PublicTechnology looks at case studies from a recent National Audit Office investigation into the public
sector’s ongoing efforts to halt billions of pounds in losses through fraud and error
Fraud and error in the public sector diverts tens of billions of pounds a year away from service-delivery and into less-deserving hands.
The Government Digital Service projects that savings of £6bn a year could be achieved through better use of technology to crack down on the problem. However annual fraud-and-error losses in the public sector could be as high as £81bn, according to the National Audit Office.
Earlier this month, the NAO said departments and their agencies had so-far realised only “modest” savings through the use of technology to combat fraud and error.
Nevertheless, it spotlighted multiple examples of how data analytics is being employed across government to identify fraud – and prevent it from happening. PublicTechnology takes a look at what departments and agencies are doing.
Early adopters
With their responsibility for huge amounts of tax and benefits, it is perhaps no surprise that HM Revenue and Customs and the Department for Work and pensions have long used data analytics to protect taxpayers’ money.
According to the NAO’s Using Data Analytics to Tackle Fraud and Error report, HMRC’s data-matching tool for identifying tax fraud was introduced in 2010 and generates savings of £3bn-£4bn a year.
The tax-collection agency currently uses a range of in-house data-analytics platforms to help it decide where to prioritise investigations, bringing together more than 100 data sets in the hunt for discrepancies between people’s declared income and their lifestyle.
The NAO said HMRC has now adopted a “more responsive approach” to using its data analytics tools, with the creation of targeted networks of individuals and the ability to target specific fraud risks more efficiently.
The type of analytics used include data sharing, data matching, network analysis, risk scoring and data rules.
DWP has so far racked up savings in excess of £1bn since 2022 through its Targeted Case Review programme, which seeks to identify incorrect Universal Credit benefits payments through risk scoring and data-rules analytics. Case workers are assigned to claims flagged as potentially “risky”, with benefits recipients required to provide a range of evidence to demonstrate that their circumstances match those on which their benefits are based.
Just under 6,000 staff are currently working on Targeted Case Reviews and the programme is forecast to save £13.6bn by 2030.
Since 2018, DWP has been using real-time information about the income and employment status of benefit recipients that is provided by HMRC through the Verify Earnings and Pensions Service. VEPS allows a claimant’s earnings to be verified before new benefit claims are approved. It also provides alerts about changes in earnings that claimants may not have reported.
DWP estimates that between 2018-19 and 2023-24 it saved £121m in Carer’s Allowance payments alone because of data provided by VEPS. The department was allocated additional resources in the 2025 Spending Review to invest in VEPS and investigate alerts it has so far been unable to review.
Apprenticeship cheats and legal-aid fraud
Elsewhere, both the Department for Education and the Legal Aid Agency are using HMRC data to crack down on fraud with the aid of the Digital Economy Act 2017.
A pilot that began in 2020 has seen DfE working with HMRC to identify training providers who claim funding for apprentices not recorded as being in work. The project has now become “business as usual” and, since August last year, has seen more than 250,000 new apprenticeships checked. Savings of up to £1m are claimed for the project.
In 2023, the LAA began a data-sharing and matching project in conjunction with HMRC that is designed to verify whether applicants for legal aid are really eligible. The LAA used the 2017 act to set up a legal gateway check on income data for around 600 legal-aid recipients suspected of making fraudulent claims.
Recipients who did not meet eligibility requirements were investigated and the LAA estimates it has saved around £500,000 in future erroneous payments.
The LAA reported that the HMRC data-check took a matter of minutes, compared with the likelihood that investigators would spend “significant time” manually verifying income details for recipients of interest. The agency is now looking to turn the data-share with HMRC into a business-as-usual operation.
AI to the rescue?
The NAO report lists several parts of the public sector employing artificial intelligence for assistance in the war on fraud, including the NHS, HM Prison and Probation Service and the Department for Transport.
The NHS Counter Fraud Authority has invested in its general data-analytics capability so that it can review fraud and error risks across different parts of the health service. Last year it introduced the Project Athena data-science and machine-learning programme, which is designed to detect “anomalous data points” that could be prioritised for investigation. Examples of data of interest include staff working elsewhere while claiming to be sick or fake invoices submitted for goods and services not supplied.
The NAO said NHSCFA reports it is currently achieving a 3:1 return on investment, due in part to Project Athena. The range of expected savings from the project is £10m-£100m.
The Ministry of Justice data-science team has developed a machine-learning model that flags possible fraud, corruption and bribery using HMPPS staff misconduct records.
