Public sector trails industries in AI maturity: McKinsey

Government agencies need to redesign services and workflows to move AI projects beyond the pilot stage.

The public sector received an AI maturity score of 28 out of 100, below the global cross-sector average of 33, according to a McKinsey report.

It ranked lowest amongst the sectors assessed, behind consumer and packaged goods and life sciences, which each scored 29. Technology recorded the highest score at 44.

McKinsey said many public-sector AI projects remain stuck at the pilot stage because of fragmented data, workflow integration issues, model risks and high operating costs.

Government agencies also face rigid procurement and budgeting rules, data-sharing restrictions, slow workforce redesign and stricter requirements for explainability and human oversight.

Programmes that redesign an entire service or operational domain are more likely to progress. About 70% of domain-based programmes reach production, compared with 30% of projects focused on individual use cases.

The report identified four measures for scaling public-sector AI: setting strategies around mission and resident outcomes, redesigning workflows from end to end, building organisational capabilities around the technology and retaining human oversight for consequential decisions.

McKinsey estimated that approximately 60% of AI’s value comes from workflow redesign rather than adding the technology to existing processes.

Workforce confidence is another challenge. Only one in five public-sector employees expects AI to significantly affect their daily work, whilst 31% trust their employer to develop the technology safely. This compares with 71% across industries.

The report said agencies should invest in training, adoption and capability-building alongside technology. McKinsey’s research suggests that every $1 spent on technology may require $5 in change-management investment.

Human review should remain in place for decisions such as benefit denials, licence revocations and public-safety dispatches. Lower-risk administrative processes could have higher levels of automation.

McKinsey said agencies should measure AI performance using resident-facing outcomes, including shorter waiting times, fewer errors and faster responses, rather than the number of pilots introduced.