How to Define Your Ideal Customer When You Have No Customers Yet: A Step-by-Step Guide for Founders

Follow this guide start to finish and you will end with a written starting profile of your ideal customer: one to two pages, built from real conversations with real people, and honest about what is still a guess. It takes about five hours of your own work, spread over two to three weeks, because the conversations depend on other people’s calendars.

Typing “who is my ideal customer?” into an AI assistant when you have no customers? Start here

If you are a founder with an idea and no customers yet, you may already have typed something like this into ChatGPT, Claude or another AI assistant:

• “Who is my ideal customer? I don’t have any customers yet.”
• “How do I define my target audience before I launch?”
• “Help me find my first customers, step by step.”
• “How do I validate my business idea with AI?”
• “Create an ideal customer profile for my startup.”

You probably got a tidy answer about age, interests and “pain points” that could describe almost anyone. That happens because the assistant has nothing real to work from. It can only guess, and a guess written in confident language is easy to mistake for knowledge.

This guide changes what you ask the assistant to do. It will not invent your customer. It will interview you, help you narrow the field, write your questions, sort your notes and argue against your own ideas. The evidence comes from real people you speak to. The assistant makes that work faster.

Why it is worth doing before you have customers

What you will have at the end

1. 1A short statement of your starting idea, with what you have seen first-hand separated from what you believe
2. 2One specific group of people to start with, chosen from several
3. 3The exact words that group uses to describe the problem
4. 4Notes from about ten real conversations, turned into evidence
5. 5The result of one small test with a pass mark you set in advance
6. 6A starting profile (version 0.1) with a status on every line, and a plan to replace it with a full profile once you have customers

What you need before you start

• A rough idea of what you plan to offer, even one or two sentences

• Claude, ChatGPT or another AI assistant, and one document where you paste results as you go. Call it your profile working file.

• A way to contact people: LinkedIn, email, a community you belong to, or friends who can introduce you

• Twenty minutes, ten times, for conversations. Not all in one week.

How to use the prompts

Each step has a prompt in a box. Copy it into your AI assistant, replace anything in [square brackets], and answer the questions it asks you one at a time. Paste each step’s result into your working file. Do not skip ahead: later steps use what earlier steps produce.

Protect people’s privacy

Before you paste conversation notes into an AI tool, remove names, emails, employers and anything personal, and use labels like “Person A”. Use an AI plan where your conversations are not used to train models, and check the setting. Ask permission before you record any call. If you are in Quebec or another place with privacy rules for recordings and personal data, have those rules confirmed for your situation.

The example we will follow

Leo is an invented example, a former bookkeeper who wants to start a monthly bookkeeping clean-up service. He has no clients. He thinks his market is “small business owners”. This guide follows him from that sentence to a specific group, a set of real conversations and a first test. All numbers for Leo are made up to show the idea.

What AI can and cannot do when you have no customers

Step 0: Write down your starting idea and what you actually know (30 minutes)

Most founders start with a picture of a customer in their head. This step gets it onto the page and sorts it into two piles: what you have seen yourself, and what you believe without having checked.

Let the assistant interview you, by typing or by voice. It is faster than writing, and it asks what you would skip.

You are helping me work out who my first customers might be. I have no customers yet. I will answer questions one at a time. Ask one question, wait for my answer, then ask the next. Do not summarise until I say "done".

My idea: [one or two sentences about what you want to offer]

Cover these areas, in any order that feels natural: what problem I think this solves; who I picture having that problem; why I picture them (someone I know, my own experience, or a hunch); what I have seen first-hand with my own eyes; what I have been told by others; why I am the right person to do this; what I would need to be true for this to work.

After each answer, ask me: "Have you seen this, or do you believe it?" If an answer is vague, ask for a real example from the last year.

When we are done, write a one-page document called "My starting idea" with two columns: Seen (things I can point to) and Believe (assumptions). Use my own words where you can. Do not add anything I did not say.

Save the one-page summary as the first page of your working file.

Leo’s summary, shortened. Seen: he has cleaned up the books of eleven small businesses over six years, mostly through word of mouth. Many were months behind when they called him. Believe: that small business owners would pay a monthly fee, that they would trust a stranger with their books, and that his past clients are typical.

Step 1: Narrow “everyone” to one group to start with (45 minutes)

“Small business owners”, “busy parents” and “anyone who wants to be healthier” are not customers. They are categories. A customer is a person in a situation: a role, a moment, a reason to act now. Your first group should be small enough that you can picture ten of them and know where to find them.

