone they know.
2
List the candidates
Write down every group that plausibly has the problem. Most ideas have four to eight, and the list is usually longer than the founder first thinks. Describe each one as a situation, not a label: not small businesses, but ten-person agencies that bill by the hour and reconcile time by hand every Friday.
Keep the unit straight. A customer segment vs persona mix-up starts here: the segment is the group of companies or people who share the situation, the persona is one kind of person inside it. For a product sold to companies you are choosing the segment first. It also helps to notice which way your idea leans, because vertical vs horizontal software changes what a candidate looks like: one industry served deeply, or one job served across many industries.
Check that each candidate is really one group. Three tests help: would they all buy the same product, could you sell to all of them the same way with the same words, and do they talk to each other. If the answer to any of these is no, it is two candidates sharing a label, and they should be split before you compare them. The third test matters most later, because a group that talks to itself is one where the tenth customer has already heard of you from the first.
Do this in Lev
Not sure who the candidates are? Describe your idea and Lev drafts two or three candidate customer groups, each with how badly it feels the problem, so you start from a list instead of a blank page.
List your candidate customers with Lev
3
Compare them on the same questions
Put the candidates side by side and ask the same five questions of each. How much does the problem hurt, which is the painkiller vs vitamin test applied to a group. Can they pay, and do they already spend on it, which is the first read on willingness to pay. Can you reach them, by name, this month, through channels you can actually use. How fast do they decide, since a long sales cycle length can outlast an early company's runway. And are there enough of them to matter, which a rough bottom-up market sizing answers better than a market report.
Rate each high, medium or low, and where you do not know, write unknown. The unknowns are the point: they are the list of things to find out before you commit, and a table with honest gaps is more useful than one filled in from optimism.
Then look for the candidate that is strong on pain and reach, even if it is not the largest. Size is the question founders weight most and the one that matters least for a first customer group.
- How badly does the problem hurt them, and what have they already done about it?
- Do they have a budget, and is anything being spent on this today?
- Can you name twenty of them and reach them this month?
- How long do they take to decide, and how many people are involved?
- Are there enough of them to build the next stage of the company on?
4
Write it in three parts, with numbers
An ideal customer profile is only useful if it can be applied. Write it in things you can observe from outside: the kind of company, its size, what it sells, the tools it uses, and the event that makes the problem urgent, such as a new hire, a funding round, a regulation, or a missed deadline. Words like innovative or growth-minded sort nothing.
A complete profile has three parts. The first is attributes you can measure: who they are, with a number or a threshold wherever one exists, and a note of which direction is better. The second is propensity to pay: what they already pay for, how they like to pay for it, and which roles in the group hold budget. The third is channels to market: where this group finds out about new things, the people, publications, communities and events they actually follow. Founders write the first part, skip the second, and forget the third, which is the one that tells you how to reach anybody on the list.
Here is the shape, for a contact management product sold to professionals. Attributes: places high value on relationships, eight or more out of ten, because the job depends on them; between five hundred and ten thousand contacts across address book and social networks; more than five meetings a week with people outside the company; uses technology to get an information advantage. Propensity to pay: already pays monthly for information tools, such as a CRM, a contact data service or a research subscription, and more is better; works in a role like sales, recruiting, public relations or law, or influences a small team that does. Channels: also uses personal productivity tools; reads productivity newsletters and listens to productivity podcasts; goes to networking events to build their network. Every line can be checked against a real person.
Two kinds of attribute do the work inside the first part. Demographic attributes are the facts: for a company, its industry, size, location, stage and tools; for a person, their role, seniority, age or income. They are easy to search on and they get you a list. Psychographic attributes are how the customer thinks: what they value, what they worry about, how they like to buy, whether they try new tools early or wait for proof. They are harder to search on and often the better predictor, since two companies with identical facts can want opposite things. Use both, and for each psychographic attribute write down the outside sign that gives it away: growth-minded sorts nothing, but has posted three sales roles this quarter does.
