The UK's Office for National Statistics (ONS) is integrating artificial intelligence to reduce costs and enhance data quality, addressing tasks such as repairing crucial economic statistics and preparing for the next population census under fiscal constraints. The agency indicates that new AI tools are projected to save thousands of hours annually, enabling a reallocation of resources toward core economic data production while urgently improving efficiency.

James Benford, the ONS's director of economic statistics, stated that significant quality issues remain unresolved, requiring the agency to implement new national accounts standards, advance the next census, refine the labour force survey, and modernise the statistical business register. He noted that applying AI in statistical production not only boosts output efficiency but also improves data quality, with the agency currently in a transitional phase.

Recent deep-seated data problems have led to frequent errors in employment figures and other key economic indicators, forcing the statistics body to scale back work in areas such as health and crime. This month, the agency warned that efforts to replace the flawed labour market survey are increasingly diverting investment away from broader improvements to the statistical system. Although the ONS receives an additional £100 million annually to prepare for the 2031 census, its core funding is set to be cut by nearly 10% in real terms over the next two years.

Benford emphasised that AI is primarily a productivity initiative, with the goal of achieving more with the existing workforce without reducing headcount, while reinvesting efficiency gains back into the statistical infrastructure. Last year, the ONS became the first statistical agency globally to use AI in producing official data, introducing a tool based on Google's enterprise-grade large language model to classify industries and occupations from survey responses.

Andrew Banks, the chief data scientist, reported that the tool is estimated to save 350 hours annually across two surveys, while improving industry classification accuracy from 71% to 80%. Early next year, the agency plans to deploy AI scanners to automatically process household receipts collected in the living costs and food surveys. Banks estimates this tool could compress per-household receipt review time from three hours to seconds, saving roughly 7,500 hours annually, and by eliminating manual data entry, researchers could significantly expand survey sample sizes.

AI is also expected to enhance the new labour force survey by dynamically adding occupation-related questions for respondents to improve data accuracy. A pilot study involving 1,000 people showed that 43% of respondents who received follow-up questions had more precise occupation classifications, with no impact on questionnaire completion rates. Benford acknowledged that the rollout of the new labour force survey faces challenges due to its digital-first approach, which typically requires in-person interviews to determine industry codes, and the agency is attempting to replicate this process with AI.

Benford also revealed that AI is being widely deployed across the ONS workforce. More than 5,000 of the agency's 5,980 employees now use standardised AI tools, while 120 data scientists use coding assistants daily. He stressed that the approach is to accomplish more with the existing team rather than downsize, reinforcing the agency's commitment to efficiency without compromising its workforce.