---
title: 'UK Department for Work and Pensions: ML risk referral with human fraud decision'
slug: dwp-uc-advances-fraud
stable_id: 957e48fc908e26a9
company: UK Department for Work and Pensions
function_code: insurance
pattern_codes:
  - threshold_as_control
  - continuous_decisioning
evidence_strength: verified
publication_tier: showcase
freshness: current
reviewed_at: '2026-08-23'
updated_at: '2026-08-22'
source_quality_summary: 1 independent, 1 primary; publication outcomes are verified.
caveat_summary: Department/NAO estimates; subgroup disparities require monitoring.
collections:
  - regulated-autonomy
bundle_version: 1.0.0
bundle_fingerprint: sha256:c23c6cc2b88153d008ea8fda928f632ce0011fc2d4c5036672a16e5d895bab93
canonical_url: https://brianletort.ai/transformations/dwp-uc-advances-fraud
---

# UK Department for Work and Pensions: ML risk referral with human fraud decision

Fraud review can focus on model-high-risk claims if adverse decisions remain human.

Function: Insurance. Patterns: Threshold as the human control; Continuous decisioning. Evidence: verified.

Freshness: current. Reviewed: 2026-08-23. Updated: 2026-08-22.


Source quality: 1 independent, 1 primary; publication outcomes are verified.

## Before

1. **Universal Credit advance process** — Receives a new claimant request and applies existing eligibility and fraud controls. (control: The fairness report does not document a dated pre-model operating queue; random sampling is the evaluation benchmark, not claimed as the full former workflow.)
2. **Human decision maker** — Reviews evidence and alone determines whether an advance is fraudulent. (control: Human adverse-action authority.)

## After

1. **Advances machine-learning model** — Scores new advance requests and refers predicted high-risk cases. (control: Model is a prevention control; it does not make the fraud decision.)
2. **Human decision maker and assurance sample** — Reviews referred cases; equal high/low-risk random samples monitor hit rate and false negatives. (control: Legitimate referrals are prioritized for timeliness; DWP conducts fairness assessment.)

## Decision rights

A human confirms fraud; DWP owns thresholds and fairness review.

## Exception path

Legitimate referred cases are prioritized; random sampling checks misses.

## Outcomes

- **Fraud identification** (verified): Random sample → 2.9 times as effective; estimated £4.4m saved. Department/NAO estimates; subgroup disparities require monitoring.

## Executive lesson

A defensible model narrows review; it does not own denial.

## Anti-pattern

Do not automate adverse action from risk alone.

## Questions for leaders

- Where is the operating threshold set and who can override it?
- What measured result would trigger rollback or retraining?
- Which residual decisions must remain human-owned?

## Sources

- [Universal Credit Advances Model Fairness Assessment](https://www.gov.uk/government/publications/fairness-assessment-including-statistical-analysis-of-the-universal-credit-advances-machine-learning-model-1-april-2024-to-31-march-2025/universal-credit-advances-model-fairness-assessment) — UK DWP
- [DWP begins to make headway tackling fraud and error](https://www.nao.org.uk/press-releases/dwp-begins-to-make-headway-tackling-benefit-fraud-and-error/) — UK NAO