Learn · Work · 5 min
Does AI actually save you time? What the trials found
Real studies, not sales claims: where AI made people faster, where it made them slower, and how to tell which kind of task you have.
Yes, on the right tasks
In a trial with 758 Boston Consulting Group consultants, those using GPT-4 finished 12.2% more tasks, 25.1% faster and to a higher standard, on work the AI was good at. In a call centre with 5,179 support agents, AI help raised the number of problems solved per hour by 14% on average and 34% for newer staff.
No, on the wrong ones
In the same consulting trial, on a task outside what the AI did well, people using it were 19 percentage points less likely to get the right answer. In a 2025 trial, experienced software developers working on their own mature projects took 19% longer with AI, while believing it had made them faster.
What people say versus what changes
In the UK government's trial of Microsoft 365 Copilot with 20,000 civil servants, people reported saving 26 minutes a day, and 17% saw no clear saving. A large Danish study linking surveys to payroll records found no change in earnings or hours worked two years after ChatGPT arrived. Feeling faster and being paid for it are different things.
Be careful with vendor numbers
Some of the biggest figures come from companies selling the tools, or from the AI estimating its own value. They can still be useful, but read them as a best case, not your case.
Find your own number
Pick one repeat task: a weekly email, a report, a plan. Time it without AI, then with AI including the time you spend checking. Keep AI where the second number is clearly smaller, and watch for "workslop": polished output that someone else has to fix.
Try it now
Try it on a real task now: say a weekly job in one line and let Ontold draft it, then time your check.
Reviewed from: Navigating the jagged technological frontier (Harvard Business School), Generative AI at Work (NBER), Experienced open-source developer productivity (METR (arXiv)), The impact of AI on developer productivity (arXiv), Microsoft 365 Copilot experiment: cross-government findings (UK Government Digital Service), Still Waters, Rapid Currents (NBER), Estimating AI productivity gains (Anthropic), AI-generated workslop (Harvard Business Review). Written in our own words.
Drafted with AI from the linked sources. Check the links for the full detail.