About the AI model

The harmful gambling detection model is a machine-learning model designed to assess the player's betting behavior and other supporting data to determine the level of risk associated with that player.

The Risk score is measured on a scale from 1 to 100. It indicates the severity of the player’s gambling behavior. Players with higher scores exhibit more patterns of problematic gambling.

Based on the Risk score, each player is assigned a Risk category based on the following ranges:

Category name	Risk score value range
No risk	< 50
Low risk	[50, 68)
Medium risk	[68, 95)
High risk	[95, 99)
Very high risk	[99, 100)

To further contribute to the explainability of the Risk score, the actual contribution to it is broken down into 7 different contributing factors. Below you can find some examples of metrics for each.

Contributing factors	Examples of metrics
Betting behavior	Tickets placed, turnover etc
Depositing behavior	Deposits, deposit amount etc.
Withdrawal-related behavior	Withdrawals, withdrawal amount etc.
Speed of play	Live tickets, live stake etc.
Time	Active hours
Losses	Losses, tickets lost etc.
RG activity	Limit changes

The values can be positive or negative:

• The positive values influence the model in a »positive manner«, that is, they add to the model output. In other words: »They increase the likelihood that someone is a problematic gambler.«

• The negative values influence the model in a »negative manner«, they subtract from the model output. Or: »They decrease the likelihood that someone is a problematic gambler.«

Update frequency

The risk scores are updated daily, typically by 4 AM UTC.

Risk scores are only updated for players who had at least some betting activity the previous day (i.e., placed at least one bet). Therefore, requesting a risk score for any player multiple times per day will always return the same risk score.

More details about specific metrics, how they are constructed and how the data is trained is available on request.