An internet service provider in a mid-sized town has a small customer-support team. One veteran, Sipho, has been there eight years. He knows every router quirk, every billing glitch and exactly how to calm an angry customer. New staff take months to get close to his level, and many leave before they do.
The owner wonders whether AI could help. Not to replace the team, but to give newer staff instant access to the kind of answers Sipho carries in his head: suggested replies, troubleshooting steps, the right tone for a frustrated caller.
It turns out that idea has been tested at scale. One of the most rigorous studies of AI in a real workplace found that AI assistance helped customer-support agents, and helped the least experienced ones most of all, by essentially spreading the know-how of the best performers across the whole team.
One Like insight: AI’s biggest gift to a small team may be passing on what your best people already know.
AI at the help desk
Erik Brynjolfsson of Stanford, with Danielle Li and Lindsey Raymond of MIT, studied the rollout of an AI assistant at a company providing customer support for software, covering 5,179 support agents. The AI tool, based on a large language model, suggested responses and links to relevant technical information in real time as agents chatted with customers. Agents could use or ignore the suggestions. The research, first released in 2023, was later published in the Quarterly Journal of Economics.
On average, agents with access to the AI resolved about 14% more customer issues per hour. But the gains were very uneven. The least experienced and lowest-skilled agents improved by around a third, while the most experienced and highest-skilled agents saw little improvement, and in some measures slightly lower quality.
The researchers’ explanation was that the AI had effectively learned from the conversations of the company’s best agents and was passing their approaches on to newer staff. New agents with AI assistance reached the performance level of more experienced agents much faster. Customers also responded better: they were more positive in their messages and less likely to ask to speak to a manager. And agent turnover fell.
The study involved one large company and one tool, in a setting where the AI could learn from many past conversations. Smaller businesses won’t have the same volume of data. But the core finding, that AI can help newer or less experienced staff perform closer to the level of your best people, is highly relevant to small teams.
Spreading your best people’s knowledge
Small businesses can apply the same principle without building their own AI system. The key is capturing what your most experienced people know and making it available when others need it.
Write down the expert answers. Ask your most experienced staff to record how they handle the twenty most common customer questions or problems: what they say, what they check, what they do. This is valuable on its own, AI or not.
Give the AI your knowledge. Many AI tools let you upload your own documents, such as FAQs, procedures, product guides and example replies, so that suggestions are based on your business, not generic information.
Use AI as a coach, not a replacement. Let newer staff use AI to suggest a first reply or troubleshooting steps, then adapt it. They learn faster by seeing good examples in context.
Keep experts involved. Experienced staff should review the AI’s suggestions regularly and correct anything wrong. Their judgement is what makes the system work.
Watch the customer experience. Track whether response times, resolution rates and customer feedback improve, as the study measured.
[Infographic placement] Who gains most from AI assistance? A bar chart from the study. All agents: about 14% more issues resolved per hour. Least experienced agents: around 30% or more. Most experienced agents: little change. Caption: Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 5,179 customer support agents.
Getting started
Choose one area where newer staff struggle most: answering customer questions, writing quotes, handling complaints or explaining products. Sit down with your most experienced person and capture how they do it in a short document. Then try an AI tool, with that document as its reference, to help newer staff draft their responses. After a month, compare speed, quality and confidence.
Be thoughtful about the human side. Experienced staff may worry that sharing their knowledge makes them less valuable. Recognise and reward them for it; their expertise is what makes the AI useful in the first place.
A few questions can help:
- Where do new staff take longest to become confident?
- What knowledge lives only in our most experienced people’s heads?
- Could AI, using our own documents, help newer staff with first drafts or answers?
- How will we check the quality of AI-assisted work?
Which brings us back to the internet provider and its veteran, Sipho. Capturing his eight years of know-how in a guide, and giving new staff an AI assistant that draws on it, wouldn’t replace him. It would extend his reach, so that customers calling on his day off still got something close to his kind of help, and new staff became confident in weeks rather than months.
Sources
- Erik Brynjolfsson, Danielle Li and Lindsey R. Raymond, “Generative AI at Work”, The Quarterly Journal of Economics, 2025 (NBER Working Paper 31161, 2023).
