Turning a tracker into a habit
New owners were active for a few days, then drifted. We built AI-generated goals, streaks and timely nudges into the Sportstech app, and defined retention carefully enough that I could trust it when it moved.
At a glance
Users were very active in the first days after purchase, then dropped off between weeks two and four.
They didn't lack intent, they lacked a target. Without one, the app was a tracking tool, not a motivator.
AI-generated 7-day, 14-day and monthly goals, streaks on the home screen, and nudges before someone drifts away.
30-day user retention +32% and workout completion +28% within six weeks of launch.
Our best-selling machines had a week-two problem.
Walking pads and treadmills were Sportstech's best sellers and the devices people used most through the app. The retention data had a clear shape: very active for the first few days after purchase, then a significant drop.
My KRA was 30-day retention. The question wasn't how to get people to try the app. It was how to still have them in week four.
People bought with good intentions. The app gave them nothing to aim at.
Momentum faded in week two
A new machine carries the first few days. After that, nothing gave people a reason to train today.
No clear target
Users could see what they'd done, but not what “good” looked like this week.
A tracker, not a motivator
The app recorded activity faithfully. It never asked for any.
Intent wasn't the problem. Direction was.
“They didn't lack intent. They lacked a target.”
When we spoke to users, most had bought the machine meaning to use it. After a week or two they lost momentum, because nothing told them what to do next.
That reframed the brief. Instead of just tracking activity, help users build a habit.
Goals the AI sets for you, streaks you can see, and a nudge before you drift.
AI-generated goals
Personalised 7-day, 14-day and monthly movement goals, built from onboarding info and early activity.
Streaks & progress, up front
On the home screen, so users see how they're doing against their goal every time they open the app.
Nudges at the right moment
With our GTM integration, falling behind or going quiet triggers a reminder before the user drops off.
- OnboardingGoal: lose weight, move daily
- DeviceWalking pad
- Week one4 walks · avg 28 min
- AI goal5 walks this week
Rule — the tick on the ring is where you should be by today. Fall behind it and a nudge goes out before the streak breaks.
I defined retention so it couldn't flatter me.
“Retention went up 32%” means nothing until you say which cohort, which window and why. Before we argued about whether it moved, I wanted the team to agree on what it meant.
- Definition
- The resume reports a 32% improvement in 30-day user retention. Reported absolute rates: 28% before, 37% after (about 32% relative lift).
- Cohort
- The resume does not specify the cohort denominator. The index chart communicates the relative lift only.
- Window
- The reported result was observed within six weeks of launch.
- Honesty check
- Illustrative definitions: the demo below compares day-30, bracket and rolling retention; it does not establish which definition produced the reported 32% lift.
Why not “active on day 30”?Nobody works out on a fixed calendar day. If I measured day 30 exactly, I'd mostly be measuring which weekday day 30 landed on.
5/ 10 retained
Tap a row to see why that user counts, or doesn't.
Why 30 days, not 7 or 90
It maps to money
The subscription billed monthly. Zero workouts in a month meant no renewal, almost a straight line. D30 was our closest leading indicator of revenue.
It's where the drop was
People fell off between weeks two and four. D7 would have shown almost nothing.
It's iterable
D90 is the truer test of a habit, and we tracked it. But one learning loop on D90 is a quarter.
The number we steered by, cohort after cohort.
The slower check that the habit actually stuck.
Six weeks is a long time to wait for one number
The first full D30 result arrived about six weeks after launch: the first week's users needed their full 30 days, plus a few days of measurement lag. That's one complete cohort, not a settled result. Until then I watched three leading indicators. They moved first, which is what gave us confidence early.
- Week 0
Launch
Goals, streaks and nudges go live for new purchase cohorts.
- Week 1
% setting a goal in week one
Are people accepting an AI goal at all?
Moved first - Weeks 1–2
Streak continuation rate
Once started, do streaks survive the second week?
Moved first - Week 2
Week-two retention
The earliest point the old drop used to show.
Moved first - ≈ Week 6
First full D30 cohort
Launch + 30 days + lag. Then cohort after cohort.
+32% retention
Retention moved, and the honesty check moved with it.
Read the chart - 28% before and 37% after are the reported absolute rates, an increase of 9 percentage points and approximately 32% relative lift. The interactive definitions demo is illustrative.
It's easy to win retention by annoying people. So I named the counter-metrics first.
Notification opt-outs & uninstalls
The nudges were the risky part. Irritate people into turning notifications off and you lose every future re-engagement channel, while retention looks fine on paper.
Support tickets & store reviews
If we were nagging, it would show up in reviews before it showed up in any dashboard.
Goal completion rate
Goals that are too hard make people fail and feel worse. If completion fell too low, the goals were wrong, not the users.
A tracker tells people what they did; a goal tells them what to do next. And a retention number is only as good as its definition, so I write it down before I look.