A good definition reads like a good database query. Each attribute is another condition: industry, headcount, the tool they use, the event that just happened. Every condition you add returns fewer companies and better ones. Too few conditions and the list is long and mostly wrong; too many and it is empty. The right number returns a list you can work through without running out and without drowning: enough people for thirty or so conversations, few enough that most of them turn out to have the problem. Fewer than about twenty and you will run out before you learn anything; tens of thousands and the list is too broad to learn from. If you cannot run the definition against a real list of companies and get names back, it is still a description and not yet a profile.
Three attributes is a sensible floor. Founders who later built large business software companies tend to describe their first profile with at least three specific conditions, and some started with five or six, narrow to a degree that sounds almost absurd when said aloud: a size band, a single region, one way of paying staff. They widened later, from a position of having won. Two habits keep a first list comparable: one or two company sizes, not four, and one country or region, so every call lands inside your working day. Starting narrow and widening is routine. Starting wide and narrowing usually happens only after the money has been spent.
Said aloud, a good definition has a recognizable shape: my ICP is X, even better if A, B and C, and not D. X is the must-have, the condition every target meets. A, B and C are the things that make a target better without being required, and they are how you rank the list once you have it. D is the look-alike you have learned to leave alone. For example: my ICP is agencies of ten to thirty people that bill by the hour; even better if they have just hired an operations lead, run more than one office, and still track time in spreadsheets; not agencies that work on fixed-fee retainers. If you cannot say yours in that one sentence, it is not finished.
Write the definition at two levels: the company, and the economic buyer inside it. The company level says which organizations to look at. The buyer level says whose name goes on the list: the person who owns the budget this would come out of, described the same way, by title, team, how long they have been in the role, and what they are measured on. The same sentence works for both. My buyer is the head of operations; even better if they joined in the last year and report to the founder; not an office manager without budget. A profile that stops at the company gives you a building to call, not a person.
Then note who else is inside it. The user vs buyer vs champion distinction belongs in the profile, because a segment where one person feels the problem and another holds the budget is a different sale from one where they are the same person.
Finish with who is out, the not D in that sentence. A short list of disqualifiers, the companies that look right and are not, saves more time than any other line in the document.
Do this in Lev
Lev drafts your profile from your customer discovery work, then walks you through it a question at a time: the buyer, their seniority, company size, location and industry. As you answer, it counts how many real people match and flags the choices that blur the list, such as any company size, or a buyer who rarely feels the problem at that size.
Build your ICP with Lev
5
Fit the profile to how you sell
The shape of the profile depends on the kind of business. Four cases cover most founders.
Selling to large companies, there is an economic buyer and there are influencers around them: the people who will use it, the ones who can block it in security, legal or procurement, and the one inside who wants it to happen. The user vs buyer vs champion distinction is the short version. The profile names the company, then the economic buyer, then the two or three other roles that have to be on side, because a deal with the right buyer and an unconvinced influencer stalls just as surely as one with no buyer.
Selling to consumers, you still have an ICP. There is no company to describe, so the definition moves to the person and the moment: who they are, what they already do or pay for, and what just happened in their life that makes the problem live. The same sentence applies, a must-have, a few even-betters and a not. Everyone who shops online is no more a profile than all small businesses is.
Running a marketplace, you have two ICPs, one for each side, and both need the same care. In a two-sided marketplace model the side that pays is the one to get right first. Define the paying side as tightly as any other profile, win it, and let the definition of the other side follow from who those payers most want to find.
Where the person who uses the product is not the person who pays, you have buyers and users, and you will likely need to talk to both. An advertising-supported model is the clearest case: the audience uses it and the advertiser pays. Some products sold to companies work the same way, used by a team and bought by someone who never opens it. Write a profile for each and keep them apart. Users tell you whether the product is worth using; buyers tell you whether it is worth paying for. The mistake is to hear enthusiasm from one and count it as demand from the other.
6
Treat the choice as a hypothesis and go and test it
At this point the profile is a guess with good reasons behind it. The test is conversations with people who match it, and the first thing each conversation checks is the profile itself: is this person who I thought my customer was, and is the problem in their top three.
