Call center ai · Production

DoorDash builds a generative AI voice self-service contact center with Amazon Bedrock and Anthropic's Claude

The problem

DoorDash handles hundreds of thousands of support calls per day from Consumers, Merchants, and Dashers, but most calls were still being redirected to live agents despite an existing IVR. Dashers' preference for phone support while driving made response latency a critical constraint, and the team needed a scalable self-service solution that could resolve common inquiries quickly without sacrificing quality.

Workflow diagram · grounded in source
1
Dasher calls support
Trigger
Dashers call into support rather than chatting while on the road.
▾ source quote
“Dashers generally prefer calling into support rather than chatting while they're on the road driving to or from Merchant or Consumer locations.”
2
IVR self-service experience
Ai action
Users are guided through a self-service interactive voice response (IVR) experience powered by Connect Customer and Amazon Lex.
▾ source quote
“users are guided through a self-service interactive voice response (IVR) experience, powered by Connect Customer and Amazon Lex”
3
RAG knowledge retrieval
Integration
RAG fetches data from DoorDash's publicly available help center and enriches the prompt to provide more relevant and accurate responses for Dashers.
▾ source quote
“DoorDash added data from its publicly available help center to use retrieval-augmented generation (RAG), a technique that fetches data from company sources and enriches the prompt to provide more relevant and accurate responses for Dashers.”
4
Claude generates response
Ai action
Anthropic's Claude models in Amazon Bedrock generate voice responses, with capability to mitigate hallucinations, prompt injection events, and detect abusive language.
▾ source quote
“Claude was instrumental to the project because it has the capability to mitigate hallucinations, prompt injection events, and detect abusive language.”
5
Low-latency voice response delivered
Output
The solution delivers voice responses with a latency of 2.5 seconds or less.
▾ source quote
“achieving a response latency of 2.5 seconds or less”
6
Complex issues routed to live agents
Routing
Addressing routine questions with generative AI frees up live agents to solve higher-complexity issues that benefit from human troubleshooting.
▾ source quote
“Addressing routine questions with generative AI has also freed up DoorDash's live agents to solve higher-complexity issues that benefit from human troubleshooting.”
7
Automated test and evaluation
Feedback loop
A SageMaker-based framework completes thousands of automated tests per hour and semantically evaluates responses against ground-truth data.
▾ source quote
“complete thousands of automated tests per hour—a 50x increase in capacity—and semantically evaluates responses against ground-truth data.”
Reported outcome

DoorDash completed rollout in early 2024 of a generative AI voice self-service solution—built in only 2 months—that now handles hundreds of thousands of Dasher support calls per day, driving large and material reductions in call volumes and reducing escalations to live agents by thousands per day.
The prior IVR had already achieved a 49 percent reduction in agent transfers and $3M in YoY operational cost savings; the new Bedrock solution reduced AI development time by 50 percent and achieved a response latency of 2.5 seconds or less.

Reported metrics
agent transfers reduction (existing IVR)49 percent
first contact resolution increase (existing IVR)12 percent
YoY operational cost savings (existing IVR)$3M
Time to build and test solution2 months
Show all 10 reported metrics
agent transfers reduction (existing IVR)49 percent
first contact resolution increase (existing IVR)12 percent
YoY operational cost savings (existing IVR)$3M
time to build and test solution2 months
generative AI development time reduction50 percent
voice response latency2.5 seconds or less
automated testing capacity increase50x increase in capacity
daily Dasher support calls handledhundreds of thousands
escalations to live agents reductionreduced by thousands per day
Dasher support call volume reductionlarge and material reductions
Reported stack
Amazon BedrockAmazon Connect CustomerAnthropic's ClaudeClaude 3 HaikuAmazon LexKnowledge Bases for Amazon BedrockAmazon SageMakerRAG
◆ Does this fit your context?

Compare to your context

Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.

Compare to your context →
~30 seconds · free
Source
https://aws.amazon.com/solutions/case-studies/doordash-bedrock-case-study?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

DoorDash completed rollout in early 2024 of a generative AI voice self-service solution—built in only 2 months—that now handles hundreds of thousands of Dasher support calls per day, driving large and material reducti…

What tools did this team use?

Amazon Bedrock, Amazon Connect Customer, Anthropic's Claude, Claude 3 Haiku, Amazon Lex, Knowledge Bases for Amazon Bedrock, Amazon SageMaker, RAG.

What results were reported?

agent transfers reduction (existing IVR): 49 percent; first contact resolution increase (existing IVR): 12 percent; YoY operational cost savings (existing IVR): $3M; Time to build and test solution: 2 months (source-reported, not independently verified).

How is this call center ai AI workflow structured?

Dasher calls support → IVR self-service experience → RAG knowledge retrieval → Claude generates response → Low-latency voice response delivered → Complex issues routed to live agents → Automated test and evaluation.

Related call center ai cases

WHAT TO DO WITH THIS

Now compare it to your context

This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.

Browse the Call center ai cases →