- Benefit overpayments remain too high, but levels are reducing after a substantial rise during the COVID-19 pandemic.
- The Department for Work & Pensions (DWP) is making headway, saving an estimated £4.5 billion from April 2022 to March 2025 through its counter-fraud interventions.
- To make further progress, the NAO recommends that DWP should finalise its approach to implementing its fraud and error strategy and should progress its ambition to reduce the overpayment rate to the pre-pandemic level.
A new NAO (National Audit Office) report finds that the proportion of benefit expenditure overpaid remains too high,1 but the figures for 2024-25 suggest that overpayment levels are now reducing due to DWP’s recent interventions.
The estimated proportion of benefit expenditure overpaid fell from 3.6% (£9.7 billion) in 2023-24 to 3.3% (£9.5 billion) in 2024-25, while the estimated Universal Credit overpayment rate dropped significantly from 12.4% in 2023-24 to 9.7% in 2024-25.2
The government has given DWP £6.7 billion of dedicated funding for fraud and error activity over the nine years from 2020-21 to 2028-29, enabling the department to increase the scale and impact of its approach.
Since April 2022, DWP has mainly used the funding to scale up its programme of Targeted Case Review of Universal Credit claims, increase its counter-fraud staff and expand its use of data analytics to tackle fraud and error.
Its new strategy incorporates a greater focus on prevention alongside its ongoing detection activity. The report finds that stopping overpayments before they occur is the best way to secure value for money in this area.
As part of its efforts to reduce fraud and error, DWP is working to expand its innovative use of data analytics. For instance, since May 2022, DWP has used a machine learning model to flag potentially fraudulent claims for Universal Credit advances, saving an estimated £4.4 million.
However, concerns have been raised by the Public Accounts Committee about the potential impact of machine learning on vulnerable claimants. In July 2025, DWP published, for the first time, detailed information on its fairness analysis.3
This showed that older claimants (in age groups 45 to 54 and above) and non-UK nationals were being over referred for review, with these groups being more likely to be asked to find and provide additional evidence for their claim.
In terms of performance, DWP found the machine learning model to be around three times more effective at identifying fraud risk than a randomised control group sample. As such, the department concluded that it remains reasonable and proportionate to continue operating the Universal Credit advances model as a fraud prevention control, but it would continue to seek to improve its effectiveness.
DWP also successfully scaled up its Targeted Case Review (TCR) programme to detect and correct fraud and error in existing Universal Credit claims, with around 6,000 staff (in-house and outsourced) carrying out reviews by March 2025.4
A total of 1.15 million claims have been reviewed, generating estimated total savings of £581 million from TCR, by March 2025.5
This exceeded DWP’s savings expectation by 11% although it did not meet its expectation for the proportion of reviewed cases found to be incorrect.
The total that DWP expects to save from TCR has increased significantly over time – from an initial target of £2 billion in savings by 2026-27, to £13.6 billion by March 2030.
The NAO recommends that the DWP should finalise its approach to implementing its fraud and error strategy and should progress its ambition to reduce the overpayment rate to the pre-pandemic level. Beyond this, DWP should focus on getting the overpayment rate down to a level that represents a cost-effective control environment.
The report also recommends that DWP should build on its existing use of data analytics to explore how these emerging technologies may help to detect and prevent fraud and error.
"The Department of Work & Pensions has made real progress in tackling the levels of benefit overpayments due to fraud and error, but there is still a way to go.
"With the increase in funding and the greater focus on prevention, the next few years will be key to its success in addressing this long-standing issue.
"The government should carefully consider the challenges and the recommendations outlined in today’s report if it is to build on its progress so far."
Gareth Davies, head of the NAO
Read the full report
Tackling benefit overpayments due to fraud and error
Notes for editors
1. The Comptroller and Auditor General qualified his opinion on the regularity of DWP’s 2024-25 accounts due to the material level of fraud and error in benefit expenditure (except for State Pension, for which the level of fraud and error was significantly lower). This was the 37th year in which DWP’s accounts had been qualified due to material fraud and error.
2. DWP estimates the monetary value of fraud and error in the benefit system annually through direct measurement of five or six benefits each year using a statistical sampling exercise. For those benefits not covered, it typically rolls forward the rate from when the benefit was last tested or uses a similar benefit as a proxy.
3. DWP’s fairness assessment for 2024-25, including its statistical analysis of the Universal Credit advances machine learning model, can be found online at: Fairness assessment including statistical analysis of the Universal Credit advances machine learning model: 1 April 2024 to 31 March 2025 – GOV.UK
4. The TCR programme began in February 2022 with seven agents and by April 2024 involved 3,100 DWP staff. In 2023-24, DWP decided it would use a contracted-out route to scale up its TCR workforce and appointed TP (formerly Teleperformance) to provide additional capacity.
5. DWP’s estimated savings from TCR activity include the detection and recovery of historical error, as well as savings associated with the prevention of future error.