How much more do customer support agents get done with AI?
In the largest real-world study, chat agents at a Fortune 500 software firm resolved 14 to 15 percent more issues per hour with an AI assistant, and the least skilled gained 30 to 36 percent. A larger randomized trial at Alibaba found much smaller speed gains, with chats 1 to 8 percent faster.
Across all 2 studies: 1 to 15 percent, counting the Alibaba trial.
- Low confidence
- 2 primary sources
- 2025 to 2026
- Last checked
Every study on one scale
Chart values as text
- Scale: % faster or more issues resolved. Shaded range: 14–15%; dashed, both studies: 1.1% to 15.2%.
- Brynjolfsson, Li and Raymond (2025): 13.8–15.2% more resolved
- Ni et al., Alibaba (2026): 1.1–8.2% faster
Why low confidence: The 14 to 15 percent comes from a single study at a single firm, in two versions of the same paper. The only other large field experiment, at Alibaba, measures different things and finds much smaller speed gains.
Sources
| Study | Creator | Year | Sample | Method | Figure | Link |
|---|---|---|---|---|---|---|
| Generative AI at Work (Quarterly Journal of Economics) | Erik Brynjolfsson (Stanford), Danielle Li and Lindsey Raymond (MIT) | 2025 | 5,172 chat support agents at one Fortune 500 software firm, mostly in the Philippines, 3 million chats | Staggered rollout of a GPT-3-based reply assistant, 2020 to 2021, compared before and after with the firm's own records. The 2023 NBER version gives 13.8%. | 15.2% more issues resolved per hour; handle time down 8.5%; customer satisfaction unchanged “increases worker productivity, as measured by issues resolved per hour, by 15% on average” | doi.org ↗ |
| Generative AI in Action: Field Experimental Evidence from Alibaba's Customer Service Operations | Xiao Ni, Yiwei Wang, Tianjun Feng, Lauren Xiaoyuan Lu, Yitong Wang and Congyi Zhou (Fudan, Zhejiang University, Dartmouth Tuck, Alibaba Group) | 2026 | 5,940 new agents with under a year of tenure, about 2.56 million chats on Taobao, China | Four-week randomized trial run by Alibaba, January to February 2024. The assistant drafts a diagnosis and solution at the start of each chat. One author works at Alibaba. | Issues identified 8.2% faster and chats 1.1% shorter; dissatisfaction down 3.4% “reduced issue identification time and chat duration by 8.2% and 1.1%, respectively” | arxiv.org ↗ |
Which support agents gain the most from AI?
The weakest. At the software firm, less skilled and less experienced agents gained 30% and the lowest-skilled fifth 36%, while the most skilled saw no significant change and their quality slipped slightly. At Alibaba, customer ratings for the lowest-performing fifth of agents rose 0.81 points on a 5-point scale and fell 0.28 points for the top fifth. F-12 compares this pattern across studies.
Numbers you may have seen
These figures circulate widely. Here is what the original source says.
- AI makes support agents 14% more productive, and novices 34% more. Those are the 2023 working-paper figures. The peer-reviewed 2025 version says 15% on average, 30% for less skilled and less experienced agents, and 36% for the lowest-skilled fifth.
- Klarna's AI assistant does the work of 700 agents. Klarna's 2024 press release describes a chatbot answering chats on its own. It gives no method and says nothing about how productive human agents are with AI.
More from What AI Saves
- F-083–5.4%How much time does AI save knowledge workers?
- F-0925–40%How much faster do people finish writing tasks with AI, and is the work better?
- F-1121–26%How much faster do developers code with AI assistants?
- F-1227–46%Does AI help less-experienced workers more than experienced ones?
- F-1314–25%When does AI make work slower or worse?
The whole chapter · Chapter poster · All of The AI at Work Fact Book