1. Knowledge Base
  2. Retain
  3. Game Recommendations (Casino)

Game Recommendations - How It Works

Our AI recommendation engine personalises game recommendations by analysing player behaviour, preferences and game attributes, using a set of machine learning models to deliver relevant results

How Our AI Models Drive Game Recommendations

To deliver accurate and relevant game recommendations, the engine leverages a set of AI models that analyse signals from player behaviours, player preferences, session patterns and game attributes. Each model contributes a different perspective on what makes a game likely to resonate with a player.

  • Significant Activity & Favourite Games

    • Detects significant player sessions.

    • Filters out low-engagement or promotional sessions.

    • Defines "favourite games" based on player activity, recency and other behavioural factors.

  • Collaborative Filtering

    • Utilises the game interactions of similar players.

    • Uses collective data to recommend games aligned with player preferences.

  • Game Similarity

    • Groups games based on common characteristics: such as RTP, volatility, theme, game features and paylines.

    • Suggests games similar to players' favourite or frequently played games.

  • Game Sequencing

    • Predicts future choices based on a player's sequence of played games.

    • Adjusts for lifecycle phases (e.g. new player conversions).

Key Considerations

These are key factors the system considers when generating personalised recommendations

  • Stake Amounts: The amount wagered by players in their sessions.
  • Session Frequency: Frequency of game sessions.
  • Preferred Games: Regularly chosen games.
  • Frequently Played Titles: Top game titles revisited by player.

Recommendation Process

The recommendation process balances personalised curation with fallback options to ensure a full list of relevant games.

  1. Curated Game List: By design the system typically suggests up to 20 games per player, ranked by preference, although this is configurable.

  2. Customisable Backfill: If fewer than 20 recommendations exist, the system backfills games that align with the player base preferences.

  3. Model Outputs: The recommendations can be delivered as independent feeds, e.g. via individual API requests, or blended sets, e.g. singular API request.

Updates & Frequency

Recommendations are regularly updated based on player activity and system schedules to keep suggestions fresh and relevant.

  • Frequency: Recommendations can be updated in real-time or batch, with the most common update frequency being daily.

  • Player rescoring: Occurs based on significant new player activity or after 7 days of inactivity.

  • Maintenance with lack of activity: Recommendations may remain similar if no significant new activity is detected.

  • Last updated date: Each recommendation batch includes the most recent rescored date.

For more information:

Still have questions? Check out our FAQs for further assistance.