◾Churn & LTV prediction
Last updated
Last updated
The Churn Prediction model is a dynamic tool explicitly created for casino & sport operators to identify and categorize players based on their risk of discontinuing play.
This tool is particularly focused on predicting the likelihood of players not depositing within the next 7, 14 or 30 days. The model assigns each player a churn probability score daily. This scoring system is integral to categorizing players into risk groups, thereby providing the possibility to enhance player retention.
Churn definition
In the context of this model, a player is considered to have churned if he has not deposited within the defined by the operator number of days - 7, 14 or 30. Based on the custom needs of the clients, Smartico can train models for different no-deposit periods or different definitions (e.g., no deposits and no gameplay)
The LTV prediction model is a statistical model projecting the LTV value of the player for the next 15,30 and 60 days.
During the evaluation, the mode assigns a risk score for each player with values between 0 and 1. Where 0 is the lowest probability of churn and 1 is the highest.
All players are assigned to one of 6 churn ranks for a more straightforward interpretation.
| Rank name | Explanation |
|---|---|
Not set | Player didn't have any deposit yet |
Low | Low probability of chrun, no action required. Risk value 0-0.4 |
Medium | Medium probabilty of churn, no action required. Risk value 0.4-0.6 |
High | There is a high probability of churn; action is recommended. Risk value 0.6-0.85 |
Critical | Risk 0.85-1, action is highly recommended. |
Churned | Player is churned by definition. Didn't deposit for X days |
Invalidated | The churn rank is invalidated because of the user activity. See the "Churn rank invalidation" case below. |
You can find the current distribution of the players by rank in the "Churn risk ranks" report.
This report shows the number of players for each rank and the "Probability to churn" for this rank, calculated based on historical predictions.
For example, a probability of 94.3% for Critical rank indicates that out of 100 players predicted, 94 players have churned.
The "30-day net-deposit projections" show the monetary value of the net-deposit for the next 30 days (average, based on user's history) if these players continue to be active.
To see the "Probability of churn" metric, the model should be active for at least X-day period so the model can compare its evaluation with actual churn data and calculate this metric.
The X here depends on the operator's choice of a particular model; it could be 7, 14, or 30 days.
The churn rank of the player is updated by a special event."Core: churn prediction updated.".
This event is sent when a player migrates between the ranks, for example, from Low to High.
Using this event, the Operator can build a real-time campaign when the user enters the rank requiring retention action.
It also can be used inside the campaign to stop marketing insensitivity when the user rank is going to low-risk ranks.
Churn risk can also create a segment of users that can later be used in the Campaign, Automation rules, Missions, Mini-games, and Tournaments.
For example, you can target specific mini-games only to the users with a high probability of churn or expose them to specific missions from the Gamification area.
If a user is initially classified as "Churned" or "Critical" but later takes an action that likely changes their rank, the system will temporarily label them as "Invalidated." This new rank signals to the Operator that the user no longer belongs to the "Churned" or "Critical" categories. The rank will be reviewed and updated in the next evaluation cycle.
By default, the system is moving to an "Invalidated" state only for Churned or Critical users, but this configuration can be adjusted per the Operators' request.
Example
In a sports betting service, suppose a user named Sarah is labeled as "Churned" because she hasn't made any deposits or placed bets for several months. If Sarah then makes a substantial deposit into her betting account, this indicates a potential revival of her interest in betting. In response, the system updates her status to "Invalidated." This change serves as a notification to the service operators that Sarah's previous "Churned" status might no longer reflect her current activity level, pending a re-evaluation in the next cycle.
There are three reports available for the churn model.
Churn risk ranks
Showing the current distribution of players by risk rank and the historical size of each risk group of players.
Users vs. Ranks
Reporting shows the history of user migration between the ranks within the selected period.
Actual churn report
This report shows the actual churn rate of players who did FTD and reactivated players.
The LTV Prediction AI model projects User Lifetime Value (LTV) over the next 15, 30, and 60 days.
The LTV is defined as the “Net Deposit,” which is the sum of all deposits minus all withdrawals.
The player profile in the Smartico backoffice shows projected LTV for 15, 30, and 60 days (blue line), as well as the upper and lower confidence bounds (red and grey lines).
The mode is optimized for accuracy, ensuring that 80% of users’ actual LTV falls within the provided bounds (between lower and upper limits for each predicted time interval).
Based on the marketing strategy, the LTV projections can be used to identify users with low LTV growth expectations, enabling targeted promotional incentives to encourage retention and engagement.
Notes to consider:
For users with high initial LTV, the model often projects a decline due to a pattern of significant withdrawals that is usually visible for such users.
All LTV values are calculated and displayed in Euro currency.
