At a glance
- Source
- Brynjolfsson, Li and Raymond (2025). Generative AI at Work. The Quarterly Journal of Economics, 140(2), 889–942. doi:10.1093/qje/qjae044
- How it was done
- The company gave the assistant to different groups of agents at different times. The researchers compared agents who already had it with agents who did not yet. Field study. It follows real work inside a company and compares staff who had the tool with staff who did not yet have it.
- Who took part
- 5,172 customer-support agents.
- Where
- Customer support for a company studied by researchers in the United States; some agents worked from other countries.
“Less experienced and lower-skilled workers improve both the speed and quality of their output, while the most experienced and highest-skilled workers see small gains in speed and small declines in quality.”
What it found
- Agents with the assistant resolved 15% more issues per hour on average.
- Less experienced and lower-skilled agents got faster and their work got better.
- The most experienced agents got slightly faster, and the quality of their work fell slightly.
- The assistant appeared to help agents learn. It also improved the English of agents working from other countries.
- Customers were more polite and asked for a manager less often.
An earlier working-paper version of the same study, covering 5,179 agents, reported a 14% average gain and a 34% gain for novice and low-skilled agents (NBER Working Paper 31161). The figures on this page come from the published journal version.
When the result holds
The work was customer support. It is frequent, well defined and easy to measure. Gains were largest on problems that were moderately rare: uncommon enough that agents had little practice, common enough that the AI had seen examples.
Limits
- Agents were not assigned to the tool by chance. The comparison relies on the order in which the company rolled it out.
- The study measures speed and service quality. It does not measure profit.
Practices that use this study
This study supports:
- Measure time saved per task against a baseline Record how long a task takes and how good the result is before you add AI. Without that starting figure you cannot tell a real gain from a hopeful one. Do first · Effort: low (days, one person)
- Start with one costly process, not a company-wide rollout Pick one process that visibly costs time or money. Apply AI there, with a person checking the output, and expand only when the numbers move. Do first · Effort: medium (weeks, a few people)
- Give AI assistance to newer staff first When you roll out an AI assistant for a well-defined task, start with your less experienced people. Studies find they gain the most. Do next · Effort: low (days, one person)
This study limits or contradicts: