How to Build a Buyer Persona from Real Data, Not Guesswork

Most buyer personas start with a stock photo, a made-up name and a line like "25–34, likes travel and coffee". They look good in a presentation and change nothing. A useful persona is different: it is a short summary of evidence about how a specific group of customers makes a buying decision. If it does not change what you write, where you advertise or what your sales team says, it is decoration.

This guide shows how to build personas from data you already have, plus a handful of customer conversations.

What a persona should describe

Demographics help with media targeting, but they rarely explain why someone buys. Two 35-year-old managers in the same city can buy for completely different reasons. A working persona describes a decision:

• The job to be done — what the customer is trying to achieve, in their own words.
• The trigger — what happened that made them start looking.
• Decision criteria — what they compare, and in which order.
• Objections and fears — what almost stopped them.
• Influencers — who else has a say: a partner, a boss, an accountant, a friend.
• Information sources — where they search, whom they ask, what they check.

In B2B, build one persona per role in the buying group. The person who uses your service, the person who signs the contract and the finance person who approves the payment have different questions.

Step 1: Start with the data you already have

Before you talk to anyone, look at what is already sitting in your systems. It costs less than new research and keeps your interviews focused.

• Sales and CRM data. Who buys, what, how often, through which channel and at what average order value. Sort customers by revenue and look for clusters: for example, frequent small orders versus rare large ones. These are often different personas.
• Support tickets and chats. This is where customers describe problems in their own words. In many markets in our region, much of this lives in WhatsApp and Instagram direct messages rather than in a ticket system. Export or simply read a sample of 50–100 recent conversations and note the questions that keep coming back.
• Sales team notes. Ask for the reasons behind lost deals and the objections heard most often. In my experience, salespeople know patterns that never reach the marketing team.
• Website analytics and search queries. The pages people view before contacting you, and the words they search for, show what they need to know before buying.
• Reviews — yours and your competitors'. Complaints about competitors are an honest list of unmet needs.

Step 2: Talk to real customers

Data shows what people do; conversations explain why. Start with around ten interviews per segment and stop when you no longer hear anything new.

Interview a mix: recent buyers, people who considered you and chose someone else, and customers who stopped buying. If you only talk to loyal fans, you get a flattering portrait, not a useful one.

Ask about past behaviour, not opinions about the future:

1. "Tell me about the last time you bought something like this. How did it start?"
2. "What made you start looking at that particular moment?"
3. "What other options did you consider?"
4. "What almost stopped you from buying?"
5. "Who else was involved in the decision?"
6. "Where did you look for information or advice?"

Avoid questions like "Would you buy this?" or "What features do you want?" People are poor predictors of their own future behaviour, and they tend to be polite to the interviewer.

Record the calls with permission. AI tools can transcribe and summarise them quickly, but read the original quotes: summaries smooth out exactly the phrases that later make good copy. If your market is multilingual, interview people in the language they buy in — customers who speak different languages may describe the same need in different words.

Step 3: Find patterns, then check the size

Put quotes and data points side by side and group them by motivation and behaviour, not by age or gender. You are looking for groups that share a trigger, criteria and objections.

For most small and mid-size companies, two to four personas are enough. If you end up with nine, you have a list of individuals, not a tool. Then go back to the sales data and check how much revenue each group represents. A vivid persona that brings a tiny share of revenue should not drive your strategy.

Step 4: Write it on one page

Use the same template for every persona so they are easy to compare:

• Name and situation — one line, e.g. "First-time founder hiring an agency".
• Goal — the job to be done.
• Trigger — what starts the search.
• Top three criteria — ranked.
• Main objections.
• Channels and sources — where they look and whom they trust.
• Influencers in the decision.
• Their words — three or four real quotes.
• What this means for us — key message, best channels, offer, what sales should say.
• Evidence — number of interviews, data sources used, date of the last update.

The last line matters. It tells everyone that this is research, not opinion, and when it was last checked.

Common mistakes

1. Inventing instead of researching. A persona written in a workshop reflects the team's assumptions, not the market.
2. Demographics over motivation. Age and income do not explain decisions.
3. Too many personas. Nobody can keep nine people in their head.
4. Interviewing only happy customers. Lost and former customers teach you more.
5. Filing it away. If the persona is not used in briefs, ad copy and sales scripts, the work is wasted.
6. Never updating it. Markets, prices and channels change. Review personas at least once a year, or after any major shift.
7. Treating it as an ad-targeting setting. A persona guides the message and the offer; audience targeting on ad platforms is a separate decision.

Where to start this week

Pull the last three months of sales data, read 50 recent customer conversations and book five customer calls. Then write one persona on one page and use it to change one thing — a landing page headline, an ad, or the first message your sales team sends. If that change performs better, you have proof the persona is real.