Realised productivity gain is one of the most contested measurements in AI economics because it sits at the intersection of a technical outcome (the AI works) and a management decision (the freed capacity is used for something economically valuable).
The distinction matters because the two are not the same thing and do not follow automatically from each other. An AI system that reduces document review time by 30% has produced a real technical result. Whether that translates into £X of economic value depends entirely on what happens to the 30% of capacity that was freed. If the reviewer takes on additional higher-value work, the saving is real. If the organisation reduces headcount to match, the saving is real. If the freed time is absorbed into the same work at a slower pace, the saving is theoretical.
This gap between theoretical and realised productivity gain is one of the primary reasons AI ROI cases overstate actual returns. Many business cases count time saved at the hourly cost of the people doing the work, assume that all freed time will be redeployed productively, and do not plan explicitly for how that redeployment will be managed. Finance leaders should ask for the workforce planning assumption behind any productivity claim before accepting it as economic evidence.
For the full treatment of how productivity claims should be evaluated, see What CFOs Should Ask of AI ROI Claims.