You win back lapsed players by contacting a ranked short list, not the whole dormant base, by matching the offer to the player, and by measuring the result against people you did not contact. Most winback budget is lost in the first step: the same message and bonus go to everyone who went quiet, including players who would have returned alone and players who were never going to. Here is how I would set it up, and where I would be careful.
What is the difference between reactivation and winback?
The words are used loosely, so I will fix them. Reactivation is getting a dormant player active again, often before they have fully gone. Winback is bringing back a player who has already left. Both start from a quiet account, but the chance of success and the cost are different.
The earlier you act, the cheaper the attempt. A player who has been quiet for a short stretch often needs a small nudge or nothing at all. A player who left long ago needs a stronger reason and a message that does not feel like a mass mailing. For definitions of the neighbouring terms, see what player churn means.
Who is actually worth winning back?
Not everyone. The first step is to rank the dormant base by two things: how likely the player is to come back, and what they were worth. A player who deposited every week for a year and then went quiet is a different case from a one-time depositor who took a bonus and disappeared.
A scoring model can give you the first number. It estimates, for each player who has already gone quiet, the probability of returning. That is a separate question from the probability that an active player will leave, and it should be a separate model. The second number is the value, and I would show it as a range, because a single figure on a 90-day horizon pretends to a precision nobody has. I explain why in what predicted lifetime value means.
The output you want is a list sorted by expected value, with the players most likely to return and worth the effort at the top. A manager should be able to work down from the first row and stop when the effort no longer pays.
Which offer should a lapsed player get?
The one from your own promotion catalogue that fits the player's stage and value. A cashback offer, a reload and a spins bundle appeal to different people. A single offer for the whole dormant base is the most expensive way to run winback, because part of it goes to players who would have come back anyway.
In Sellrise the idea is the same one we use for active players: a promotion matched to the player's stage and value, with the reason shown, so a manager can check it before sending. I would also keep the option to send nothing. If the reason a player went quiet is that they have just had a win and are enjoying it elsewhere, a bonus is a cost, not a hook.
Decide the economics of the offer yourself. Wagering terms, margins and legal limits belong to you. A forecast can say who and when. It cannot know your bonus rules.
Which channel and what timing?
Use the cheapest channel that can reach the player, and move to a more expensive one only when the cheaper one has not worked. Email and push cost little. SMS, Viber, WhatsApp and Telegram cost more per contact, and a call costs the most, so a call should go to the players whose value justifies it.
In Sellrise the channels are SMS, Viber, WhatsApp, email, Telegram and calls, with push and in-app going through your platform. Paid channels are billed at cost. I discuss the choice in more detail in how to choose messaging channels for casino retention.
On timing, use the player's own rhythm. If someone usually plays on Friday evening, a message on Tuesday morning is noise. A fixed wait of the same number of days for every player is the simplest rule and the most wasteful one.
Who should you not contact at all?
Two groups. First, players who are self-excluded or flagged for responsible gaming. They can appear in analysis, but they must never reach the outreach queue, and the check should run both when you score and when you send, because a status can change in between. I wrote more on this in responsible gaming and retention.
Second, players without consent for the channel you want to use. A strong offer is not a reason to skip this check.
A third group is softer: players whose expected value is below the cost of the contact. It is fine to leave them alone.
How do you know winback worked?
Hold back a control group of lapsed players that nobody contacts. Compare how many returned and how much they played in the group you contacted with the group you left alone. The difference is the effect of the campaign. Without a control group you will count players who would have returned anyway as your success.
This is the same test we run in a Sellrise pilot: a control group that nobody contacts, your team working as it does today, and your team following the suggestions, for one month, with average turnover per player as the metric and group sizes agreed beforehand. We do not quote a percentage before the test, because any figure would come from somebody else's players. The method is explained in how to run a control group test.
What mistakes do teams make most often?
- Treating every dormant player as equally valuable.
- Using one offer for the whole list.
- Measuring clicks and opens instead of returns and turnover.
- Contacting players without checking responsible gaming status and consent at send time.
- Waiting until a player has been quiet for months, when the cheap window has gone.
None of these needs a model to fix. A ranked list, a short catalogue of offers and a control group get you most of the way.
Where does Sellrise fit?
Sellrise gives your team the ranked list and a suggested action for each player, and your team decides and sends. Nothing goes out on its own. If you want to see how it looks, the demo runs on synthetic data, and you can start a pilot to measure it on your own players. The use case page for player retention has the short version.