Work the candidates in serial, not in parallel. Take the top one from the table and give it your full attention for a fixed run of about ten conversations before you look at the next. Three profiles tested at once means three sets of outreach, three versions of the pitch and four conversations each, which is too few to tell a pattern from luck in any of them. One at a time is slower on paper and faster in practice, because each run ends in an answer. Ten is enough to rule a candidate out. Once one survives, the full run is thirty to fifty conversations, enough to hear the same story in the same words from people who have never met.
Alongside the conversations, test demand, because it costs little and reaches people you would never meet. Write two or three messages provocative enough that the right person would share them, each naming the problem in the profile's own terms. Put each on a simple page with a short sign-up form built from the profile's attributes, and ask for some effort in return for early access: fill in the form and pass the page on. Who responds, which message they responded to, and how they answered the form is a second, larger read on the same hypothesis, and the best respondents become your next conversations.
You will know a profile is closer to right by how the conversations change. More of the people you approach agree to talk. They get more animated when the problem comes up. They want to do something about it now, not next quarter. And they nod along as you describe it, sometimes finishing the description for you. Judge this from people you reached cold, not from introductions through friends and investors, who are kind in ways that tell you little.
Watch for the early adopter inside the group, the ones who have already tried to solve it themselves. The fullest version of the test has five parts: they have the problem, they know they have it, they have been looking for a solution, they have patched one together themselves, and they have or can get the money to pay for a better one. If you cannot find any after ten conversations, the pain score you gave this candidate was wrong. Close it out, write down why, and go back to the table for the next one. That is problem validation doing its job, not a failure of the process.
Do this in Lev
Once your profile is saved, Lev finds twenty real people who match it, looks up how to reach each one, and drafts the outreach, so your first ten conversations start from a list of names rather than a research project.
Find people who match your ICP with Lev
7
Score every lead against it
Once the profile is written in measured attributes, it becomes a scoring system. Every person or company you meet gets a row, and every attribute gets a column: does it meet the must-have, how many of the even-betters, any disqualifier. Keep it somewhere you can change easily, because you will change the columns every week.
The test of a lead score is whether it predicts. Do the highest-scored leads engage first, close more often, stay longer and expand? Compare the score with what actually happened, and you will find attributes that matter more than you thought and some that do not matter at all. Cohort retention is the slowest but most honest of those checks, because it shows which kinds of customer are still there months later.
Re-sort the list every week and adjust the weights. The ranking moving is not noise; it is the profile learning. The early adopter you are looking for is usually near the top of a score that has been corrected a few times.
8
Choose the first customers from inside it
A profile describes a group; you still have to choose the first few companies. Pick an ideal first customer on purpose: the one whose situation is most typical of the group, not the friendliest and not the largest. An unusual first customer will pull the product toward needs nobody else shares. Three to five who pay is the target: enough to see what they have in common, few enough to serve properly.
Decide what you will sell them first. The wedge is the narrow thing that gets you in the door of this kind of customer, small enough to buy quickly and useful enough to earn the next conversation. A clear profile is what makes a wedge possible, because you know exactly whose door it is.
9
Revisit it on a rhythm, and when the evidence says so
Once there are customers, the profile stops being a guess and becomes something you can check. Win-loss analysis shows which kinds of company buy and which stall. Cohort retention shows which ones stay. Put together, they often reveal that the customers getting the most value are a narrower group than you set out to serve, or a different one.
Put the checking on a calendar: a little each week, a deeper look each month and each quarter. How Customer Discovery Actually Works lays out that rhythm. The part that belongs here is the question each pass asks: does the profile still describe who actually buys and stays, and does the lead score still predict it?
There is a direct way to find the group that matters most. Ask the people already using the product how they would feel if they could no longer use it, and look closely at the ones who say very disappointed. Who are they, what do they have in common, and what do they say is the main benefit. That description, of your most demanding and most satisfied customer, is often a sharper profile than the one you started with, and it comes from behavior, not from a planning session.
Change the profile when that pattern is clear, and not before. Three good customers of a new kind is a reason to look; it is not a reason to rewrite the homepage. And widen it only after the first group is won, because the next segment will ask who else like them is already using it.
Do this in Lev
When the evidence says the profile should change, change one answer and Lev refreshes the people it found, leaving everyone you have already contacted where they are.
Update your ICP in Lev

