GovTech Benefits Fraud Detection Platform — UK Local Authority

DWP data match

National Fraud Initiative (NFI) — CIPFA-administered data matching exercise (biannual).

HMRC PAYE data match

employed claimants with employment income not declared.

Companies House data match

claimants who are company directors (undeclared income).

Local authority own data

council tax records (number of occupants cross-referenced against single person discount and occupant declarations).

ML Anomaly Detection

Gradient boosting model (XGBoost, Python scikit-learn): trained on 3 years of historical fraud investigations.

Features

claim age, benefit amount vs area median, address change frequency, employment history gaps, number of address co-habitants (council tax cross-reference).

Anomaly score

• 0–100 risk score per claim.
• High-risk threshold: claims scoring &gt
• 75 sent to investigator queue.
• Model recalibrated quarterly.

ICO Guidance on Data Matching in the Public Sector

• local authorities have explicit statutory powers for benefit fraud detection (Social Security Administration Act 1992, Housing Act 1996).
• Article 6(1)(e) public task basis.

Proportionality

data matching only for Housing Benefit fraud detection — no broader profiling.

Transparency

• privacy notice updated to describe data matching activities.
• Automated decision-making: ML anomaly score is investigator support only — no automated sanction without human investigation.
• CIPFA Counter-

CIPFA Fraud and Corruption Tracker

platform generates data for annual counter-fraud return.

Investigation case management

evidence collection, witness statements, prosecution referral tracking.

Overpayment recovery

automated calculation of overpayment amount and penalty for prosecuted cases.

Sanction tracking

caution, prosecution, administrative penalty — CIPFA reporting categories.