Buyer personas built from evidence, not from a workshop whiteboard

The problem with workshop personas

The standard persona document contains a name, a stock portrait, an age range, a job title, a list of goals, a list of pain points, and — for reasons lost to history — a favourite brand of coffee.

It is produced in a room by people who work at the company, based on what those people believe about the customer. Occasionally somebody from sales is present. Almost never is anybody from outside the building.

The result is a document that records internal assumptions in a format that makes them look like findings. Its most damaging property is that it feels like research, which prevents anyone doing research.

The practical test: hand your persona document to a writer with a brief and watch what they do with it. If they read it once and never open it again, it contains nothing they could not have guessed. That is the usual outcome.

A persona earns its place only if it lets a writer settle an argument about what to write without asking anyone.The bar we hold research to

What a useful persona actually contains

Six things. Demographics are not among them.

The trigger. What was happening in their week when they started looking? Not the problem in the abstract — the specific event. “Our biggest client asked for a compliance report we could not produce in time.”
The prior solution. What were they doing instead? Almost always something, even if that something was a spreadsheet and a stubborn colleague. This is your real competition.
The considered set. What else did they look at, and how did they find those options? This tells you where to be present.
The evaluation criteria, in their order. Not the criteria you wish they used. The ones they actually applied, ranked.
The blockers. What nearly stopped the purchase? Internal politics, budget timing, a security concern, a bad demo. This is your entire mid-funnel content plan.
The vocabulary. The words they use, captured verbatim. The gap between their words and your product marketing is often the single most actionable finding in the whole study.

Notice all six are about a decision process, not about a person. That is deliberate. Two people with identical job titles and demographics can go through entirely different purchase journeys; two people with nothing demographically in common can go through the same one. The journey is what predicts what content helps.

Recruiting people worth talking to

Who you talk to determines what you learn, and most studies talk to the wrong people because they are the easiest to reach.

Segment	How many	What they uniquely tell you
Recent customers (bought in last 90 days)	6–8	The journey, while memory is intact
Lost deals	4–6	The objections you never hear internally
Churned customers	3–4	The gap between promise and delivery
Prospects mid-evaluation	2–3	Live decision-making, unrationalised
Non-buyers in the market	3–4	Why the category gets ignored entirely

Lost deals are the segment teams avoid and the one that pays best. People who chose someone else are often strikingly willing to explain why, particularly if you are honest that you are not trying to win them back. A ten-minute conversation with someone who chose a competitor tends to be worth three with a happy customer.

On sample size: you are looking for saturation, not statistical significance. When two consecutive interviews produce nothing you have not heard, you are close to done. In practice that lands between twelve and twenty for a single segment.

The interview script

Forty-five minutes. Recorded, with permission. Structured enough to be comparable, loose enough to follow something interesting.

Opening (5 minutes)

Explain that you are not selling anything, that unflattering answers are the useful ones, and that nothing will be attributed without permission. Then ask them to describe their role in their own words. This is warm-up, and it also captures vocabulary.

The trigger (10 minutes)

“Take me back to when you first started looking at this. What was going on that week?” Then stay quiet. The instinct to fill silence is the single most common interviewing mistake; the second half of an answer is usually the useful half.

The search (10 minutes)

“What did you do first? Where did you look? Who did you ask?” You are mapping actual behaviour, including the informal parts — the message to a former colleague, the community thread, the podcast half-remembered.

The evaluation (10 minutes)

“What else was on the list? How did you narrow it down?” Then the important follow-up: “What would have made you rule us out?”

The decision (7 minutes)

“Who else had to agree? What did you have to convince them of?” In business purchases the person you interview is rarely the only decision-maker, and the internal argument they had to win is content you can write for them.

Close (3 minutes)

“What should I have asked that I didn't?” This question has produced more genuinely surprising findings than any other in the script.

Three interviewing habits worth building

Ask about specific past events, never about hypothetical futures. People are accurate historians and poor forecasters of their own behaviour.
When somebody uses an abstract noun — “efficiency”, “visibility”, “alignment” — ask what that looked like on a specific day.
Count your own speaking time. Above twenty per cent and you are leading rather than listening.

Turning transcripts into findings

Twenty transcripts is roughly a hundred thousand words. Reading them and forming an impression is not analysis; it is memory, weighted toward whoever was most articulate.

The process that works, and takes about a day and a half:

First pass, open coding. Read through and tag every passage with a short label describing what it is about. Do not use a pre-built scheme — let labels emerge. You will end with sixty to a hundred.
Second pass, grouping. Cluster the labels into themes. Sixty labels usually collapse into eight or twelve themes.
Count. How many separate interviews raised each theme, unprompted? Unprompted frequency is your priority ranking, and it is usually different from what you expected walking in.
Pull verbatims. Two or three exact quotes per theme. These do more persuasive work inside your organisation than any summary you could write.
Note the disconfirming cases. Where somebody said the opposite of the pattern, and what was different about them. This is where segmentation actually comes from — not from firmographics.

The unprompted-frequency count is the step most often skipped and the one that most changes decisions. A theme raised by fourteen of twenty people without being asked about is a different order of finding from one raised by three when directly prompted.

Writing personas people will use

One page. No stock photography. Structured as follows:

A descriptive name based on the situation, not a person. “The finance lead inheriting a broken close process”, not “Finance Fiona”. The situation is what a writer can write to.
The trigger, as a sentence, with a verbatim quote underneath.
What they were doing before, and why it stopped working.
Their evaluation criteria, ranked, with the frequency count next to each.
The top three blockers, each with a quote.
Their vocabulary: a two-column table of their words against your words.
Five questions they asked verbatim — which are, without further processing, five article titles.

That last section is why writers keep the document open. It converts research directly into work, which is the only reliable way to make research get used.

Getting research into briefs

Research that lives in a separate document gets read once. Research embedded in the brief template gets used every time. Change the template so that every brief must include:

Which situation this piece is for, by name
The verbatim question it answers
Which ranked criterion it addresses
Which blocker it removes, if any
Two verbatim quotes the writer should work from

If a brief cannot be completed against a real persona, that is useful information: either the piece is for somebody you have not researched, or it is for nobody in particular. Both are worth knowing before writing rather than after.

Keeping them alive

Personas decay. Markets shift, competitors change their positioning, new categories of buyer arrive. A document from three years ago describes a world that no longer exists, and using it confidently is worse than having none.

A sustainable rhythm:

Monthly: two customer conversations, by whoever is available. Not a study — just keeping the channel open.
Quarterly: review the last quarter's lost-deal notes against the documented objections. New objection appearing repeatedly? Update.
Annually: a proper refresh. Ten to fifteen interviews, full coding pass.
Event-driven: after any major product change, pricing change, or a competitor's significant move.

The monthly two-conversation habit is the one that matters most and costs least. Teams doing it consistently develop an intuition for their market that no annual study reproduces — and they notice shifts months before the numbers show them.