Where founders get the process wrong

Describing the customer as anyone with the problem. That is a description of the market, and it gives you nobody to call tomorrow.
Choosing the biggest segment. The largest group is usually the hardest to reach and the slowest to decide, and a first customer group has to be winnable before it is large.
Choosing the segment you happen to know people in, without checking that it has the pain. Easy access is a real advantage and a poor reason on its own.
Writing the profile in adjectives. If two people on the team would sort the same company differently, it is not a profile yet.
Writing attributes without numbers. Uses a lot of contacts sorts nothing; more than five hundred contacts sorts every person into yes or no.
Skipping channels to market. A perfect profile of people you have no way to reach produces a beautiful list and an empty calendar.
Leaving out who is disqualified, so the pipeline fills with companies that look right and never buy.
Keeping two or three profiles open to avoid deciding. Each one halves the attention the others get, and none of them is tested properly. Work them one after another.
Stopping at two attributes. Software companies with fifty to two hundred people returns thousands of names and tells you nothing about which ones to call.
Scoring leads once and never re-sorting. A score that is never checked against who actually closed and stayed is a guess with decimals.
Changing the profile after every conversation. One enthusiastic outlier is not a segment.
Confusing users with buyers. Where one group uses the product and another pays, love from users is not a budget, and a buyer's approval is not adoption.
Never changing it. A profile written before the first customer and untouched after the twentieth is describing a guess, not the business.

Related guides

The process, in order

Accept that you have to choose

A product for everyone is sold to no one in particular, and the cost shows up everywhere at once: a homepage that speaks to nobody, a roadmap pulled five ways, and interviews that cannot be compared because no two people shared a situation. Narrowing is what makes the rest of the work possible.

The choice is a starting point, not a ceiling. A beachhead market is the first small group you can win completely, chosen because winning it gives you the references and the product to reach the next one. Crossing the chasm is the long version of the same argument: later customers buy on the evidence of people like them, so you need a group small enough that everyone in it can hear about you from someone they know.

List the candidates

Write down every group that plausibly has the problem. Most ideas have four to eight, and the list is usually longer than the founder first thinks. Describe each one as a situation, not a label: not small businesses, but ten-person agencies that bill by the hour and reconcile time by hand every Friday.

Keep the unit straight. A customer segment vs persona mix-up starts here: the segment is the group of companies or people who share the situation, the persona is one kind of person inside it. For a product sold to companies you are choosing the segment first. It also helps to notice which way your idea leans, because vertical vs horizontal software changes what a candidate looks like: one industry served deeply, or one job served across many industries.

Check that each candidate is really one group. Three tests help: would they all buy the same product, could you sell to all of them the same way with the same words, and do they talk to each other. If the answer to any of these is no, it is two candidates sharing a label, and they should be split before you compare them. The third test matters most later, because a group that talks to itself is one where the tenth customer has already heard of you from the first.

Do this in Lev

Not sure who the candidates are? Describe your idea and Lev drafts two or three candidate customer groups, each with how badly it feels the problem, so you start from a list instead of a blank page.

List your candidate customers with Lev

Compare them on the same questions

Put the candidates side by side and ask the same five questions of each. How much does the problem hurt, which is the painkiller vs vitamin test applied to a group. Can they pay, and do they already spend on it, which is the first read on willingness to pay. Can you reach them, by name, this month, through channels you can actually use. How fast do they decide, since a long sales cycle length can outlast an early company's runway. And are there enough of them to matter, which a rough bottom-up market sizing answers better than a market report.

Rate each high, medium or low, and where you do not know, write unknown. The unknowns are the point: they are the list of things to find out before you commit, and a table with honest gaps is more useful than one filled in from optimism.

Then look for the candidate that is strong on pain and reach, even if it is not the largest. Size is the question founders weight most and the one that matters least for a first customer group.

How badly does the problem hurt them, and what have they already done about it?
Do they have a budget, and is anything being spent on this today?
Can you name twenty of them and reach them this month?
How long do they take to decide, and how many people are involved?
Are there enough of them to build the next stage of the company on?