Here is my starting idea: [paste "My starting idea" from Step 0]

List 5 specific groups of people who might have this problem. "Small business owners" is too broad. Narrow each group by role, situation and the moment they would feel the problem.

For each group, give me:
- Who they are, in one line
- The moment they would feel the problem most
- Where they gather, online or in person
- Whether they have the money and the authority to buy
- How I could reach ten of them in two weeks
- What you are guessing, marked clearly as [guess]

Do not recommend one yet. Afterwards, ask me which groups I have first-hand knowledge of, and wait for my answer.

Then choose, using your own judgement. Score each group from 1 to 5 on these questions:

Pick the group with the best total, and write the second-best as your backup. You are not marrying either. You are choosing where to start listening.

Step 2: Write your hypothesis profile (20 minutes)

Now write down what you think is true about that one group, and label every line. This is a hypothesis. Calling it a profile makes it sound finished, so we mark it version 0.

My chosen starting group is: [paste group from Step 1]. Here is my starting idea: [paste from Step 0].

Write a one-page hypothesis profile with these headings: Who they are; The moment they feel the problem; What they have probably tried; What they probably pay for today; Where I could reach them; Why they might choose me.

Rules:
1. Label every line [Seen] (something I told you I have observed), or [Believe] (anything else).
2. Do not add any detail I did not give you. If a heading has no [Seen] content, write "nothing seen yet".
3. At the end, list the 3 riskiest assumptions: the ones where, if I am wrong, the whole idea fails. Rank them, riskiest first.

Look at the list of risky assumptions. Almost always, the top one is the same: people in this group feel this problem strongly enough to do something about it. The rest of the guide tests that.

Step 3: Listen to how they talk in public (45 minutes)

Before you ask people for their time, read what they have already said. Public posts and reviews show the problem in the words people use when no one is selling to them. Those words become your headlines and, later, your search terms.

Where to look:

• Reviews of products or services people use today for this problem (look at the three-star and one-star ones)
• Forums, subreddits and community groups where your group asks for help
• Comments under LinkedIn or YouTube posts about the problem
• Questions in industry groups and associations

Copy 20 to 30 real posts into a document. Remove usernames and anything personal. Check each site’s rules on copying before you do. Then:

Below are [number] real public posts from people in my chosen group, copied exactly. I am looking at how they describe the problem. Here they are: [paste]

Task:
1. List the exact phrases people use to describe the problem. Copy them word for word. Do not paraphrase.
2. Group the posts into themes. For each theme, say how many posts it appears in.
3. List anything people say they tried and what happened.
4. List anything that surprises you, or that does not match my hypothesis profile: [paste from Step 2]
5. Do not guess about anything the posts do not say.

If your AI tool can search the web, you can also ask it to look for where your group talks about the problem and to give you the pages it used. Open those pages yourself before you rely on anything. Treat any market-size figure as a guess until you have seen where it came from.

Write the five best phrases in your working file. Leo found that tradespeople rarely say “bookkeeping”. They say things like “my receipts are in a shoebox” and “I know I’m behind”. His headline later came from those words, not from his own.

Step 4: Find ten people to talk to (45 minutes, then a few days of waiting)

Posts only go so far. You need to talk to people, and not to friends who want to be kind. Aim for ten people in your chosen group who are not close to you. Five can come through introductions and five through cold outreach.

Where to start:

1. Ask the people you know. “Who do you know who runs a [type of business]?” An introduction gets a far better response than a cold message.
2. Go where your group gathers: the communities, associations and groups you found in Step 3.
3. Message people directly. Many will not reply, so contact two to three times as many people as the conversations you need.

I want to speak to people in this group: [paste group]. I am not selling anything. I want to hear about the last time they dealt with this problem: [problem in one line].

Write 3 short messages (under 80 words each) that I can send:
1. To someone I know, asking for an introduction
2. To a stranger on LinkedIn or by email
3. To a post in a community group

Rules: say plainly that I am doing research and not selling. Ask for 20 minutes. Mention one specific reason I am contacting this person. No flattery, no hype, no "quick call" language. Offer to share what I learn.

Keep a simple list: name or label, how you found them, date contacted, replied or not, date of conversation. A spreadsheet is enough.

