Signals: learning from friction without surveillance
Signal acquisition, done safely
Many teams want to measure stress or confidence, but most approaches drift into invasive monitoring. A safer approach is to observe system-level signals and invite user-controlled feedback.
The aim is not to guess emotion. The aim is to find where people get stuck, rush, abandon, or loop, then redesign to reduce pressure and improve recovery.
Core principles
- Measure the interaction, not the person. Prefer loops, retries, delays, and help-seeking signals.
- Keep signals explainable. Every metric should map to a design action.
- Separate quality signals from security signals. Do not treat calm design telemetry as fraud scoring.
- Make consent visible. Offer opt-in feedback and make data minimisation explicit.
Minimum viable telemetry
If you want a simple starting set, use these events with short retention and ephemeral session IDs.
| Signal | What it looks like | What it may indicate | Best response |
|---|---|---|---|
| Hesitation | Long pause before continuing | Uncertainty or fear of consequences | Add consequence preview, add “review” step, reduce ambiguity |
| Backtracking | Repeated toggles, going back and forth | Confusing wording or missing context | Improve labels, add examples, show “why we ask” |
| Error loop | Same error repeated 2+ times | Instruction mismatch or hidden requirement | Show corrective guidance, offer alternate path, preserve progress |
| Help seeking | Help opened repeatedly on a step | Local clarity gap | Move help closer, shorten explanations, show one example |
| Abandonment | Drop-off during a high-stakes step | Pressure spike or perceived risk | Add “save and return”, reduce urgency, clarify outcomes |
| Recovery success | User completes after a hint | Guidance works | Standardise that hint pattern and reuse it elsewhere |
Consentful feedback patterns
Analytics shows where friction happens. Consentful feedback explains why it happens, without surveillance.
- Ask “Was this step clear?” at natural stopping points.
- Offer one click answers like “Clear” and “Unsure”.
- Make prompts dismissible with no penalty.
- Explain collection in one sentence, and keep retention short.
Implementation checklist (drift prevention)
- Retention: keep quality telemetry short (for example 14 to 30 days).
- Sampling: measure enough to learn, not enough to profile.
- No content capture: do not store typed input, photos, clipboard content, or free-text logs.
- No emotion inference: avoid “emotion scores” and mental state predictions.
- Transparency: document your signals and the design actions they trigger.
If you want a ready-to-ship blueprint, read the Note: Signal acquisition without surveillance.