Lyft cut average support resolution time by 87% with a Claude-based assistant

Lyft built a customer care assistant on Claude through Amazon Bedrock. It tries to solve rider and driver issues first, and hands hard cases to people.

Organization: Lyft
Industry: Ride-hailing
When: Feb 2025
Evidence: Vendor case study
Agents and tools: Claude Amazon Bedrock

Reported outcome

Average resolution time fell by more than 87%, with thousands of requests resolved per day (company claim).

-87%
Resolution time
+30% (vendor)
Decision accuracy

Caveats

Published by the company that sells the agent. Treat numbers as marketing until confirmed elsewhere.

Numbers come from Lyft and Anthropic, not an independent audit.
An issue counts as 'resolved' when the customer answers yes to 'Did we resolve your issue?'. This is self-reported.
Safety, deactivation, and fraud issues still go to human agents.

What the agent does

The assistant talks to riders and drivers in the Lyft app. It answers common questions and solves simple issues. It passes sensitive cases to a human agent.

Why this works

Before the change, people reportedly waited 30 to 40 minutes to reach an agent. Most of that wait was for simple issues. The assistant answers those at once, so humans can focus on hard cases.

What to copy

Write a clear list of issue types the agent must never handle alone (safety, fraud, account bans).
Define "resolved" up front, and know its limits.
Measure time-to-resolution, not only containment.

Starter prompt

Written by us, not by Lyft.

You help {riders/drivers} with support issues.

You may resolve: {list of allowed intents}.
You must hand off to a human immediately for: safety incidents,
account deactivation, suspected fraud, injury, or legal threats.
When you hand off, write a 3-line summary for the human agent:
issue, what you checked, what the customer wants.

Use only the trip and account data provided. Never guess amounts.