Personas
Date: 2026-08-17
Fictional characters standing in for user segments, built from research. The technique is sound and its usual execution isn’t — a persona with a stock photo, a name and an invented coffee preference is a costume on an assumption.
A persona is a composite character representing a recognisable group of users, built from research, used to keep decisions anchored to someone other than the person making them.
What a persona is for
One job: making “the user” specific enough to argue about.
WITHOUT
"users want more filters"
→ whose? which users? how do we
know? no one can disagree usefully
WITH
"Priya, who shops on her phone during
a commute and knows exactly what she
wants, needs to filter by brand
before anything else"
→ now it's checkable, and someone can
say "that's not what the research
showed"
The value is in constraining the argument, not in the character. A persona nobody references in a decision has cost more than it returned.
What a useful one contains
BEHAVIOUR what they actually do
→ from research
CONTEXT device, place, time,
interruptions
GOALS what they're trying to
achieve
FRUSTRATIONS what currently blocks it
DECISION CRITERIA what makes them choose
KNOWLEDGE vocabulary, expertise level
A REAL QUOTE in their words
Nothing about their coffee order, their car, or their weekend. Demographic and lifestyle detail is where personas earn their reputation, and it’s almost never load-bearing — knowing someone is 34 predicts far less about their behaviour than knowing they always shop on a phone.
Include the evidence. A persona should be traceable to the sessions and data it came from, or it’s indistinguishable from a guess — Research Repositories.
The usual failures
- Invented from assumption. The workshop persona, produced in an afternoon, describing the team’s mental model back to itself
- Demographic rather than behavioural. “Female, 25–34, urban” groups people who behave completely differently and separates people who behave identically
- Too many. Seven personas means nobody remembers any of them. Three is usually the ceiling
- Never updated. A persona from a research round four years ago describes a market that has moved
- Designed for, rather than checked against. They’re a reference during a decision, not an input to a brief
The behavioural alternative
For ecommerce specifically, grouping by behaviour and situation usually beats grouping by person:
| Demographic | Behavioural |
|---|---|
| 35–44, professional | Knows the exact product, wants to reorder fast |
| 25–34, urban | Researching for the first time, needs reassurance about suitability |
The second pair describes what the interface must do; the first doesn’t. And the same human is often both, in different weeks — which is the strongest argument against person-shaped segmentation and for situation-shaped — Jobs To Be Done.
Related artefacts worth distinguishing
PERSONA a person-shaped summary
of a segment
JOB a situation someone is
trying to make progress in
— Jobs To Be Done
SEGMENT a measurable group in your
data — Segmentation (analysis)
PROTO-PERSONA an explicitly assumption-based
persona, labelled as such,
to be validated
See: Jobs To Be Done · Segmentation (analysis)
Proto-personas are the honest version of the workshop exercise. Building one is fine; the failure is producing one and then treating it as research. Labelling it as an assumption means someone can later ask whether it held.
Whether to bother
A qualified yes, with conditions. They’re worth building when:
- They come from actual research — User Interviews, Diary Studies
- There are three or fewer
- They’re behavioural, not demographic
- They get referenced in real decisions
If any of those fails, the effort is better spent on the research itself. A team that has watched eight customers attempt to buy something has the understanding a persona was meant to transmit — and the artefact matters most when the people making decisions weren’t in the room.