Write it in three parts, with numbers

An ideal customer profile is only useful if it can be applied. Write it in things you can observe from outside: the kind of company, its size, what it sells, the tools it uses, and the event that makes the problem urgent, such as a new hire, a funding round, a regulation, or a missed deadline. Words like innovative or growth-minded sort nothing.

A complete profile has three parts. The first is attributes you can measure: who they are, with a number or a threshold wherever one exists, and a note of which direction is better. The second is propensity to pay: what they already pay for, how they like to pay for it, and which roles in the group hold budget. The third is channels to market: where this group finds out about new things, the people, publications, communities and events they actually follow. Founders write the first part, skip the second, and forget the third, which is the one that tells you how to reach anybody on the list.

Here is the shape, for a contact management product sold to professionals. Attributes: places high value on relationships, eight or more out of ten, because the job depends on them; between five hundred and ten thousand contacts across address book and social networks; more than five meetings a week with people outside the company; uses technology to get an information advantage. Propensity to pay: already pays monthly for information tools, such as a CRM, a contact data service or a research subscription, and more is better; works in a role like sales, recruiting, public relations or law, or influences a small team that does. Channels: also uses personal productivity tools; reads productivity newsletters and listens to productivity podcasts; goes to networking events to build their network. Every line can be checked against a real person.

Two kinds of attribute do the work inside the first part. Demographic attributes are the facts: for a company, its industry, size, location, stage and tools; for a person, their role, seniority, age or income. They are easy to search on and they get you a list. Psychographic attributes are how the customer thinks: what they value, what they worry about, how they like to buy, whether they try new tools early or wait for proof. They are harder to search on and often the better predictor, since two companies with identical facts can want opposite things. Use both, and for each psychographic attribute write down the outside sign that gives it away: growth-minded sorts nothing, but has posted three sales roles this quarter does.

A good definition reads like a good database query. Each attribute is another condition: industry, headcount, the tool they use, the event that just happened. Every condition you add returns fewer companies and better ones. Too few conditions and the list is long and mostly wrong; too many and it is empty. The right number returns a list you can work through without running out and without drowning: enough people for thirty or so conversations, few enough that most of them turn out to have the problem. Fewer than about twenty and you will run out before you learn anything; tens of thousands and the list is too broad to learn from. If you cannot run the definition against a real list of companies and get names back, it is still a description and not yet a profile.

Three attributes is a sensible floor. Founders who later built large business software companies tend to describe their first profile with at least three specific conditions, and some started with five or six, narrow to a degree that sounds almost absurd when said aloud: a size band, a single region, one way of paying staff. They widened later, from a position of having won. Two habits keep a first list comparable: one or two company sizes, not four, and one country or region, so every call lands inside your working day. Starting narrow and widening is routine. Starting wide and narrowing usually happens only after the money has been spent.

Said aloud, a good definition has a recognizable shape: my ICP is X, even better if A, B and C, and not D. X is the must-have, the condition every target meets. A, B and C are the things that make a target better without being required, and they are how you rank the list once you have it. D is the look-alike you have learned to leave alone. For example: my ICP is agencies of ten to thirty people that bill by the hour; even better if they have just hired an operations lead, run more than one office, and still track time in spreadsheets; not agencies that work on fixed-fee retainers. If you cannot say yours in that one sentence, it is not finished.

Write the definition at two levels: the company, and the economic buyer inside it. The company level says which organizations to look at. The buyer level says whose name goes on the list: the person who owns the budget this would come out of, described the same way, by title, team, how long they have been in the role, and what they are measured on. The same sentence works for both. My buyer is the head of operations; even better if they joined in the last year and report to the founder; not an office manager without budget. A profile that stops at the company gives you a building to call, not a person.

Then note who else is inside it. The user vs buyer vs champion distinction belongs in the profile, because a segment where one person feels the problem and another holds the budget is a different sale from one where they are the same person.

Finish with who is out, the not D in that sentence. A short list of disqualifiers, the companies that look right and are not, saves more time than any other line in the document.

Do this in Lev

Lev drafts your profile from your customer discovery work, then walks you through it a question at a time: the buyer, their seniority, company size, location and industry. As you answer, it counts how many real people match and flags the choices that blur the list, such as any company size, or a buyer who rarely feels the problem at that size.

