DWP Unveils AI-Driven Tool to Streamline ESA Benefit Decisions

The Department for Work and Pensions (DWP) has started using an artificial intelligence tool called ‘online medical matching’ to help determine the eligibility of applicants for Employment Support Allowance (ESA). This benefit, which supports approximately 1.5 million claimants affected by health conditions in their ability to work, is now partially processed by comparing the condition entered by a claimant with a centrally maintained list, achieving an 87 per cent correct match rate. Although the system automatically registers claims based on these results, final decisions are still reviewed by human agents to ensure an additional layer of oversight.

The tool has been operational since July 2020 and, since its deployment, it has processed over 780,000 cases while saving 42,500 operational hours. However, earlier versions of the tool, which were in use until 2024, faced significant issues with only a 35 per cent accuracy rate due to an algorithm that focused too narrowly on matching spelling rather than considering context. This shortfall required human agents to intervene and correct mismatches in about 65 per cent of cases.

Concerns have been raised by experts regarding the reliance on automation in critical benefit decision-making processes. Critics argue that while speeding up operations could be beneficial in an environment where claimants struggle to reach the DWP, there is a risk of bias and unfair outcomes if these AI systems are not properly managed. Addressing these issues, the DWP has emphasized that personal details are not provided to the tool, which focuses solely on the medical conditions submitted, and that staff receive training to review and verify the tool’s outputs.

This development comes amid broader governmental discussions on the expansion of artificial intelligence within public services. While the prime minister recently highlighted plans to revolutionize services like welfare, education, and health, the DWP is the first department to publicly share details on its AI algorithm, even as it maintains an internal inventory of other tools. Despite pressures for greater transparency, the agency has defended its approach, stressing the importance of controlling how such information is disseminated to the public.