The UK Department for Work and Pensions (DWP) is using artificial intelligence (AI), predictive analytics and new working methods to improve public services, reduce administrative pressure and help staff make better decisions. However, a new report from the National Audit Office (NAO) says the department needs a stronger culture of innovation to make these improvements more consistent across its operations.
The report, DWP’s Approach to Innovation, recognises that several initiatives are already delivering measurable benefits for customers and employees. These include AI systems that help identify child maintenance cases at risk of falling into arrears, tools that flag potentially vulnerable customers, and technology that reduces the time staff spend preparing training materials.
However, the NAO found that the department could do more to explain acceptable levels of risk, encourage staff to test new ideas safely and share lessons from projects that succeed or fail.
The findings come as DWP faces the challenge of modernising services used by around 23 million people while managing substantial public spending and growing demands on its workforce. According to the figures provided in the report, the department spent £309.5 billion on welfare payments and £10.6 billion on running costs and service delivery in 2025-26.
The review highlights an important distinction: introducing new technology is only one part of innovation. Public organisations also need clear safeguards, effective management and a way to ensure that successful approaches benefit teams across the organisation.
Results
The NAO examined six examples of innovation at DWP, covering artificial intelligence, benefits administration and employment support. The projects illustrate how targeted changes can improve administrative processes and help staff respond to customers more effectively.
One of the clearest examples involves a machine learning model developed by the Child Maintenance Service. The system identifies cases that may be at risk of payment breakdown, helping caseworkers recognise situations where early intervention could prevent arrears.
DWP estimates that the model has prevented an additional 715 cases a month from moving into arrears. It has also saved staff time equivalent to 49 full-time employees.
These figures suggest that predictive analytics can help public services identify potential problems before they become more difficult to resolve. Rather than relying entirely on manual reviews, caseworkers can use risk indicators to prioritise cases that may need attention.
However, the system supports caseworkers rather than removing the need for human judgement. Its value depends on how accurately it identifies risk and how effectively staff use the information to determine appropriate next steps.
AItools
Another initiative is the Whitemail Insights and Vulnerability Scanner, which processes approximately 25,000 documents each day.
The tool helps identify correspondence that may indicate a customer is vulnerable. This can enable staff to review cases more quickly and determine whether additional support or a different approach is appropriate.
Handling large volumes of correspondence can place considerable pressure on public service teams. Technology that helps identify relevant information may reduce the time required to locate important details and allow employees to concentrate on decisions that require individual assessment.
DWP has also developed a generative AI tool known as GAIL, designed to reduce the time staff spend creating training materials.
Preparing training content can involve substantial administrative work, particularly in a large organisation where employees need regular updates on procedures and policy changes. Generative AI may help speed up drafting and content preparation, although appropriate review remains important to ensure that materials are accurate and consistent with official guidance.
These examples demonstrate different uses of AI within government. Predictive analytics helps identify potential risks, document processing supports the handling of customer correspondence, and generative AI assists employees with routine content creation.
The potential benefits are practical rather than purely technological: less time spent on repetitive tasks, more focused casework and better use of staff resources.
Employment
DWP’s innovation programme also extends beyond artificial intelligence.
One initiative involves assigning Jobcentre work coaches to particular economic sectors. The aim is to give Universal Credit claimants more tailored employment support by connecting advice and guidance with the needs of specific industries.
A work coach with a better understanding of a sector may be better positioned to discuss relevant vacancies, employer expectations and the skills required for available roles.
This approach could help make employment support more closely aligned with local and sector-specific labour market conditions. Its effectiveness, however, depends on the availability of suitable jobs, the needs of individual claimants and the quality of relationships with employers.
Other projects include text message pilots encouraging Carer’s Allowance claimants to report changes in their circumstances. Timely reporting can help reduce the risk of overpayments by ensuring that relevant information reaches the department sooner.
DWP is also looking into how the benefits system could be redesigned from the ground up. Such work offers an opportunity to examine whether existing processes can be simplified, made more efficient or organised around the needs of people using the service.