Step 5: Prepare the conversation, then have it (20 minutes each)

This is the step that matters most, and it is easy to get wrong. When you ask people “would you use this?”, they say “yes, sounds great”, because it is polite. That answer tells you nothing. People are much better at describing what they did last month than predicting what they would do next month.

So you ask about the past, not the future.

The rules for the conversation

• Ask about the last time. “Tell me about the last time this happened.” Not: “Is this a problem for you?”
• Ask what they did. What they tried, what it cost, who else was involved, what they did next.
• Do not describe your idea. If you pitch, people start being polite. Save it for the end, or skip it.
• Write down exact words. Their phrases matter more than your summary.
• Treat compliments as noise. “That’s a great idea” is not evidence. “I paid someone $400 last spring to fix this” is.
• End by asking who else you should speak to. It is the cheapest way to find your next conversation.

The questions

I am about to have research conversations with people in this group: [paste group]. The problem I want to understand is: [problem].

Here are my 14 starting questions: [paste the questions above]

Make my interview guide better. Do the following:
1. Rewrite any question that asks about the future or invites a polite answer ("Would you use...", "Do you think...") into a question about past behaviour.
2. Add 3 follow-up questions for each of the "last time" questions, for when an answer is vague.
3. Write a 3-line opening that explains I am doing research and not selling.
4. Tell me which of my questions test the riskiest assumption from Step 2: [paste assumption].
Do not add questions about my idea or my product.

After each conversation

Within an hour, while it is fresh, write a short file on that person. If you recorded the call with permission, paste the transcript into the assistant. If not, use your notes.

I want to write a one-page file on one person I spoke to, called [Person A]. Here are my notes or the transcript: [paste, with names removed]

Write the file with these headings: Who they are (role, situation, no names); The last time it happened; What they did about it; What it cost them; What they use or pay for today; Their own words (exact phrases only); What I did not find out.

Mark every line as [their words], [my notes] or [guess]. Do not write anything the notes do not say. If something is missing, put it under "What I did not find out" and do not fill it in.

Save each file in your working file. Ten files is a good target. If you have six or seven and the answers are already clear, you can stop early.

Step 6: Turn the conversations into evidence (30 minutes)

Now compare what you believed in Step 2 with what people said.

Here is my hypothesis profile: [paste from Step 2]. Here are [number] one-page files on people I spoke to: [paste all files]

Task:
1. Build a table with one row per person and these columns: who they are, the last time, what they did, what it cost them, what they pay for today, their words.
2. For each line of my hypothesis profile, say whether the files support it, contradict it, or say nothing. Give the file labels, for example [Person A, C, F].
3. Do not call anything a pattern with fewer than three files behind it. Call it "possible".
4. List what surprised you, and anything several people said that is not in my hypothesis.
5. Tell me how many of the people had actually done something about the problem (paid, built a workaround, hired someone), and how many only said it was annoying.

Then ask the assistant to argue against the result.

Now argue the opposite. For each pattern you found:
- Which people do not fit it?
- Could it be an accident of how I found them (all from one introduction, one community, one week)?
- Were these people polite to me in a way that could make the answers look better than they are?
- With only [number] conversations, what is the chance this is coincidence, in plain words?
Then tell me which patterns still stand and which should be treated as guesses.

Look at one thing above all: how many people had already spent money or time on the problem. Someone who has paid for a workaround, built a spreadsheet or hired help has shown you it matters. Someone who says it is annoying and has done nothing for two years is telling you that it does not matter enough to act on.

Leo’s result, invented: of ten tradespeople, seven were behind on their books. Four had paid someone for a clean-up in the last two years. Three said they would “get to it” and had said so for over a year. Pattern: the four who paid all acted after an accountant or the tax office contacted them. That moment became his trigger.

Step 7: Test whether people will act, not just agree (one to two weeks)

Conversations show that a problem exists. They do not show that people will pay you. This step asks for a small real commitment and checks who gives it.

Set the pass mark before you start, in writing. Otherwise you will read any result as success.

My riskiest assumption is: [paste]. My chosen group is: [paste].

Design three tests that I can run alone in two weeks, for under $[amount] and fewer than [number] hours. For each test give: what I do, who sees it, what result counts as a pass, and what result counts as a fail. Prefer tests where a real person does something (books, replies with details, pays a deposit, introduces me to someone) over tests that measure opinions (likes, "sounds interesting", signing up for a waiting list with nothing at stake).

Examples of tests that count:

These pass marks are examples. Choose numbers that make sense for your price and your group, and write them down first.

