Recommendation 24

The Department explained that, in order to manage the risk of unintended bias in the use of data analytics to identify fraud, it ensured that there was always meaningful human involvement in decision-making, and that it undertook ‘fairness analysis’ to identify any disproportionate impacts. We asked the Department which groups the fairness analysis had been conducted for. It explained it had undertaken the analysis for a number of protected characteristics but that this was “still very much in the test phase” and did not tell us which groups or the results of this analysis so far. It explained that its testing included identifying both fraudulent activities, but also any false positives where the model suggests activity might be fraudulent but on further investigation is not. It further explained that it will need to regularly review and improve its methods for as long as it uses them to ensure that they were being applied fairly and to stay ahead of fraudsters.46 Following our evidence session, the Department wrote to the Committee to explain than an Equality Analysis had been completed for the overall project which initially considered fairness metrics for age, gender and pregnancy. It told us that it continuously monitored for bias and accuracy, and will update the Equality Assessment to consider other protected characteristics following further testing.47 45 DWP ARA 2021–22, page 229 46 Qq 66–68; DWP ARA 2021–22, page 229 47 DWP Follow up Correspondence to the Public Accounts Committee Hearing on DWP Annual Report & Accounts 2021–22 on 20 July, 6 September 2022 18 The Department for Work and Pensions’ Accounts 2021–22 – Fraud and error in the benefits system 3 The systemic underpayment of State Pension The Department’s progress in correcting the systemic underpayment of State Pension