Build your ICP with Lev

Fit the profile to how you sell

The shape of the profile depends on the kind of business. Four cases cover most founders.

Selling to large companies, there is an economic buyer and there are influencers around them: the people who will use it, the ones who can block it in security, legal or procurement, and the one inside who wants it to happen. The user vs buyer vs champion distinction is the short version. The profile names the company, then the economic buyer, then the two or three other roles that have to be on side, because a deal with the right buyer and an unconvinced influencer stalls just as surely as one with no buyer.

Selling to consumers, you still have an ICP. There is no company to describe, so the definition moves to the person and the moment: who they are, what they already do or pay for, and what just happened in their life that makes the problem live. The same sentence applies, a must-have, a few even-betters and a not. Everyone who shops online is no more a profile than all small businesses is.

Running a marketplace, you have two ICPs, one for each side, and both need the same care. In a two-sided marketplace model the side that pays is the one to get right first. Define the paying side as tightly as any other profile, win it, and let the definition of the other side follow from who those payers most want to find.

Where the person who uses the product is not the person who pays, you have buyers and users, and you will likely need to talk to both. An advertising-supported model is the clearest case: the audience uses it and the advertiser pays. Some products sold to companies work the same way, used by a team and bought by someone who never opens it. Write a profile for each and keep them apart. Users tell you whether the product is worth using; buyers tell you whether it is worth paying for. The mistake is to hear enthusiasm from one and count it as demand from the other.

Treat the choice as a hypothesis and go and test it

At this point the profile is a guess with good reasons behind it. The test is conversations with people who match it, and the first thing each conversation checks is the profile itself: is this person who I thought my customer was, and is the problem in their top three.

Work the candidates in serial, not in parallel. Take the top one from the table and give it your full attention for a fixed run of about ten conversations before you look at the next. Three profiles tested at once means three sets of outreach, three versions of the pitch and four conversations each, which is too few to tell a pattern from luck in any of them. One at a time is slower on paper and faster in practice, because each run ends in an answer. Ten is enough to rule a candidate out. Once one survives, the full run is thirty to fifty conversations, enough to hear the same story in the same words from people who have never met.

Alongside the conversations, test demand, because it costs little and reaches people you would never meet. Write two or three messages provocative enough that the right person would share them, each naming the problem in the profile's own terms. Put each on a simple page with a short sign-up form built from the profile's attributes, and ask for some effort in return for early access: fill in the form and pass the page on. Who responds, which message they responded to, and how they answered the form is a second, larger read on the same hypothesis, and the best respondents become your next conversations.

You will know a profile is closer to right by how the conversations change. More of the people you approach agree to talk. They get more animated when the problem comes up. They want to do something about it now, not next quarter. And they nod along as you describe it, sometimes finishing the description for you. Judge this from people you reached cold, not from introductions through friends and investors, who are kind in ways that tell you little.

Watch for the early adopter inside the group, the ones who have already tried to solve it themselves. The fullest version of the test has five parts: they have the problem, they know they have it, they have been looking for a solution, they have patched one together themselves, and they have or can get the money to pay for a better one. If you cannot find any after ten conversations, the pain score you gave this candidate was wrong. Close it out, write down why, and go back to the table for the next one. That is problem validation doing its job, not a failure of the process.

Do this in Lev

Once your profile is saved, Lev finds twenty real people who match it, looks up how to reach each one, and drafts the outreach, so your first ten conversations start from a list of names rather than a research project.

Find people who match your ICP with Lev

Score every lead against it

Once the profile is written in measured attributes, it becomes a scoring system. Every person or company you meet gets a row, and every attribute gets a column: does it meet the must-have, how many of the even-betters, any disqualifier. Keep it somewhere you can change easily, because you will change the columns every week.

The test of a lead score is whether it predicts. Do the highest-scored leads engage first, close more often, stay longer and expand? Compare the score with what actually happened, and you will find attributes that matter more than you thought and some that do not matter at all. Cohort retention is the slowest but most honest of those checks, because it shows which kinds of customer are still there months later.

Re-sort the list every week and adjust the weights. The ranking moving is not noise; it is the profile learning. The early adopter you are looking for is usually near the top of a score that has been corrected a few times.

