The Stanford Study: How AI Helped Resolve 15% More Issues
AI assistance increased issues resolved per hour by 14–15% on average, with up to 36% gains for less experienced workers, while improving customer satisfaction and retention.
– Brynjolfsson, Li, and Raymond (2023)One of the most common questions business owners ask is: "Will AI actually help my customer-facing teams, or will it just frustrate our clients?"
To answer this, researchers from Stanford University and MIT conducted a massive field study at a Fortune 500 enterprise software company. They tracked over 5,000 customer support agents, some of whom were given access to a generative AI assistant that provided real-time suggestions for responding to customers. The findings provide a blueprint for how businesses should think about AI augmentation.
The 15% Boost in Output
The headline finding from the Stanford study was clear: agents using the AI tool saw a 14% to 15% increase in issues resolved per hour. They handled more chats, spent less time searching for answers, and successfully closed more tickets.
But more importantly, they didn't sacrifice quality for speed. The researchers found that AI assistance actually improved customer satisfaction metrics and reduced the likelihood of customers requesting to speak to a manager. The AI wasn't replacing the human; it was acting as a highly competent sidekick, instantly surfacing technical documentation and drafting polite, accurate responses that the agent could review and send.
The Stanford Study Impact
The True Value: Capturing Institutional Knowledge
The most profound insight from the study mirrors findings from MIT Sloan: the benefits of AI were not distributed equally. The AI tool had minimal impact on the company's most experienced, highest-performing agents. They already knew the answers.
However, for novice and low-skilled workers, the AI drove a 34% to 36% improvement. The researchers noted that the AI system effectively captured the tacit knowledge of the best workers and distributed it to newer employees in real-time. It allowed junior staff to perform like veterans within weeks instead of months.
Furthermore, the study found that employee retention improved among the newer workers. Because they felt supported and capable, they were less likely to quit due to frustration.
Applying This to Your Operations
You don't need to be a Fortune 500 company to leverage these findings. The principle is: use AI to structure unstructured knowledge.
- Knowledge Retrieval: Instead of making staff dig through shared drives, an AI agent can instantly retrieve policy details or technical specs.
- Email & Document Drafting: Drafting standard responses or extracting data from forms (like our Federal Form Automation tools) saves hours of manual typing.
Augmentation vs. Automation
A critical takeaway from the Stanford study is the approach to implementation. The company didn't build an AI chatbot to replace its human agents and force customers to talk to a robot. Instead, they built an AI to empower their human agents. This "human-in-the-loop" augmentation is consistently proven to be the most effective, least risky way to adopt AI.
Build Your AI Infrastructure
At Dakota AI, we build the technical pipelines that turn academic theory into operational reality. From custom data extraction to automated workflows, we help businesses implement AI correctly.
About the Author
Neil Olson is the founder of Dakota AI. Holding an MS in Data Science, he specializes in translating cutting-edge AI research into practical, automated workflows for businesses. From custom computer vision to multi-agent document processing, he builds the systems that drive real productivity gains.