Duty of care, data science, and gambling harm: A scoping review of risk assessment models

Abstract

Aims

Duty of care policies mandate gambling operators to identify problematic gambling behaviours amongst their customers. Online operators often employ risk detection algorithms to accomplish this task. This scoping review focuses on how such data science applications can perform from a duty of care perspective.

Methods

In line with the PRISMA guidelines for scoping reviews, we systematically retrieved academic studies, reports, and industry initiatives that used statistical methodologies to predict, model, or forecast gambling behaviour. The final sample consists of 31 academic studies published between 2015 and 2025, and 11 commercial solutions. Our analysis focuses on three critical stages of model development: 1) selection of estimation data; 2) decisions related to the model estimation process; and 3) assessment and interpretation of prediction model results.

Results

Models vary in terms of predictors, dependent variables, methodological approaches and assessment. Most models attempt to identify harm that has already occurred rather than forecasting future harm. Data are typically aggregated despite higher granularity in original datasets. Measures to assess the prediction ability of models are not optimal. Industry funding or involvement is prevalent in model development.

Conclusions

Currently, risk assessment algorithms do not function pre-emptively and are unlikely to capture the full extent of harm occurring in digital gambling. As such, their usability within the duty of care framework remains limited. Ways forward would entail openness and standardisation in terms of choice of variables, forecasting horizons, assessment of methods, and evaluation of results to improve models and regulatory oversight.