£81bn
Potential scale of fraud-and-error losses to the public sector each year, according to the NAO
15:1
Claimed return on investment of anti-fraud dashboards operated by Network Rail
11 out of 14
Proportion of case studies considered by the NAO that are designed to detect, rather than prevent, fraud
£6bn
Annual savings that could be achieved by better use of tech, according to GDS
1996
Date of establishment of the National Fraud Initiative data-sharing and -matching programme
The model, which was introduced last year, was trained on thousands of staff records – including investigations and disciplinary cases across multiple sites. It analyses free-text summaries and details of allegations to assign labels for types of fraud, corruption and bribery, bringing automation to what used to be a manual task. MoJ’s data-science team plans to roll out similar analysis across the department.
DfT has piloted the use of deep learning to identify cases of fraud related to grants for the installation of electric-vehicle charging points. Its image-recognition tool was developed after the department came to the realisation that pictures of the same charger could be fraudulently submitted as evidence to support multiple grant claims.
DfT told the NAO that it has so-far identified and recovered small amounts of fraudulently-obtained funding and has blocked further dealings with the vendors involved.
According to the department, the savings generated by the pilot were less than £100,000, however the tool is now being used in other grant schemes. Consideration is being given to ways that live data can be integrated into the tool to prevent fraud.
Counter-fraud on track
Anti-fraud dashboards developed by Network Rail have helped the infrastructure operator achieve a 15:1 return on its anti-fraud investment and savings bracketed at up to £1m.
The system, introduced last year, brings together data on money spent across the organisation and shows the level of fraud risk for each area of spending. The dashboards flag anomalies such as a fuel card being used twice in the same day.
The counter-fraud team investigates cases where there is a risk of fraud, prioritising work based on the size of the payment.
Network Rail’s dashboards were developed in response to risk assessments that identified procurement processes most susceptible to fraud. The current system depends on supplier data that is not delivered in real time, however the organisation said the tools could be optimised with automatic data transfer – but such a system would require resources to develop.
Authority SNAPs to work
The Public Sector Fraud Authority’s work to help recover Bounce Back Loan scheme losses from the Covid-19 pandemic has led to the development of an analytics platform for company data intended for cross-government use.
Its Single Network Analytics Platform, or SNAP for short, uses public and non-public government data sets to provide a clear picture of UK-registered companies.
SNAP was developed while PSFA was supporting the government with work on BBL scheme analytics, a task which began in 2021. It said that by March 2023, around £268m had been saved through the data analytics work. Savings were achieved by measures such as identifying companies being fraudulently dissolved to avoid paying back BBLs.
Data is brought into SNAP in batches and matched to data already held in the system. Users are provided with instant risk scores and information on links between companies and their directors, so they can understand potential fraud risk.
The NAO said PSFA expects to report a “continued significant impact” from SNAP.
‘Prevention is better than detection’
One theme of the NAO’s report is the extent to which initiatives to prevent fraud are preferable – and cheaper – than those that merely detect it, because processes to recover lost funds can be “costly, time-consuming and often unsuccessful”.
The public-spending watchdog pointed out that 11 of the 14 public sector data-analytics case studies included in its report were “detective” rather than “preventative”.
Of the others, two are preventative and one is a mix of preventative and detective.
One of the purely preventative measures is the Legal Aid Agency data-sharing and data-matching project discussed above, run in conjunction with HMRC. The other is an HMRC initiative that gives people the option to use open-banking data to verify their accounts and enable faster payments.
The tool checks that account details provided to HMRC for a repayment match the details held by UK banks, giving assurance that repayments are not being fraudulently redirected to the wrong bank account. People owed refunds, including from pay-as-you-earn, can consent to HMRC’s open-banking provider getting a one-off access to their account details so they can be verified.
The NAO said savings from the initiative, which has been running since 2018 were “unquantified”. But it said that the system allowed for direct electronic payments to be made to customers instead of cheques being sent out, with associated cost savings in addition to the assurance provided by the system.
HMRC reported that the use of open banking instead of cheques would lead to annual efficiency savings of £2.5m in the first year and result in taxpayers getting repayments more quickly.
The project that is a mix of detective and preventative elements is the National Fraud Initiative, which is the largest fraud and error sharing exercise across the public sector.
Established in 1996, the NFI is a long-standing data-sharing and matching service that is now overseen by the Public Sector Fraud Authority.
Local authorities are required to use the NFI and share data about housing benefits, council tax, payroll and right-to-buy. Preventative checks can also be conducted at the point of application for housing benefit or jobs.
Additionally, the NFI conducts a “national exercise” every two years that compares more than 8,000 data sets from central government, local authorities and private-sector organisations to identify data inconsistencies that may indicate fraud.
The NAO said the initiative had generated UK-wide savings of around £1.8bn since 2015.
