Your static persona is actually five different users.
Adaptive AI doesn’t care who someone is on paper: it needs to know what state they’re in right now. Audience Factors maps your users across five behavioural stages, grounded in Prochaska’s Stages of Change and Fogg’s B=MAP, so the system adapts to readiness, not a demographic.
A persona is a snapshot. Your user is in motion.
The same person is a different user on Monday morning than on Friday night: more motivated, less able, open to a nudge or desperate to be left alone. A persona freezes them at one demographic; an adaptive system has to read the state they’re actually in.
The persona trap
- Static PDFs, frozen in demographics
- “A busy 34-year-old”, and nothing your AI can act on
- Says nothing about motivation, stress, or readiness
- Stable goals that assume people never change
The Audience Factors way
- Attribute-based, dynamic state modelling
- Real-time signals over assumed intent
- The exact intervention the moment calls for
- A micro-commitment, a deep challenge, or silence
Model the state, not the stereotype, and your system finally has something to adapt to: a reading it can act on, moment to moment, instead of a profile it has to guess around.
Two proven behavioural models, one map
The trajectory comes from Prochaska’s Stages of Change; the momentary read comes from Fogg’s behaviour model. Together they tell your system where a user is heading and whether now is the moment to act.
Prochaska’s 5 stages · the trajectory
Pre-contemplation
Unaware or unready: they don't yet see a reason to change.
System action: Zero pushing, no guilt: only subtle, educational content.
Contemplation
Ambivalent and easily overwhelmed: weighing it up, not committed.
System action: Reduce friction and surface the barriers, without shaming.
Preparation
Committed to act, and needing a push over the line.
System action: Offer micro-commitments: “just 10 minutes today.”
Action
Forming the habit, and at the highest risk of dropping off.
System action: Daily check-ins, immediate reinforcement, relapse prevention.
Maintenance
Routine established: the behaviour has stuck.
System action: Advanced challenges and community. Turn off the patronising reminders.
Fogg’s behaviour model · the momentary read
Behaviour = Motivation × Ability × Prompt
A behaviour only fires when motivation, ability, and a prompt converge at the same moment. The Builder maps those three variables onto each stage of change, so your team knows exactly when a prompt will land, and when it’s better to stay quiet.
Turn human behaviour into AI system logic.
Detect
Behavioural signals
Capture real-time telemetry: app-launch frequency, prompt-rejection rates, browse-vs-act duration, session skips.
Infer
State & readiness
Translate raw telemetry into human context: current stage of change, motivation score, ability barriers, relapse risk.
Adapt
System interventions
Automate the right response: UI affordances, notification frequency, tone, and prompt intensity.
Deliverables built for design and engineering alignment
Audience Factors matrix
Attribute-based mapping that replaces the static persona slide with something an adaptive system can actually read.
Detect–Infer–Adapt ruleset
Exportable logic rules your product managers and ML engineers can build against, not prose to reinterpret.
Vulnerability & edge-case guardrails
Map situational vulnerability instead of leaning on harmful demographic stereotypes.
One-click Jira & spec export
Bridge design research and model training: the map lands as tickets and specs, not a deck.
Built for teams designing adaptive experiences
Senior product designers
Design stage-specific affordances and progressive disclosure, instead of one static interface for everyone.
AI PMs & data scientists
Turn qualitative research into quantitative signals and automation rules your model can train on.
Behavioural design researchers
Validate behaviour-change products (fitness, fintech, learning, health) with scientific rigour.
Behaviour is dynamic, not demographic.
Who’s in front of your AI isn’t a fixed profile: it’s a state that shifts by the hour. Reading it in real time is one of four questions every AI team keeps asking, in no fixed order: what you anticipate, whether people trust it, and whether it meets the rules all depend on knowing who’s actually there right now. Map the state and the rest sharpens: you anticipate the right need, measure the real experience, and can show a vulnerable user was met with care.
Before you start
Do I need to throw away our existing user personas?
No. Keep them as shared vocabulary if they help, but treat them as a starting hypothesis, not the model your AI reads. Audience Factors sits underneath: it turns “who someone is” into “what state they're in, right now,” which is the part a static persona can't give an adaptive system.
How is this different from event-tracking or telemetry analytics?
Analytics tells you what happened: clicks, sessions, funnels. Audience Factors tells you what it means: which stage of change a user is in, their motivation and ability, and what the system should do about it. It's the interpretation layer on top of telemetry, not another dashboard of raw events.
How do we collect behavioural signals without over-tracking or hurting privacy?
Map from the signals you already legitimately have, and keep the model deliberately coarse: a stage and a readiness read, not a surveillance profile. The goal is to prompt less and better, so “collect everything” works against you: map the minimum signal that reliably infers state, held in the EU-hosted, GDPR-first workspace.
Can these factor maps export into product specs for data science teams?
Yes. The Detect–Infer–Adapt ruleset and the factors matrix export as structured specs and tickets, so your data-science team gets automation rules to build against rather than a slide deck to interpret.
How does this prevent annoying or patronising prompts, the “Clippy effect”?
That's the whole reason to map state. Fogg's B=MAP says a prompt only fires behaviour when motivation and ability are already present, so the model's default is often silence. It tells the system when NOT to prompt (pre-contemplation, low ability, settled maintenance) as clearly as when to, exactly what Clippy never learned.
More questions? See the full FAQ →
Ready to upgrade from static personas to AI logic?
Build your first Audience Factors map in under five minutes. One account, no credit card required.
Grounded in Prochaska & Fogg · Export to Jira & Figma · Free tier included