Using Machine Learning to optimize mobile game experiences

Halfbrick and Google used a custom on-device machine learning model in Jetpack Joyride to personalize the digital rewards offered to players for watching rewarded video ads, increasing interactions with the ad unit by 36% in an initial experiment. Rather than relying on random selection or hard-coded rules, the model evaluates gameplay signals such as skill level, recent session details, cause of a player’s last loss, and selected power-ups to recommend a reward suited to each player. Halfbrick collected 129 signals through Google Analytics for Firebase, prepared the data in BigQuery, trained and refined a TensorFlow-based model, and deployed it through Firebase ML, with Remote Config directing devices to current model versions. The model runs locally through TensorFlow Lite, avoiding cloud inference requests and associated per-inference costs, while a portion of sessions continues to receive random rewards to generate fresh training data. A companion codelab demonstrates a simplified version of the workflow with six input features for developers seeking to implement similar personalization.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.