Churn Prediction Model (AI-Powered)
This page outlines how our AI-Powered Churn Prediction Model works, what are the requirements and the restrictions in order to enable it on your environment.
🖌️ Redefining Churn Prevention
Predictive models help redefine how we approach churn strategies. The Churn Prediction Model no longer focuses solely on deposits but instead evaluates bets and wins aggregated in sessions. The model assesses factors such as session frequency, game engagement, and indicators of a positive or negative experience. Every 24 hours, players receive a score that determines how likely they are to churn. The focus is no longer just on reactivating customers who are already churned or close to churning, but also on preventing these scenarios, allowing you to react swiftly and prevent churn while avoiding unnecessary and intrusive outreach.
More information on the basic Churn Prevention Model can be found here:
📩 How Does the Model Work
First, the model is trained with at least six months of historical data, allowing it to understand player behaviour and adapt to their habits.
Once activated, here’s an example of how it works:
A player frequently plays slots and makes deposits
The last session showed signs of frustration (e.g., consecutive losses, early session exits).
The model detects this negative experience and segments the player accordingly.
Based on your strategy, you can then trigger a tailored re-engagement offer—such as free spins on their favourite game or a personalised bonus—to bring them back.
📌 Definition
We define a player as churned when they have at least 30 days with no real bets or deposits. The model is applied once a day on every inactive player between 1-29 days to detect potential churn risks early and allow for timely intervention.
📈 Segmentation Based on Churn Rate and Usage
The predictive model assigns each player a Churn % Rate (e.g., 20% churn rate, 30% churn rate) based on their risk level. This segmentation enables a more targeted and structured churn prevention strategy, allowing, for example, for:
Offer Generosity Scaling (recommended use) – The churn rate can determine the generosity of incentives, where higher-risk players receive more compelling offers. This allows for dynamic campaign structures where promotions trigger at strategic intervals to encourage re-engagement at critical moments.
Tailored Churn Prevention Campaigns – Players can be placed into specific retention campaigns based on their churn probability, ensuring messaging and incentives are relevant to their risk level.
By leveraging these churn segments, operators can fine-tune their engagement strategies, optimising retention efforts while minimising unnecessary costs on incentives for lower-risk players.
Incorporating Control Groups in Churn Prevention campaigns is essential to maintaining the model's integrity, ensuring it accurately predicts each player's churn rate over time.
📊 Data Requirements
For the Churn Prediction Model to function optimally, the following data points should be available:
Casino events – Tracking overall player engagement and success in gaming sessions.
Payment events – Monitoring deposits and withdrawals to understand financial behaviour.
Mandatory fields for the model to work
amount
exchange_rate
payment_id
status
timestamp
user_id
type
external_user_id
Blocked events – Identifying account restrictions that might contribute to churn. (Not mandatory, but recommended)
Mandatory fields for the model to work with Blocked events
timestamp
blocked
excluded
user_id
deposit_limit
valid_until
Historical data – To ensure the model is reliable and produces accurate results, the following data thresholds must be met:
A minimum of 6 months of player activity
At least 65,000 player sessions* in total (used for training the model)
➡️ As a general rule of thumb, if you have 3,000 unique active players (UAPs) per month, you’re likely to meet the required number of sessions.
📌 Why does this matter? The model needs enough behavioural history to recognise patterns that indicate churn. Insufficient data will limit its ability to identify trends and reduce accuracy.
*What is a Player Session?
A player session is defined as a player’s game activity (e.g. betting on a game), separated by 30 minutes of no activity.
If a player is inactive for 30 minutes, we consider the session closed.
The first bet placed after that period starts a new session.
What if you don’t meet the UAP threshold?
Please note: the 65,000 player sessions are the true requirement for the model to operate effectively.
Even if you have fewer than 3,000 UAPs per month, your model may still work — provided you’re generating enough player sessions overall.
On the other hand, if you have 3,000+ UAPs but don’t hit the session count, the model will not have sufficient data to perform accurately.
🚧 Restrictions
To ensure the accuracy and effectiveness of the model, the following restrictions apply:
Casino data only – The model currently applies exclusively to casino gaming data.
Minimum player activity – A player must have at least 4 gaming sessions recorded.
Minimum account duration – The player must have been registered for at least 30 days before they can be assessed.
For further information on the data points highlighted above, kindly refer to our Integration Knowledge Base.
🏁 Getting Started
To run the requirement checks and enable the model in your Fast Track environment, go to FT Singularity Model → Player Features. Under Available Player Features, select Churn Prediction Model.
Click Run Requirements Check to start an automated validation of your data for the model. If any issues are found, a summary of errors will be displayed.
Once your data is validated, the system will generate a new Player Feature called Churn Prediction Model. This feature comes with pre-built logic and will automatically assign each player to a classification.
Once the process is complete, a new Segment field will be available in the Player Features section. This allows you to accurately Segment players based on their likelihood of churn. The Segment options are listed below.
Available Segmentation
Active
Player was active in the last 24 hours
Very Low Risk
Probability of churn between 0% and 20%
Low Risk
Probability of churn between 20% and 30%
Medium Risk
Probability of churn between 30% and 40%
Medium High Risk
Probability of churn between 40% and 50%
High Risk
Probability of churn between 50% and 60%
Very High Risk
Probability of churn between 60% and 70%
Critical
Probability of churn between 70% and 80%
Very Critical
Probability of churn between 80% and 100%
Inactive
Player has reached 30 days inactivity
Initialised
When a player is eligible to the player feature but has not been evaluated yet.
Last updated
Was this helpful?