Fraud and error cost the government an estimated £55-81bn in 2023-24. In an environment where public finances are under increasing scrutiny, data analytics are key to tackling these losses. So why are they not already being used more widely?
In our recent report, Using data analytics to fight fraud, there are plenty of examples where public bodies are using data analytics to good effect to tackle fraud and error.
These include making better use of existing data, establishing new data-sharing and matching systems, and experimenting and rolling out emerging technologies such as AI to identify high-risk cases.
Departments like HMRC and DWP have been using data analytics for a long time. HMRC saves £3-£4bn annually by using its ‘network analytics tool’ which join up over 100 data sets to help it prioritise and support tax investigations. Network Rail reports a 15:1 return on investment (ROI) from its counter-fraud data analytics work, while the NHS Counter Fraud Authority meets its 3:1 ROI target, partly due to its machine learning pilot Project Athena.
Alongside these, other parts of government are experimenting with data analytics. DfT has piloted an AI image recognition tool to detect duplicate grant claims and HM Prison and Probation Service uses machine learning to detect internal fraud.
Here are some more useful examples:
Targeting which cases to review under Universal Credit: To reduce incorrect Universal Credit payments, DWP is using data analytics to prioritise which cases to review. The system flags risky claims which are then sent to case workers, who look for previous incorrect payments, and correct future payments. DWP also uses the insights gathered from this approach to strengthen its preventative controls.
Using data matching to verify Legal Aid payments: The Legal Aid Agency (LAA) used the Digital Economy Act 2017 to pilot data-sharing with HMRC, verifying income for around 600 suspected fraudulent claims. LAA investigated recipients who did not meet income eligibility requirements, estimating that it prevented around £500,000 of incorrect payments. Senior stakeholders backed the pilot because it saved investigators significant time trying to establish eligibility without the HMRC data-share.
Deep learning to scan photographic evidence for electric vehicle chargers: DfT piloted an AI tool using image recognition and deep learning to detect duplicate claims in a grant scheme for electric vehicle charger installations. It identified and recovered small amounts of fraudulently obtained funding and blocked further dealings with those vendors. The tool is now being used in other grant schemes, and DfT is exploring live data integration to prevent fraud.
These case studies show that there are already great practices to learn from. But they also highlight a troubling reality, namely that the savings achieved so far have been modest compared to the amount potentially achievable. Why?
We have set out ten challenges that public bodies face when implementing data analytics to tackle fraud and error. Notably, we found a lack of central planning to help public bodies to adopt new technologies. Officials also told us they find it difficult to make the case for investment and the need to build greater public trust.
It is worth reading the full list of challenges and the recommendations we made to the Public Sector Fraud Authority (PSFA) and Government Digital Servies. But here are five key takeaways that those in public bodies should consider when implementing data analytics in their counter-fraud efforts.
1. The ultimate aim should be prevention rather than detection
Preventative controls are often more effective than detective controls. They can be harder to implement and often require real-time data sharing but, crucially, they can stop incorrect payments before they are made. This prevents money being lost to fraud and error while also saving the time, money and effort spent recovering amounts that should never have been paid out. Nonetheless, detective controls remain important and can be used for root cause analyses to strengthen preventative measures.
2. Before developing your own, consider using centrally provided tools
Cabinet Office offers several centrally provided data analytics tools, but these are not widely used. Developing and implementing a data analytics tool can seem challenging, but departments do not always need to start from scratch.
Central tools that are available include: The National Fraud Initiative, which uses data-matching to detect inconsistencies that may indicate fraud; SNAP, a network analytics tool that links data sets to uncover hidden risks; Payee Verification, which checks whether the details associated with a bank account match those being provided; and Spotlight, which checks grant and contract eligibility for companies and charities by automatically performing due diligence checks.
3. Cross-government data can improve the way you verify the eligibility of participants in your programmes
Key data sets, such as HMRC’s employment income data or Companies House data on UK-registered companies, can help tackle fraud and error. Officials told us that while they understood sharing data to be crucial to tackling fraud and error, they found it difficult and bureaucratic. PSFA can advise public bodies on how best to follow and take advantage of the data-sharing process under the Digital Economy Act.
4. Develop a multidisciplinary team to get the most out of data analytics
The most effective tools we saw were developed by dedicated, multidisciplinary teams combining digital skills and fraud subject-matter expertise. Departments investing in data analytics should prioritise building cross-functional teams to deliver more effective and scalable solutions. This includes building counter-fraud data analytics into new digital programmes, and pulling together counter-fraud and digital expertise to develop new analytical solutions.
5. To make a difference to the bottom line, invest in people who can investigate data analytics results, and not just in digital tools
Tools that flag payments for review require people to investigate further, otherwise the data analytics will be a wasted investment. In our experience, departments have not always maximised their ROI due to an under-resourcing of qualified staff. Proper levels of staffing, coupled with tools being reviewed iteratively to ensure that algorithms are tweaked and AI keeps learning, can produce the maximum return in terms of detected and prevented fraud and error.
If you are looking for further inspiration, read our full report and the 20 case studies contained within.
Joshua Reddaway is director of fraud and propriety at the National Audit Office