Why the best players were leaving, and when they really decided.
High-value players (LTV ×23) were leaving without any warning sign. Retention actions came too late: the decision had already been made.
- The stakes
- Churn was rising among the highest-value players, with no signal before cancellation.
- What I found
- The decision to leave forms 3 weeks earlier. It shows in play frequency and in support contacts.
- The decision
- Roadmap reprioritised on expert features, plus a 3-stage retention plan.
- The impact
- Roll-out in progress; the effect on HVU churn is being measured.
Why this study was urgent.
Since early 2025, the Poker team had seen churn rising among its High-Value Users: experts, VIPs and semi-pros who play several times a week. They account for a share of revenue far beyond their numbers (LTV ×23 versus a casual player at 24 months). The problem: the team only saw players leave at the moment of cancellation. The usual levers (offers, reminders, incentives) were reaching players who had already left in their heads.
- Sponsors
- Head of Product, Head of Marketing, Poker PM
- My role
- Research lead, sole researcher, from framing to readout
- Deliverables
- Churn dashboard, priority matrix, 90-day plan
Identify when and why High-Value Users disengage, early enough to act before they churn.
- Data & customer support1 week
Behavioural cohorts, early signals, review of support tickets.
- 12 1:1 interviews2 weeks
HVU players who left in the last 90 days. Rebuilding their decision journey.
- Quantitative survey1 week
267 players surveyed to validate and size the causes identified.
- Synthesis & readout1 week
Priority matrix, recommendations, briefing of the product team.
Why this approach
- Start from the data before asking
One week on cohorts, LTV curves and support tickets to learn where and when churn happens. The interviews then start from hypotheses, not a blank page.
- Talk to players who already left
12 HVU players who churned in the last 90 days, 6 on mobile and 6 on desktop. Their decision is real and recent: we rebuild a timeline, not a stated intention.
- Quantify what the qualitative surfaced
A survey of 267 players with declining activity, split into 4 segments by size of the drop, to measure the weight of each cause.
What we needed to know.
- When does the decision to leave really take shape?
- Which signals, in the data and in support, come before it?
- What makes expert players disengage?
- What do they find at competitors that they don't find here?
What users told us.
- D-21The decision is made 3 weeks before churning
Play frequency drops by 23 % at D-21, 50 % at D-13 and 77 % at D-6. Support contacts are multiplied by 4.5 between D-14 and D-7: by D-7, the decision has been made.
We could see churn coming, but we didn't know when to look.
Play frequency of churning HVU players, as a share of their usual level, in the weeks before cancellation. - 12/12Expert features are missing
All 12 expert profiles mention at least one missing feature: in-game notes, a step-by-step hand replayer, opening several tables at once.
It's an app for beginners. I need a work tool.
- 49.8 %They already compare competitors' generosity
Almost one in two players whose activity is dropping already plays elsewhere. First criterion: promotions and bonuses (57.6 %). 76.8 % say they lose more than expected.
At the competitors, I feel recognised. Here I feel like any other player.
Every cause, checked against two sources.
| Cause | Interviews (12) | Data / survey (267) | Status |
|---|---|---|---|
| Decision made ~3 weeks before cancellation | 12/12 | Play frequency −23% from D-21 | Confirmed |
| No rakeback or volume-based rewards | 11/12 | Promotions: #1 choice criterion (57.6%) | Confirmed |
| Feeling of losing more than expected | 10/12 | 76.8% say they lose more than expected | Confirmed |
| Competitors seen as more generous | 10/12 | 49.8% already play elsewhere | Confirmed |
| Missing expert features | 12/12 | Not measured in the survey | Qual only |
What to do, in order.
- P1Reprioritise the roadmap on expert features
In-game notes, hand replayer, multi-tabling: the gaps cited by 12 experts out of 12.
- P1A risk score based on play frequency
Detect the D-21 / D-7 window instead of reacting to the cancellation.
- P2A volume-based rewards programme
Answer the first criterion players use to compare with competitors.
- P2A dedicated support journey for at-risk profiles
The peak of contacts between D-14 and D-7 is the last useful touchpoint.
The roadmap was reprioritised on expert features.
In-game notes, hand replayer and multi-tabling move to the top, within a 90-day plan: detect at-risk players with a score based on play frequency, reward volume, then measure and iterate. The first releases are under way.
What this study does not tell.
Every method has blind spots. Here are this study's, and how I accounted for them.
- Players who already left rationalise after the fact.
Their accounts were checked against play-frequency data: the drop from D-21 confirms the timeline they describe.
- Expert features rest on qualitative evidence only.
It is the only cause with no quantitative measure. It has to be validated through usage, starting with a note-taking MVP.
- The non-returning segment is too small in the survey (9.1%).
It was not interpreted on its own. The conclusions rely on the other three segments, which show the same causes.
He very quickly understood our challenges and constraints, and delivered a user journey that exceeded our expectations. The new solution is now in production and is a real success with users and our business teams.
Understand your users.
Improve what matters.
GuidesHow to find out why your users churnHow many users do you need for a user test?Quant or qual? How to combine surveys and interviews
Cédric de la Chaise · Senior User Researcher · 8 years · Betclic, Roland-Garros · X ↗