Lyft says Claude cut support time by over 87%, freeing agents for harder cases

Lyft says its Claude-powered support assistant has cut resolution times by over 87%, leaving human agents more room for difficult cases. In a case study published by Anthropic on October 23, the company describes what it did with that capacity: reinvest millions in staff training and support services, including dedicated help for older riders.

This is an account of an existing deployment, not a new assistant launch. In a December 1, 2025 post, Lyft had already reported an 87% reduction in average resolution time using Claude through Amazon Bedrock. That earlier account described thousands of requests handled daily, with more than half resolved in under three minutes. The latest case study adds detail about staff workloads and reinvestment.

From driver onboarding to rider disputes

Lyft traces the effort to a support crunch in 2023. Riders and drivers could wait 30 to 40 minutes to reach an agent, while some agents managed three or four customers simultaneously. Customers faced rigid workflows and copied responses. The case study describes signs of agent burnout as the team tried to close chats quickly rather than understand each person's situation.

The company tested several models for response accuracy and their ability to match Lyft's tone, according to Elyse Hovanesian, its product lead for AI in support. She says Claude performed best for Lyft's needs on both measures. The team wanted a system that sounded warm and genuine, rather than another chatbot customers would resist using. Personality was part of the selection, not just a finishing touch.

Driver support came first. Onboarding required the assistant to navigate regional requirements alongside platform-wide rules. Lyft then extended it to riders, including charge disputes and ride experiences that had previously required human investigation. The company reports decision-making accuracy improved by over 30%, but the case study does not define that measure or disclose its starting accuracy.

Sensitive cases still go to people

The assistant now handles support interactions for both riders and drivers. Lyft says it greets customers by name, investigates their situation and can resolve issues in seconds. Its role is not to finish every interaction itself: safety concerns, complex disputes and situations requiring empathy are routed to human agents. Those agents receive AI-generated conversation summaries to help them pick up the case.

Lyft also describes a change in the pace of human support. Agents now handle one customer at a time instead of several chats at once. The company says that gives staff room to understand an issue's full context, consider solutions and anticipate other needs, rather than split their attention across a queue of simultaneous conversations.

We’ve empowered our agents to focus on those more complex issues that really require human care.

Elyse Hovanesian, Product Lead for AI in Support, Lyft

Where Lyft says the savings went

Hovanesian says using Claude saved millions of dollars, which Lyft reinvested in training its customer support agents. The case study also links the freed capacity to Lyft Silver, a service for older riders that combines a simpler app experience with dedicated human support. Trained agents can provide personal onboarding and a direct phone line, rather than leaving those customers to navigate the assistant alone.

Lyft argues that routine tickets would previously have consumed the resources needed for that service. Its stated next step is to let Claude handle increasingly complex issues while people remain focused on work requiring empathy and understanding. That is the company's direction for the deployment, not a newly announced capability or a timetable for handing more categories of support to AI.