How to review your ideal customer profile when growth has stalled and rebuild it into something you can act on Monday morning.
A guide from Inside Startups. Fifteen minutes to read. An afternoon to run against your own data with your own AI. By the end you will have two or three customer profiles built on evidence, each one telling you who to call this week, what to open with, and which offer to lead with. Take it, produce your own answers, then bring them to me.
01
First customers can make an early profile feel settled. Then growth flattens and the work moves to more outreach, more spend or a new channel. The profile that won the first revenue may no longer reach the next market.
Most ICPs describe a company by sector, size or job title. They do not tell you who to call this week, what to say in the first line or which offer to lead with. They are written in your language, organised around permanent attributes and disconnected from a purchase.
A useful profile starts with evidence: what buyers purchased, the trigger that put them in market, the words they used, the objection they raised and the reason another lookalike should be ruled out. This guide shows how to rebuild one that can be acted on.
Early customers are often unusually tolerant, unusually close to the problem or already known to the founder. Their purchase may point to a pattern, but it is not proof that every company sharing their sector or job title is ready to buy. A useful review separates the coincidences of being early from the reasons the next customer will act.
Signals matter more than permanent attributes. A new hire, a change of system, a missed handover, a growing queue or a paid workaround can tell you that a problem has become immediate. The same company without that trigger may be a poor conversation today. The profile needs to distinguish the two before outreach decides that they are the same.
It also has to connect to the offer. If three things can be sold, the profile should name which one this buyer starts with, what they are buying instead of, and what would make the purchase expand. That is the difference between a document that describes a market and one that gives the team a repeatable next move.
The journey shows where you are and where a re-checked profile takes you. The shaded gap is what the review unlocks.
“Ben has a rare ability to listen deeply and respond with clarity that cuts through complexity. His guidance helped me see challenges from new angles and make decisions with more confidence and precision.”
Juanita C
Founder, Jool
03
Rebuild the profile around the purchase, not the company. Five fields per profile.
1. What they buy. Name the offer this buyer comes in through, and the one they expand into. If you have three offers and the profile does not say which comes first, it is a description rather than a decision.
2. The trigger. What happened in the fortnight before this person would take your call. Triggers are almost always numerical or dated rather than emotional: a funding round closing or failing, a compliance deadline, a metric that turned out to be wrong, a departure, a launch, a contract renewal. This field is what turns a target list into a queue.
3. The pain, in their words. Quoted from a recorded conversation, not paraphrased. This is the material that goes into headlines and first messages, and it only works if it is verbatim.
4. The objection, and the belief underneath it. What they push back on, and then, one level down, what they would have to believe for that objection to make sense. The objection is what they say. The belief is what your content has to change.
5. Disqualifiers. Who looks identical on paper and is not, written so it can be checked before the call, rather than discovered during it.
Underpinning all five, one convention: tag every line by its source. Customer stated, verifiable record, your own assertion, or inference. Only customer-stated material goes into public copy. Without this, a document written today becomes indistinguishable from guesswork in six months.
Produce two or three profiles, not five or seven. Depth beats coverage. Two profiles built on real language will out-produce seven built on plausible reasoning, because the seven give you no basis for choosing where to spend the week.
“Ben has provided me with valuable insights around fundraising, pitch deck development, and the structuring of cap tables and SAFEs. He has also been a great sounding board for strategic planning and go-to-market.”
Michail A
Founder, Roam Network
04
This prompt is only as good as what you feed it. Two things decide whether you get a document you can act on or a generic one you can’t.
The process: an afternoon of setup and analysis, then a working session.
You need an AI that can reason over a lot of material. Claude or ChatGPT both work. Where you can, use Claude Code or a workspace version that lets the model connect directly to your tools. It goes further, and it won’t lose the thread halfway through a large analysis.
And you need your systems connected to it. This is the part that makes or breaks the exercise. With your CRM, transcripts, invoices and inboxes connected, the AI reads them itself and works from your real history. Without that connection you are reduced to pasting documents in one at a time, which caps how much it can see, fractures the analysis, and quietly drops the older records that matter most. Connected, it is an afternoon. Unconnected, it is a worse answer and twice the work.
So before you run anything: choose your AI, connect every source you can: transcripts, CRM, closed-lost, invoices, email, Slack, Linear or Jira. Only then paste the prompt.
05
Run the work in four fresh conversations. Each stage produces an evidence block for the next, so you see useful output early and a failure never costs the whole run.
The final prompt returns a branded Inside Startups document you can save as a PDF.
Stage 1, Census and map. Counts what exists, quarantines junk, maps what buyers actually purchase, and picks the six calls worth reading in full.
Stage 2, The calls. Reads the nominated calls in full, which is the only place your customers' verbatim language can come from.
Stage 3, The written record. Searches email, chat and, if connected, contracts and billing by intent, where loss reasons usually live.
Stage 4, The document. Produces the branded report from the evidence blocks alone, so every finding stays checkable.
Run each stage in a fresh conversation. Copy the evidence block at the end of each stage and paste it at the end of the next stage's prompt. If a stage fails, re-run that stage. The blocks you already have stay good.
Stage 01
About ten minutes. Counts what exists, quarantines junk, maps what your buyers actually purchase, and picks the six calls worth reading in full.
Prompt locked
Enter your name and work email above to unlock and copy the full prompt.
Each stage ends with an evidence block. Copy it, then paste it at the end of the next stage's prompt. If a stage fails, re-run that stage. The blocks you already have stay good.
Stage 02
About ten minutes. Reads the nominated calls in full. This is the only place your customers' verbatim language can come from.
Prompt locked
Enter your name and work email above to unlock and copy the full prompt.
Each stage ends with an evidence block. Copy it, then paste it at the end of the next stage's prompt. If a stage fails, re-run that stage. The blocks you already have stay good.
Stage 03
About eight minutes. Searches email, chat and, if connected, contracts and billing, by intent. Loss reasons usually live here, not in the CRM.
Prompt locked
Enter your name and work email above to unlock and copy the full prompt.
Each stage ends with an evidence block. Copy it, then paste it at the end of the next stage's prompt. If a stage fails, re-run that stage. The blocks you already have stay good.
Stage 04
About eight minutes. This is the branded-document stage, producing the report from the evidence blocks alone so what it says is checkable.
Prompt locked
Enter your name and work email above to unlock and copy the full prompt.