What makes the next check-in worth returning for?
I used return patterns to identify weak moments in a grocery app’s check-in experience, inform the redesign, and evaluate continuation.
- My role
- Diagnosis & redesign evaluation
- Original tools
- SQL · Excel · BI
- Deliverable
- Product guidance & test readout
The problem
Participation weakened after the first check-in and around the next cycle. Which moments needed a clearer reason to return?
What I did
- Separated new and existing check-in users and identified their different return patterns.
- Connected those moments to explicit next-day rewards and reward redemption, then reviewed the variant comparison.
Make the next reward clear and continued participation useful.
- Clarify the next-day benefit.
Make the reward for returning tomorrow explicit in the check-in experience.
- Support continued use.
Provide a destination where accumulated rewards can be redeemed.
- Evaluate each audience.
Review new and existing participants separately, including the transition to the next cycle.
The historical work documents the redesign and a subsequent A/B readout. The outcome measured here is check-in continuation.
The redesigned experience held more new-user check-in activity.
Day-two participation was 15.0% versus 9.0%. At the next cycle, it was 6.0% versus 1.9%. The size of the difference varied by follow-up day.
Follow the full new-user return pattern
View data & definitions
New-to-check-in participants. Day 8 represents the start of the next seven-day cycle in this reconstruction. The original deck calls the evaluation A/B, but allocation details, original sample sizes and uncertainty are unavailable. No significance or randomized causal-effect claim is made.
| Follow-up day | Original check-ins | Redesigned check-ins | Original rate | Redesigned rate | Difference (pp) |
|---|---|---|---|---|---|
| 2 | 1,350 | 2,250 | 9.0% | 15.0% | 6.0 |
| 3 | 645 | 1,425 | 4.3% | 9.5% | 5.2 |
| 4 | 495 | 1,170 | 3.3% | 7.8% | 4.5 |
| 5 | 420 | 1,050 | 2.8% | 7.0% | 4.2 |
| 6 | 360 | 990 | 2.4% | 6.6% | 4.2 |
| 7 | 330 | 960 | 2.2% | 6.4% | 4.2 |
| 8 | 285 | 900 | 1.9% | 6.0% | 4.1 |
Decision: Preserve the early return benefit and continue testing how reminders reach new participants.
Later continuation mattered as much as the first return.
Existing users started slightly lower in the redesigned variant—74.0% versus 76.0%—but held 70.4% into the next cycle, compared with 57.0% in the original.
Read existing-user continuation on its own scale
View data & definitions
A separate existing-check-in population with complete day-8 follow-up. Every initial participant stays in the denominator. These are participation rates within the check-in product, not whole-platform retention, DAU or revenue effects.
| Follow-up day | Original check-ins | Redesigned check-ins | Original rate | Redesigned rate | Difference (pp) |
|---|---|---|---|---|---|
| 2 | 7,600 | 7,400 | 76.0% | 74.0% | -2.0 |
| 3 | 7,200 | 7,320 | 72.0% | 73.2% | 1.2 |
| 4 | 6,800 | 7,260 | 68.0% | 72.6% | 4.6 |
| 5 | 6,500 | 7,200 | 65.0% | 72.0% | 7.0 |
| 6 | 6,200 | 7,150 | 62.0% | 71.5% | 9.5 |
| 7 | 6,000 | 7,100 | 60.0% | 71.0% | 11.0 |
| 8 | 5,700 | 7,040 | 57.0% | 70.4% | 13.4 |
Decision: Review cycle transitions and existing-user engagement before expanding a check-in strategy.
Use the return pattern to guide the experience and its evaluation.
What I delivered: Audience-specific diagnosis and evaluation supporting clearer rewards and a redesigned check-in experience. The work review records implementation and variant results, with reminder improvements as a further opportunity.
Participation definitions, test limits & source notes
Based on my 2019 JD Daojia Q2/Q3 reports and operations-sharing material. Original analysis used SQL, Excel, and BI; Python/vector charts are portfolio reconstructions.
- New means new to the check-in product, not necessarily newly registered on the grocery platform. Existing means a previous check-in participant.
- Daily check-ins use all initial users in each variant as denominator. These points are not conditional daily hazards or uninterrupted check-in streaks.
- The example uses a seven-day cycle with day 8 as the next cycle’s start. Reconstructed cohort counts and dates do not assert the original experiment used those exact sizes or schedule.
- The source’s A/B label does not establish random assignment, sample balance, statistical significance or a platform-wide retention effect. Differences vary across follow-up days.
Original decks, product screenshots, user records, reward systems and SQL remain private.
Download continuation counts