McKinsey: AI pilots alone won't fix government services

Programs that redesign entire service journeys reach production about 70% of the time, McKinsey found. Programs built around single use cases succeed only 30% of the time. The difference suggests the bottleneck is not technical but structural – agencies that treat AI as a bolt-on tool fail more often than those that rewire the process itself.

The report recommended a four-step approach: focus on mission outcomes rather than the technology itself, redesign end-to-end workflows, build operating models that put AI at the center, and keep humans in the loop for critical decisions.

McKinsey also said governments should start AI work alongside data-system improvements rather than waiting for full modernization. For every $1 spent on technology, about $5 should go toward adoption, training, and capability building to get AI into daily operations, the report added.

The findings echo a broader push in federal and state IT procurement. The U.S. Office of Management and Budget in March issued guidance requiring agencies to inventory AI use cases and designate chief AI officers. The McKinsey report suggests that structural changes, not more pilots, will determine whether those mandates produce results.