Concept · Design exploration · Kyool
“Still Time Today”: Explainable Show-Up Prediction
A design concept for Kyool’s planned “smart streak risk alerts.” Machine learning estimates whether a member is likely to show up today. UX turns that prediction into an explanation and a choice, so people feel helped rather than watched.
AI tells users what might happen. XAI explains why. UX makes that explanation easy to understand, trustworthy, and useful.
The Problem
In a FAB, a missed day costs more than a personal streak, because the team can’t earn its FAB Star either. Generic daily reminders quickly become noise, and nagging works against Kyool’s core principle that accountability should feel supportive, not like being caught out. The question: can we help the right person, at the right moment, in a way they understand and control?
How It Works
- Signals (the member’s own data only): usual activity time, day-of-week history, progress so far today, scheduled rest days, current streak, and time left before the day ends. Used only if the member has opted in to insights.
- Model: estimates today’s likelihood of showing up and maps it to three plain bands: on track, at risk, or likely to miss. No raw percentages are shown, because a number implies more certainty than a model has.
- Explanation (XAI): the single most important reason, in everyday words, such as “You usually walk by 7 pm. It’s 8:30 and you’re at 2,100 of 8,000 steps.” A “Why am I seeing this?” link shows the signals used.
- Action: the gentlest helpful step, chosen by the member, from a short ladder of interventions (below).
- Feedback: “Not today” and “Stop these” are respected immediately and help tune future nudges. When the model is unavailable, simple rules take over, so the experience never depends on AI.
The Intervention Ladder
Each step is less intrusive than the next one up, and the member can stop at any point.
Step 1 · At risk
A quiet in-app card
“Still time today” appears on Home with the reason. It isn’t a push notification, and it never interrupts.
Step 2 · Likely to miss
Ask before reminding
“Want a reminder at 7:30?” offers Remind me, Not today, or Stop these. No reminder is ever sent without a yes.
Step 3 · Next morning, after a miss
Explain the Lifebuoy
“Yesterday slipped. Your Lifebuoy can protect your Show Up streak. It won’t earn a FAB Star.” This is a safety net, never an excuse to skip.
Concept Mockups
Home · Morning Runners
Still time today
You usually walk by 7 pm. It’s 8:30 and you’re at 2,100 of 8,000 steps.
Start a walkLater
Why am I seeing this?
Reminder · your choice
Want a reminder at 7:30?
Based on your usual evening walks. You can change this anytime.
Remind meNot todayStop these
Next morning
Yesterday slipped. That happens.
Your Lifebuoy can protect your Show Up streak. It won’t earn a FAB Star, and you have 1 available.
Use LifebuoyKeep it for later
Concept mockups are illustrative and not screenshots of the shipped app.
Guardrails
- Private by design: predictions are visible only to the member. Teammates never see a risk level, only what Kyool already shows, which is who has shown up today.
- Opt-in: uses the same consent as Kyool’s analytics, off by default, with two equal choices and nothing preselected.
- Respectful frequency: applies the same restraint as Kyool’s notification-permission prompts (at most one ask per session, and a 14-day back-off after “Not now”), and never fires on a scheduled rest day.
- Supportive tone: no copy shames the member or suggests skipping, matching Kyool’s Lifebuoy and onboarding copy rules.
- Graceful fallback: rule-based nudges take over when the model is unavailable, the same pattern Kyool uses for its FAB impact line.
- Accessible: risk is conveyed in words, not color alone; choices are full-size buttons; explanations work with screen readers and large text.
How We Would Measure It
This is a concept, so there are no results yet. A pilot would compare an opted-in test group with a control group on:
- Show Up and FAB Star rates
- Reminder opt-in versus “Stop these” rates, a direct signal of whether nudges feel helpful
- Use of “Why am I seeing this?” and qualitative trust feedback from short interviews
- Model calibration: whether “at risk” days really are missed more often
Why It Matters
The same pattern applies to mission-critical systems: a prediction is only useful when people understand it, can act on it, and stay in control. Explainable, opt-in, and reversible AI is how intelligent features earn trust.
