Overview:
DoorDash partnered with Amazon Web Services (AWS) to build a generative AI-enabled self-service contact center to improve support for Consumers, Merchants, and Dashers. The goal was to reduce reliance on live agents, shorten response times, and elevate the user experience—especially for Dashers who frequently use voice support while on the road.
DoorDash integrated Amazon Bedrock with Amazon Connect and Amazon Lex to create a voice-operated interactive experience powered by foundation models like Anthropic’s Claude. Retrieval-augmented generation (RAG) was used to bring relevant help content into AI responses, and Amazon SageMaker supported automated testing and evaluation at scale.
Key Features:
- Integrates generative AI models via Amazon Bedrock into interactive voice response (IVR) with Amazon Connect and Lex.
- Applies Claude models to generate accurate, real-time answers with low latency (≤2.5 seconds).
- Implements RAG using Bedrock Knowledge Bases to deliver contextually relevant responses from help content.
- Automates large-scale testing and validation of AI responses using Amazon SageMaker.
- Reduces development time for generative AI features with Bedrock’s unified access to multiple models.
- Monitors performance and minimizes hallucination or inappropriate content with model guardrails and built-in safeguards.
Results & Impact:
- Accelerated development: generative AI application development time reduced by 50% through Amazon Bedrock.
- Enhanced testing: testing capacity improved by ~50× via automated evaluation frameworks.
- Fast responses: achieved response latencies of 2.5 seconds or less for AI responses.
- High daily volume: the solution now handles hundreds of thousands of support calls per day.
- Operational efficiency: significantly fewer calls require escalation to live agents, freeing agents for complex issues.
- Improved support outcomes: Dashers can get immediate assistance for common questions, boosting efficiency and satisfaction.
AI Technology:
AI Model Types: Generative foundation models (Anthropic Claude 3 Haiku), RAG workflows
AI Purpose: Automate conversational responses, answer retrieval, reduce live agent load
Application Type: Customer support automation, voice IVR enhancement
Target Users:
- Dasher support teams
- Contact center product and operations teams
- AI/ML engineering teams implementing self-service solutions
- Customer experience and contact center strategists
Sources:
- aws.amazon.com/solutions/case-studies/doordash-bedrock-case-study