Player Data Science: Value, Harm, Fraud and Experimentation

The models an online gambling operator runs on its customers, and how to build them so they can be defended. The course covers the data an operator holds and the feature layer built on it, lifetime value and churn models and the uplift experiments that make interventions pay, responsible gambling risk models from markers of harm through the label problem to the decision layer regulators inspect, fraud and integrity models against an adversary (bonus abuse, multi-accounting, payment fraud, laundering, suspicious betting) and the rules, supervised, anomaly and graph methods that stack against them, experimentation under extreme variance and regulatory constraint, and model governance: inventory, documentation, validation, explainability, fairness, automated-decision rights and monitoring.

Written for data scientists, analysts and engineers working in or joining a gambling operator, and for the compliance and product people who own the decisions those models drive. It assumes working knowledge of statistics and machine learning.

The through-line: these models make regulated decisions about real people, so the harm model constrains the commercial models and not the reverse, and every decision must be explainable to the person it affects.