Player churn in an online casino means a player who used to deposit or bet and has stopped. The difficulty is that nobody tells you. A subscription business sees a cancellation. A casino sees silence, so you have to decide how much silence counts as gone. That one decision changes every churn number you will ever report, so it is worth making on purpose. I run Sellrise and I think about this every day, but the definition below works with or without us.
How do you define a churned player when nobody cancels?
You choose an inactivity window and write it down. A player who has not deposited or placed a bet for that many days is counted as churned. There is no correct universal number. A player who deposits every Friday and a player who deposits once a month have very different rhythms, so a single window will mislabel one of them.
A workable approach is to start with a simple rule you can explain to your team, run it for a few weeks, and then compare it with how long your players actually stay quiet before they return. The gap you see in your own data is a better guide than any figure I could give you. A 14-day gap without a deposit is a common starting point to test, not a standard.
Whatever you pick, keep it fixed when you compare months. If the window changes quietly, a drop in churn may only be a change of definition.
Who counts as an active player?
You need the same discipline on the other side of the ratio. At Sellrise an active player is a player with a deposit or at least one bet in the month. That is the definition we use for tiers and for pricing, and it is also a sensible default for churn reporting.
Pick one definition of active and use it everywhere. Teams that count logins in one report and deposits in another end up arguing about numbers instead of players. A login with no bet is a visit, not play.
What is the difference between churn, dormancy and winback?
- Churn is the event or state of having left, according to your window.
- Dormancy is the stretch of quiet before you decide the player has gone. This is where you still have a chance to act cheaply.
- Winback is bringing back a player who has already crossed the line. It is harder and usually costs more.
The useful idea is that churn is a process, not a moment. Players rarely leave in one step. They log in less, bet less, play different games, and stretch the time between deposits. By the time a report shows churn, the cheap part of the work is behind you. I wrote about how to catch the fade early in how to find players about to leave and about the expensive end in how to win back lapsed players.
How do you measure churn properly?
Measure it by cohort, not as one number for the whole base. A cohort is a group of players who share a start point, usually a registration week, so you can compare their behaviour over time. A single blended churn figure hides the fact that new players behave very differently from players who have been with you for a year.
A simple set of cuts that tells you more than a headline rate:
- By registration week, to see whether newer cohorts leave faster than older ones.
- By traffic source, because some sources bring players who never come back (more on that in how to judge an affiliate source before the hold clears).
- By value, so you can see whether you are losing many small players or a few large ones.
- By stage: players who never made a second deposit, mid-life players who faded, and VIPs who went quiet.
The last cut matters most for action. These three groups leave for different reasons and need different responses, which is why one churn rate for all of them is close to useless.
What does a churn report not tell you?
It tells you what happened. It does not tell you who is next. A report that says churn rose last month cannot say which of today's active players are cooling off. That is the gap a forecast fills: a score for every player, with a reason, before they are gone.
In Sellrise the morning list is built on exactly that idea. Each row has a reason, a priority, a channel and a time to contact. Our demo shows 86 players out of a base of 10,000, and those numbers are demo data, not a client result.
Which signals show up before a player is counted as churned?
The ones I would compute first are changes against the player's own normal, not against the average player:
- Bet size falling compared with the player's recent average.
- Shorter or fewer sessions per week.
- A change in the usual mix of games.
- A longer gap than usual since the last deposit.
These are the same four signals a model uses. You do not need a model to start watching them. A manager who notices that a regular depositor has gone quiet for longer than usual is doing churn prediction by hand, and it works for a handful of VIPs. It stops working at the scale of thousands of players, which is the point at which scoring helps.
Should you aim for zero churn?
No. Some players leave because the game was not for them, and chasing every one costs more than they would ever return. The goal is to keep the players who are worth keeping, spend less on those who would have stayed anyway, and stop paying to win back players who will not repay it.
That is why churn should be read together with value. A churned player who deposited once is not the same loss as a churned player who deposited every week. For how value is estimated, see what predicted lifetime value means.
What should you do this week?
Write down your inactivity window and your definition of an active player. Split last quarter's players into cohorts by registration week. Look at churn by stage, not as one rate. If you want the forecast version of this, you can look at the demo on synthetic data or start a pilot on your own players, with a control group so the effect is measured, not claimed. The use case page for player retention shows what the daily list looks like.
If you have fewer than six months of reliable player history, I would be cautious about any forecast, mine included. Fix the definitions first.