DoorDash built a voice AI agent for Dasher support calls with answers in under 2.5 seconds

DoorDash gets hundreds of thousands of support calls a day. With AWS, it built a voice self-service agent on Amazon Connect, Bedrock, and Claude 3 Haiku that answers routine Dasher questions.

Organization: DoorDash
Industry: Delivery
When: 2024
Evidence: Vendor case study
Agents and tools: Claude 3 Haiku Amazon Bedrock Amazon Connect Amazon Lex

Reported outcome

A voice agent ready for live testing in about two months, answering in 2.5 seconds or less, with 50x more testing capacity.

≤ 2.5 sec
Response latency
50x
Testing capacity
~2 months
Build time to live test
100Ks per day (all calls)
Call volume

Caveats

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

Published by AWS, which sold the platform.
The case study gives speed and testing numbers, not resolution rate or customer satisfaction.

What the agent does

When a Dasher (delivery driver) calls support, a voice agent answers. It handles routine topics such as app problems, sign-ups, and payments, using a knowledge base. It hands harder issues to people.

Why speed matters

On the phone, a long pause feels broken. DoorDash needed answers in 2.5 seconds or less, so it chose a small, fast model. It also built automated tests that run thousands of test calls an hour.

What to copy

For voice, choose the fastest model that is accurate enough.
Build automated conversation tests before launch, not after.
Start with the most common, simple call reasons.

Starter prompt

Written by us, not by DoorDash.

You are a phone support agent for delivery drivers. Speak in short
sentences (max 2 per turn). Use only the knowledge base results below.

If the driver mentions safety, an accident, or harassment, say:
"I'm connecting you with a specialist now." and transfer.

Knowledge base: {retrieved}
Driver said: {transcript_turn}