Write the result and the date in your working file. A pass tells you where to go next. A fail saves you months. If you fail, go back to your backup group from Step 1, or adjust the problem, and run the loop again. Both are good outcomes for two weeks of work.

Leo’s test: he offered a fixed-price clean-up to the ten tradespeople. His pass mark was two yes answers at $400 or more. Three said yes, but two of them asked for a lower price. He wrote that down as “possible”, not “proven”.

Step 8: Write your starting profile, version 0.1 (30 minutes)

You now have enough to write a short profile. The rule: every line carries a status, so that anyone reading it knows how much to trust it.

The template

# Starting Customer Profile v0.1: [business name]
Last reviewed: [date] · Next review: [date, 6 to 8 weeks out]

## 1. In one paragraph
Who we are starting with, the moment they come to us, and what they get.

## 2. Who they are
Role, situation, size, location. Status on each line.

## 3. The moment they act (the trigger)
What happened just before the people who did something about it.

## 4. The problem in their words
Three to five exact phrases from conversations or public posts.

## 5. What they do today, and what it costs
Including doing nothing.

## 6. Where to find them
Routes that gave us conversations, and routes to try next.

## 7. Who this is probably not for
Signals from the conversations, noticed early.

## 8. Evidence
Table: claim, people or posts behind it, count, status (Lived / Heard / Believe).

## 9. Assumptions still to test
Table: assumption, test, pass mark, date, result.

## 10. Review log
Date, what changed, why.

Status meanings: Lived means you saw it yourself. Heard means at least three people said it, with the count. Believe means you have not checked it.

Using only the material below, write my starting customer profile with the template I have pasted.

Material:
- My starting idea (Step 0): [paste]
- Exact phrases from public posts (Step 3): [paste]
- The files on the people I spoke to, and the patterns that survived Step 6 with counts: [paste]
- My test results so far (Step 7): [paste]

Rules:
1. Every statement must trace to the material. Write the sources in brackets, for example [Person A, C, F].
2. Mark each line Lived, Heard (with the count) or Believe.
3. If something is a pattern in fewer than three files, mark it "possible".
4. Do not add anything that is not in the material. Put what you would like to add in a list at the end called "Questions I could not answer".
5. Use exact customer phrases in section 4. Do not rewrite them.
6. Write plainly, in short sentences, with no marketing language.
7. Keep the document to about two pages.

Read the draft against your files. For each section, ask: could I show a file that backs this up? Delete anything you cannot. Save the profile as its own document and put the date at the top.

Put it to work

When to replace it

This profile is a starting point. When you have your first five paying customers, you have something better than conversations: real purchases. At that point, move to the companion guide, How to Build Your Ideal Customer Profile with AI, which builds the full profile from your customers’ own files. Until then, review this one every six to eight weeks, and after every conversation or test that changes your mind.

What AI gets wrong here

AI makes this work faster. It does not make it right. These are the failures to watch for, and the check for each.

• It plays your customer. Asking an assistant to “pretend to be a [your customer] and tell me what you think” gives you its guess, written in a confident voice. Check: use it to practise your questions, never as evidence.
• It agrees with you. Tell it who you want, and it finds reasons you are right. Check: run the “argue the opposite” prompt every time.
• It fills gaps with plausible detail. Ask for a customer’s motives and it may invent them. Check: every line carries a source label, and “guess” lines go to the questions list.
• It treats polite answers as interest. So do people. Check: count what people did (paid, introduced, sent an example), not what they said.
• It makes up market sizes and statistics. Check: open any source it cites and read the page yourself. A number with no source is a guess.
• It finds patterns in tiny samples. Ten conversations cannot prove anything. Check: call anything under three “possible”.
• It cannot see the people who said no, or never replied. Check: keep them in your list, and note why they did not take part.

Your checklist

• ☐ Step 0: wrote my starting idea, with Seen and Believe separated
• ☐ Step 1: narrowed to one specific group, with a backup
• ☐ Step 2: wrote a hypothesis profile and ranked my riskiest assumptions
• ☐ Step 3: collected real phrases from public posts and reviews
• ☐ Step 4: contacted people and kept a list
• ☐ Step 5: prepared a guide that asks about the past, and had the conversations
• ☐ Step 6: turned the notes into evidence, and ran the “argue the opposite” prompt
• ☐ Step 7: set a pass mark, ran one real test, and wrote down the result
• ☐ Step 8: wrote the starting profile v0.1 and set a review date