Client project for Entain, Mar – Aug 2025
Early VIP & High-Potential VIP Player 360
An ML and graph analytics platform that spots future VIP players within their first 3 to 21 days, with a plain-language reason for every prediction.
Overview
An end-to-end Player 360 platform for Entain that detects Early VIP and High-Potential VIP players across 3, 7, 14 and 21-day behavioural windows. It joins transactional, behavioural, product-mix and graph data in BigQuery, scores players with XGBoost and graph neural networks, and explains each classification in plain language.
The problem
CRM teams want to reach players who are likely to become high-value early, while there is still time to act. That means predicting VIP potential from only a few days of behaviour, with heavy class imbalance, many related entities per player (devices, IP addresses, affiliates, deals), and a need to explain every prediction to the people who act on it.
What I did
- Built the platform end to end, from BigQuery pipelines and feature engineering to model training, inference and the explanation layer.
- Designed the Player 360 pipelines linking players to IP Address, Device, Session, Affiliate, Tracker, Country and Deal entities.
- Developed and optimised the XGBoost, homogeneous GNN and heterogeneous GNN models.
How it works
- Behavioural, monetary, temporal, engagement and cross-product features covering deposits, stakes, active days, sessions, NGR and product mix.
- Graph features for connectivity, neighbourhood statistics, propagation signals, behavioural clustering and shared-entity patterns, with relationship filtering to cut graph noise.
- Feature selection, class-imbalance handling, threshold optimisation and PR-AUC-focused evaluation for early VIP prediction.
- SHAP-driven feature selection to keep high-value predictors and drop redundant, low-value ones.
- Leakage-safe, time-aware training and inference with frozen feature schemas, reproducible preprocessing artifacts, model checkpoints and training/inference consistency checks.
- A Gemini layer that turns the top SHAP features from each inference into a readable explanation of why a player was flagged.
Outcome
- Early VIP and High-Potential VIP predictions across four behavioural windows, supporting proactive CRM targeting.
- A human-readable explanation alongside every classification, generated from SHAP values by Gemini.