You spot a future VIP early by looking at how a player behaves in the first week after the first deposit, not at how much they have spent so far. Spend confirms a VIP. Behaviour in the first days suggests one. The practical goal is to hand likely VIPs to a manager while there is still time to shape the relationship, and to stop giving VIP treatment to players who will never repay it. I will explain the idea, what to look at, and where it goes wrong.
Why is the usual VIP process too late?
Most operators define VIP by totals: a deposit threshold over a month or a quarter. A VIP manager then works from last period's list. The result is that a fast-rising player is noticed after they have already moved through the standard funnel like everyone else, and sometimes after they have left.
The opposite mistake is just as common. A player who makes one large deposit and disappears gets VIP attention because of a single number. The attention costs real money and does not come back.
Budget goes to the wrong players in both directions. That is the problem early detection is meant to solve.
What does early VIP detection look at?
The first week after the first deposit. In Sellrise, the VIP forecast works from that window: it scores each new depositor on how they are playing and depositing, and flags those who look like they are heading toward high value. I will not claim a precise hit rate here, since any figure would be somebody else's data and your result depends on your market and your players.
The kinds of signals worth looking at in a first week are the ones a good manager would notice by eye, only applied to everyone:
- How soon after registration the first deposit came.
- Whether a second deposit followed quickly (see how to get a second deposit).
- Bet size and how it moves over the first sessions.
- How many separate days the player came back.
- The mix of games and whether the player explores or sticks.
I list these as directions to test, not as a recipe. Which of them matters in your market is something your own data answers, and a model trained on your history will weight them better than any rule I could write.
Why should predicted value be a range?
Because a single number over a long horizon is fiction. In Sellrise the predicted value is shown to day 90 as a range, never a single guess. A range tells your manager how sure the forecast is. A narrow range on a strong signal deserves a phone call. A wide range deserves a light touch and a second look next week.
If a vendor shows you one clean value per player with no spread, ask how they got there. You are probably being shown precision that the data does not support. More on this in what predicted lifetime value means.
How does a flagged player reach a manager?
A forecast that stays in a report does nothing. The flagged player has to land in front of a person, with the reason, quickly. In Sellrise a future VIP is handed to a manager in one click, and the row in the daily list shows the stage (for example a VIP candidate in week one), the priority, the channel, the offer and the time to contact.
A concrete example from our demo data, which is not a client: player 48392, a VIP candidate in the first week, with high churn risk, marked for a reload offer and a call at 19:00 the same day. The point of the example is the shape of the row, not the numbers.
I would add one rule on your side. Agree who owns a flagged player, so two managers do not contact the same person and nobody assumes the other one did.
How do you stop wasting VIP budget on players who will not pay back?
Treat early detection as a filter in both directions. The same forecast that finds likely VIPs also shows players whose profile does not support VIP cost. Those players do not need to be dropped, only left on the standard path.
Be careful with one trap. Large first deposits are not the same as value. Players who deposit only to extract a bonus look generous on day one and poor on day thirty. If you suspect this pattern in a traffic source, read casino bonus abuse and how to spot it before you invest manager time.
What data do you need?
For Sellrise, four tables keyed by Player ID: players, deposits, bets and sessions. Names, phones and emails are optional and only matter if messages and calls go through the system. Integration is at our cost and every client has a separate database.
In general, whoever builds this, I would want at least six months of reliable player history so the model has examples of players who did and did not grow. Remove test, service and bot accounts first, since a handful of them can distort the result. The reasoning is in how to find players about to leave, which uses the same data.
How do you check that it works?
Run the forecast in shadow mode first: your team sees the flags but does not change behaviour. After some weeks, look at whether the flagged players actually grew faster than the others. Then run a controlled test with a group nobody touches, and compare average turnover per player over a month. That is the structure of our pilot, and the group sizes and the metric are agreed before we start.
I would be sceptical of any vendor that quotes a VIP lift before they have seen your data. I include us in that.
Where can you see it?
The use case page for early VIP detection has the short version, the demo shows the daily list on synthetic data, and you can start a pilot to test it on your own players.