Taken together, these initiatives show that innovation can involve changes to staff responsibilities, communication methods and service design, as well as the introduction of new technology.
Risks
Despite the benefits identified, the NAO stressed that DWP faces particular challenges when testing new approaches.
The department provides essential services to millions of people, many of whom depend on timely benefit payments to meet basic living costs. Disruption to a critical system can therefore have serious consequences for individuals and households.
DWP must also protect sensitive personal information and ensure that its services comply with relevant legislation. Any use of AI or predictive analytics must be considered alongside questions of privacy, security, accuracy and accountability.
These requirements mean that innovation cannot be treated in the same way as experimentation in a business where service disruption has fewer immediate consequences.
The NAO recognised that DWP has developed an approach that differentiates between levels of risk. Different projects require different degrees of oversight, and testing and refinement can help identify problems before an innovation is introduced more widely.
However, the watchdog found that these practices need to be applied more consistently.
Its recommendation is for clearer, more practical guidance that helps staff and managers understand when they can experiment, which risks are acceptable and what safeguards must remain in place.
Such guidance could help teams distinguish between changes that can be tested relatively easily and those that require more extensive checks before implementation.
Learning
The NAO also identified weaknesses in how DWP captures and shares lessons from innovation projects.
Individual teams may develop useful solutions, but those lessons do not always reach colleagues working elsewhere in the department. This can limit the wider value of successful initiatives and create a risk that different teams spend time solving similar problems independently.
The watchdog said DWP should improve how it records and communicates the results of its innovation work, including projects that are discontinued or fail to achieve their original objectives.
Learning from unsuccessful projects is important because it can help other teams identify practical limitations, avoid repeating mistakes and make better decisions about future investment.
For a department operating at such a large scale, sharing knowledge consistently could help promising ideas move beyond individual teams and become part of wider service improvements.
It could also help managers make more informed decisions about which initiatives should be expanded, modified or stopped.
A stronger approach to sharing lessons would therefore support both efficiency and accountability. Public money spent on developing new approaches is more likely to deliver lasting value when the department understands what worked, what did not and why.
Culture
The report’s broader message is that effective innovation requires more than the right technology. Staff need the confidence to identify problems, suggest improvements and test ideas within clearly defined boundaries.
If guidance is unclear, employees may be reluctant to experiment because they are unsure whether a proposed change will be considered acceptable. Alternatively, inconsistent oversight may allow projects to progress without sufficient attention to potential risks.
A more consistent innovation culture could help address both concerns. Clear responsibilities, proportionate safeguards and accessible guidance would give teams a better understanding of how to develop ideas without compromising essential services.
It would also help ensure that innovation remains focused on outcomes for customers and taxpayers rather than on adopting technology for its own sake.
For AI projects in particular, measurable improvements in speed or staff capacity need to be considered alongside accuracy, fairness, privacy and the quality of decisions made using the technology.
Outlook
The NAO’s findings present a mixed but constructive assessment of DWP’s approach. The department has demonstrated that predictive analytics, document-processing tools and changes to service delivery can produce measurable operational benefits. At the same time, the report indicates that these successes need to be supported by more consistent guidance, stronger knowledge-sharing and a clearer framework for safe experimentation.
NAO head Gareth Davies said innovation could help DWP improve services for customers and protect taxpayers’ money. He added that the department should build on its existing foundations by giving staff the confidence to test ideas safely and ensuring that good practice and lessons are shared consistently.
The challenge now is to turn individual successes into a more systematic approach across the department. That means evaluating projects carefully, learning from unsuccessful trials and ensuring that improvements can be adopted without compromising service reliability or customer protections.
For a department responsible for major welfare programmes and employment support, the long-term test will be whether innovation produces dependable improvements for the people who rely on its services. AI may help employees work more efficiently, but clear accountability, effective safeguards and an organisation-wide willingness to learn will remain essential to achieving that goal.
