Every AI company promises hours back. The best evidence says: sometimes, on the right task, and less than you feel.
Where it helped
Harvard Business School and Boston Consulting Group gave 758 consultants GPT-4. On tasks the AI handled well they finished 12.2% more work, 25.1% faster and to a higher standard. At a call centre, 5,179 support agents solved 14% more problems an hour with an AI helper, and newer staff 34% more.
Where it hurt
On a task outside what the AI did well, the same consultants were 19 percentage points less likely to get the right answer. And in a 2025 trial by the research group METR, experienced developers working on their own projects took 19% longer with AI, while believing it had sped them up by about a fifth.
Feeling faster is not the same
In the UK government's trial of Microsoft 365 Copilot, 20,000 civil servants reported saving 26 minutes a day on average, and 17% saw no clear saving. A large Danish study matched AI use to payroll records and found no change in earnings or hours two years after ChatGPT arrived.
What to do with this
Time one repeat job without AI and with it, counting the checking. Keep AI where it clearly wins, drop it where it does not, and be wary of polished output that someone else has to fix. Read company figures as a best case.