01What an ideal customer profile is, and what it is not
An ideal customer profile, usually shortened to ICP, describes the customer who gets value from your product fastest and is able to buy it. It is a filter. Its job is to tell you who to talk to first, which signups to study, which feature requests to weigh, and which deals to walk away from.
It is not a market size slide, and it is not a persona with a stock photo and a favourite coffee order. A persona describes one human. An ICP describes the account that fits: the kind of company or team, the workflow the product changes, and the moment they start looking. Once the ICP is written, personas for the user and the person who approves the spend sit inside it.
For an AI startup the profile carries extra weight. Many AI products can technically help a huge range of people, so "anyone who writes", "any team with documents" or "every developer" feels true. That breadth is exactly what makes the early go to market vague. The GTM strategy depends on a buyer you can picture, and the positioning line only works when you know who is reading it.
02Why AI startups need a sharper ICP than most
Three things make the profile harder, and more useful, for AI products.
The job matters more than the category. Buyers rarely search for a model. They search for a result: a summary they trust, a draft they can send, a bug they can fix. The ICP should name that job in the buyer's own words.
Trust varies a lot by buyer. A solo founder may accept an occasional wrong answer in exchange for speed. A regulated team may need review steps, logs and a clear data policy before anyone is allowed to try the product. Those are different customers even if they want the same output.
Usage has a cost. Each active user can add model or compute spend. A segment that uses the product heavily but never pays can look like traction while it quietly raises the bill. The profile should favour the people who get value and can convert, not only the people who use it most.
03The fields to write down
Keep the ICP to one page. These fields are enough for an early AI startup.
| Field | What to write | Example (fictional) |
|---|---|---|
| Company or team | Size, stage, and the kind of team that owns the workflow | Seed to Series A software companies with a small support team |
| Job | The task the product does, in the buyer's words | "Answer repeat tickets without copying from old replies" |
| Trigger | The event that starts the search | Ticket volume rises after a launch and response times slip |
| Current workaround | What they do today instead | Saved replies, a shared doc, and a general chatbot in another tab |
| Buyer and user | Who uses it and who approves the spend | Support lead uses it, head of operations approves |
| Must haves | What has to be true before they can try it | Works with their helpdesk, keeps a human in the loop |
| Where they pay attention | The places they already learn from | Support communities, a few practitioners on X, helpdesk forums |
| Disqualifiers | Signals that someone is not a fit yet | No helpdesk, or legal review needed before any trial |
The example column is invented to show the format. Replace every cell with what you have actually seen in calls, replies and signups.
04A fill in template
Most teams find it easier to start from one sentence and then fill the fields underneath it.
Our ideal customer is a [company or team] that needs to [job] when [trigger]. Today they [workaround]. The [user] uses the product and the [buyer] approves it, as long as [must haves]. We are not a fit yet when [disqualifier].
Write three versions of the sentence for three candidate segments, then choose one. Keeping the other two on file is fine. Running all three at once is how a small team ends up with three half finished launches. The GTM strategy template has a slot for the chosen sentence so the rest of the plan builds on it.
05How to find the evidence
An ICP written from imagination is a hypothesis. Turn it into a profile by checking it against things you can observe.
- Look at who already activated. Among early users, find the ones who reached a first useful result without help. Write down what they have in common: team, workflow, trigger. Those patterns are stronger than anything a survey will tell you.
- Read the conversations, not just the counts. Calls, support threads and replies on X show the words people use for the job. Copy those words into the job and trigger fields.
- Note who stalled. People who signed up and never activated are as useful as the ones who did. If many of them share a trait, it may belong in the disqualifiers.
- Ask about the trigger directly. "What happened the week you started looking?" usually gets a more honest answer than "what features do you want?"
- Check who can pay. A segment that loves the product but has no budget owner is a community, not yet a customer base.
Small numbers are fine at this stage. A handful of clear conversations beats a large anonymous signup list, as long as you write down what you saw instead of what you hoped to see.
06Testing the profile on X
X is a useful place to pressure test an ICP before you spend on anything else, because the right buyers either recognize themselves in a post or scroll past.
- Write a post that describes the trigger and the workaround in plain words, without naming the product. Replies from people who say "this is us" are a signal the profile is real.
- Watch which accounts engage. If the replies come from builders and the profile says buyers are operations leads, the message reached the wrong room.
- Use the ICP to choose who to follow, reply to and learn from. The profile should make that list obvious.
- Keep a simple note of which replies came from people who match the fields. That note feeds the weekly GTM metrics review.
If nobody recognizes the problem, rewrite the trigger before you rewrite the product.
07Signs the ICP is too broad
A profile is too broad when it does not change any decision. Common signs:
- It lists industries instead of a workflow. "Healthcare, finance and education" describes a slide, not a buyer.
- It has no disqualifiers. Every real profile excludes someone.
- The job field names a capability ("uses AI to analyze documents") instead of a result the buyer wants.
- Two people on the team describe the ideal customer differently after reading it.
- Your launch posts, landing page and onboarding speak to different people.
The fix is the same each time: pick the segment with the clearest evidence and write the profile only for them. You can widen it once that segment converts and retains.
08How the ICP connects to the rest of GTM
The profile is the first block of a go to market plan, and every later block reads from it.
- Positioning says why this buyer should pick you over their current workaround.
- Channel follows the "where they pay attention" field. For many early AI products that is X, plus one or two communities.
- Motion follows the buyer and must haves. A buyer who can try alone suits self serve. A buyer who needs approval suits a sales assisted path, as the GTM motion playbook explains.
- Metrics count activation and conversion inside the ICP, not across every signup.
If you want an outside view on whether your profile, positioning and channel line up, the GTM score checks the basics in a few minutes, and the intake form is where a full plan starts. The sample plan shows how an ICP section reads when it is finished.
09Keep it alive
Put the ICP where the team will see it: the top of the strategy doc, the weekly review, the brief for anyone writing launch posts. Review it when the evidence changes, not when a week feels slow. A good profile gets more specific over time, and the most useful edits are usually new disqualifiers.
Write the profile as one sentence plus a few fields, test it against real conversations, and let it decide what you say no to.
Questions about this playbook
01What is an ideal customer profile for an AI startup?
It is a written description of the company and the person who get value from the product fastest and can pay for it. For an AI product it also names the job the model does, the trigger that starts the search, and the workaround the buyer uses today.
02Is an ICP the same as a buyer persona?
No. The ICP describes the account that fits: the kind of company, team, workflow and trigger. A persona describes one person inside that account, such as the user or the person who signs off. You usually write the ICP first.
03How narrow should an early ICP be?
Narrow enough that you could list real companies or accounts that fit it and recognize them in replies on X. If the profile fits almost any business, it will not change what you build, write or say no to.
04How often should we update the ICP?
Review it when the evidence changes: a different kind of customer keeps activating, a segment stops converting, or the product gains a job it did not do before. Changing it every week means you are guessing, not learning.
05Where does the ICP go in a GTM strategy?
It is the first block. Positioning, channel, motion and metrics all depend on who the buyer is, so a strategy written before the ICP usually has to be rewritten once the profile is clear.