Receives a new claimant request and applies existing eligibility and fraud controls. → Reviews evidence and alone determines whether an advance is fraudulent.
After
Scores new advance requests and refers predicted high-risk cases. → Reviews referred cases; equal high/low-risk random samples monitor hit rate and false negatives.
Human boundary
A human confirms fraud; DWP owns thresholds and fairness review.
Why it matters
Fraud review can focus on model-high-risk claims if adverse decisions remain human.
How the work changed
Before
How the work ran before the change.
Step 1 of 2
Universal Credit advance process
Receives a new claimant request and applies existing eligibility and fraud controls.
ControlThe fairness report does not document a dated pre-model operating queue; random sampling is the evaluation benchmark, not claimed as the full former workflow.
Step 2 of 2
Human decision maker
Reviews evidence and alone determines whether an advance is fraudulent.
ControlHuman adverse-action authority.
What changed
Fraud review can focus on model-high-risk claims if adverse decisions remain human.
Decision rightHuman sets the threshold; AI decides each instance
After
How the same work runs now.
Step 1 of 2
Advances machine-learning model
Scores new advance requests and refers predicted high-risk cases.
ControlModel is a prevention control; it does not make the fraud decision.
Step 2 of 2
Human decision maker and assurance sample
Reviews referred cases; equal high/low-risk random samples monitor hit rate and false negatives.
ControlLegitimate referrals are prioritized for timeliness; DWP conducts fairness assessment.
Process model built from the published workflow evidence for UK Department for Work and Pensions. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Legitimate referred cases are prioritized; random sampling checks misses.
Work removed
Lower-yield random referral as primary control
Decision authority
A human confirms fraud; DWP owns thresholds and fairness review.
Before
01
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.
02
Human decision maker
Reviews evidence and alone determines whether an advance is fraudulent.
Control: Human adverse-action authority.
After
01
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.
02
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.
Work that left the path
Lower-yield random referral as primary control
Human role before
Benefits staff applied existing eligibility and fraud controls to advance requests, and a human decision maker determined payment or fraud action; the source does not document a dated pre-model queue.
Human role after
Human agents review referrals; random low-risk samples monitor false negatives.
AI roleScores advance claims for fraud risk.
Outcomes
Fraud identification
Verified
Random sample→2.9 times as effective; estimated £4.4m saved
2024–25; savings since 2022 · 1.4 million advances worth £0.8bn in 2024–25
