Sophie, 34, marketing manager, "values efficiency and podcasts". The workshop persona has one merit — aligning teams — and one manufacturing defect: it rests on what you imagine about customers. Your site's data tells you what visitors do. Here is how to build personas that predict, not just decorate.
Quick Answer: how do you build data-driven personas?
In four moves:
- Start from observed behaviour — your analytics naturally segments traffic by source channel, device, OS, country, recurrence and behaviour (pages, engagement, conversion).
- Look for groupings that diverge — populations with clearly different conversion rates, journeys or favourite content: the signal that distinct treatment will pay.
- Embody each segment as a named persona with measured facts (traffic share, conversion, typical journey) rather than invented traits.
- Activate them — A/B test targeting, personalization, content priorities. A persona that drives no decision doesn't exist.
Three to five actionable personas beat ten detailed profiles.
Why the workshop persona is no longer enough
Three documented weaknesses of the classic persona:
- It freezes assumptions. Built once, pinned to the wall, never confronted with data — it outlives its own expiry by years.
- It over-weighs demographics. Age and job predict web behaviour poorly: two identical "Sophies" on paper behave differently depending on whether they arrive from a Google comparison or a LinkedIn post.
- It is not technically actionable. You can neither target nor measure "likes podcasts". You can target "mobile + first visit + arrived from search".
The behavioural persona reverses the construction: criteria are measurable by definition, hence targetable and verifiable.
The materials: what your data already knows
Without a single questionnaire, serious analytics provides per visitor (anonymously):
- Origin: channel (SEO, social, direct, referral, AI/LLM), campaign, country.
- Context: device, OS, screen size — powerful proxies of usage context (on the move, at a desk).
- Behaviour: pages seen, depth, duration, recurrence, time of visit.
- Outcome: conversion, goal type, value.
Combining these axes draws sharply differentiated populations. A real e-commerce example: "desktop + SEO + comparison pages" often converts 3 to 5 times better than "mobile + social + direct product entry" — two de facto personas, calling for two treatments.
Building: from segment to persona
Find the segments that matter
The decisive test: a segment deserves a persona if it diverges on a business metric (conversion, basket, retention) AND weighs enough to justify treatment (rule of thumb: ≥ 10–15% of traffic or value). Our behavioural segmentation guide details the splits that pay.
Write the profile — data edition
A useful profile fits on one page: name + identity sentence ("The methodical comparer — desktop, SEO, reads comparisons"), measured facts (traffic share: 18%; conversion: 4.2%; typical journey: comparison → product → FAQ → quote), main observed friction (in replay: looks for prices, finds them late), priority treatment (show prices in comparisons). No invented personality traits required.
Activate, or discard
Three concrete activations, in increasing difficulty:
- Prioritise content and UX: every major persona must have its path served (the Comparer wants tables; the Rushed mobile visitor wants the price and the button).
- Target A/B tests: testing a variant only on the relevant segment multiplies test sensitivity and avoids imposing a change on those it disserves.
- Personalize: adapt a block, a message or a section order per segment — the most profitable step once the first two are in place.
Keeping personas alive
A behavioural persona recomputes — that is its strength. Review quarterly: have traffic shares moved? Does an emerging channel (AI assistants, typically) create a new population? Have treatments narrowed the conversion gap between personas? A persona whose gap has closed has done its job; the next one is waiting.
In practice with Mirage — A/B testing personas are built directly on your real data: source, device, OS, country, with an immediate estimate of the audience share and potential of each combination. You then target variants and measurements per persona — no cookies, no questionnaires, no fiction. Free 30-day trial.
FAQ
What is a marketing persona?
A representation of a typical customer group, used to align teams and guide product, content and acquisition choices. The classic persona is built in workshops (demographics, imagined motivations); the behavioural persona is derived from real data: device, source channel, country, observed navigation and conversion behaviour.
What is the difference between a persona and a segment?
The segment is the operational split (measurable criteria: mobile + social traffic + new visitor); the persona is its narrative form, making it actionable for teams. The right flow: data creates segments, segments become personas, personas drive tests and personalization.
How many personas do you need?
Three to five almost always suffice. The criterion is not descriptive richness but actionability: two personas calling for the same treatment are one; a persona you can neither target nor measure is fiction. Every persona must map to an identifiable, addressable share of traffic.
How do you validate that a persona is right?
By prediction: a good persona predicts measurable behaviour. If your "time-pressed B2B decision-maker" persona is real, the matching segment must show short, pricing-and-contact-oriented sessions. If the data contradicts the profile, you fix the profile — that is the whole advantage of data-grounded personas.