Choose the first customers from inside it

A profile describes a group; you still have to choose the first few companies. Pick an ideal first customer on purpose: the one whose situation is most typical of the group, not the friendliest and not the largest. An unusual first customer will pull the product toward needs nobody else shares. Three to five who pay is the target: enough to see what they have in common, few enough to serve properly.

Decide what you will sell them first. The wedge is the narrow thing that gets you in the door of this kind of customer, small enough to buy quickly and useful enough to earn the next conversation. A clear profile is what makes a wedge possible, because you know exactly whose door it is.

Revisit it on a rhythm, and when the evidence says so

Once there are customers, the profile stops being a guess and becomes something you can check. Win-loss analysis shows which kinds of company buy and which stall. Cohort retention shows which ones stay. Put together, they often reveal that the customers getting the most value are a narrower group than you set out to serve, or a different one.

Put the checking on a calendar: a little each week, a deeper look each month and each quarter. How Customer Discovery Actually Works lays out that rhythm. The part that belongs here is the question each pass asks: does the profile still describe who actually buys and stays, and does the lead score still predict it?

There is a direct way to find the group that matters most. Ask the people already using the product how they would feel if they could no longer use it, and look closely at the ones who say very disappointed. Who are they, what do they have in common, and what do they say is the main benefit. That description, of your most demanding and most satisfied customer, is often a sharper profile than the one you started with, and it comes from behavior, not from a planning session.

Change the profile when that pattern is clear, and not before. Three good customers of a new kind is a reason to look; it is not a reason to rewrite the homepage. And widen it only after the first group is won, because the next segment will ask who else like them is already using it.

Do this in Lev

When the evidence says the profile should change, change one answer and Lev refreshes the people it found, leaving everyone you have already contacted where they are.

Update your ICP in Lev

Where founders get the process wrong

Describing the customer as anyone with the problem. That is a description of the market, and it gives you nobody to call tomorrow.
Choosing the biggest segment. The largest group is usually the hardest to reach and the slowest to decide, and a first customer group has to be winnable before it is large.
Choosing the segment you happen to know people in, without checking that it has the pain. Easy access is a real advantage and a poor reason on its own.
Writing the profile in adjectives. If two people on the team would sort the same company differently, it is not a profile yet.
Writing attributes without numbers. Uses a lot of contacts sorts nothing; more than five hundred contacts sorts every person into yes or no.
Skipping channels to market. A perfect profile of people you have no way to reach produces a beautiful list and an empty calendar.
Leaving out who is disqualified, so the pipeline fills with companies that look right and never buy.
Keeping two or three profiles open to avoid deciding. Each one halves the attention the others get, and none of them is tested properly. Work them one after another.
Stopping at two attributes. Software companies with fifty to two hundred people returns thousands of names and tells you nothing about which ones to call.
Scoring leads once and never re-sorting. A score that is never checked against who actually closed and stayed is a guess with decimals.
Changing the profile after every conversation. One enthusiastic outlier is not a segment.
Confusing users with buyers. Where one group uses the product and another pays, love from users is not a budget, and a buyer's approval is not adoption.
Never changing it. A profile written before the first customer and untouched after the twentieth is describing a guess, not the business.

How Picking an ICP Actually Works

The process, in order

Accept that you have to choose

List the candidates

Compare them on the same questions

Write it in three parts, with numbers

Fit the profile to how you sell

Treat the choice as a hypothesis and go and test it

Score every lead against it

Choose the first customers from inside it

Revisit it on a rhythm, and when the evidence says so

Where founders get the process wrong

Further reading

Related guides

Not seeing what you need?

Stop looking these up one at a time

How Picking an ICP Actually Works

The process, in order

Accept that you have to choose

List the candidates

Compare them on the same questions

Write it in three parts, with numbers

Fit the profile to how you sell

Treat the choice as a hypothesis and go and test it

Score every lead against it

Choose the first customers from inside it

Revisit it on a rhythm, and when the evidence says so

Where founders get the process wrong

Further reading

Related guides

Not seeing what you need?

Stop looking these up one at a time