Data Analytics and Fraud Detection Technology
This domain tests how analytics is used to find fraud and how it is governed. Examiners want technique names, data sources and an understanding of model limitations, including legal constraints.
Techniques
- Rules and scoring, anomaly and outlier detection, velocity checks.
- Predictive modelling: supervised (known fraud labels) and unsupervised (clustering, anomaly detection).
- Link and graph analysis, social network analysis, entity resolution and text mining of claims notes and transcripts.
- Geospatial hotspot mapping and network-level ring detection.
Data sources
- Internal policy, claims and CRM data; CUE, MID, MIAFTR, Insurance Fraud Register; credit and corporate data; lawful open-source intelligence.
Governance
- Precision, recall and false-positive rate; alert triage and alert fatigue.
- Explainability, bias testing and fairness; UK GDPR Article 22 on automated decisions; FCA Consumer Duty and SM&CR accountability.
Worked example
Graph analysis across four motor claims shows a shared bank account, the same repairer and overlapping addresses. This supports referral as an organised ring rather than four isolated claims, and the model output is recorded with its rationale.
Traps
- Over-reliance on black-box models without explainability.
- Poor data quality producing false positives.
- Ignoring Article 22 and Consumer Duty when automating decisions.