Building a Generative AI Contact Center Solution for DoorDash Using Amazon Bedrock, Amazon Connect, and Anthropic’s Claude

DoorDash started in 2013 with the mission to grow and empower local economies—helping local businesses succeed, conveniently connecting local consumers to the best in their neighborhood, and providing Dashers with flexible ways to earn. By the end of 2023, its user base had grown to more than 37 million active consumers who use DoorDash each month and over 2 million monthly active Dashers.

To meet the needs of Consumers, Merchants, and Dashers, DoorDash uses Amazon Connect, an AI-powered contact center from AWS. Through Amazon Connect, DoorDash handles a collective volume of hundreds of thousands of calls per day, including from Dashers who need assistance with routine topics ranging from app troubleshooting to sign-ups and payment options.

Upon contacting support, users are guided through a self-service interactive voice response (IVR) experience, powered by Amazon Connect and Amazon Lex, which has reduced DoorDash’s agent transfers by 49 percent and driven a 12 percent increase in first contact resolution leading to a better experience for DoorDash users, and $3M in YoY operational cost savings. However, with most calls still being redirected to live agents, DoorDash saw an opportunity to further enhance its self-service offering. “We wanted to empower Dashers to get help with their most common questions and issues as quickly and efficiently as possible, saving them time, effort, and increasing their trust in DoorDash's self-service capabilities,” says Chaitanya Hari, Contact Center Product Lead at DoorDash.

Dashers generally prefer calling into support rather than chatting while they’re on the road driving to or from Merchant or Consumer locations. They rely on phone support to provide fast and reliable assistance, which makes response latency in any self-service solution doubly important. DoorDash needed to make sure that Dashers can spend as little time on the phone as possible, so low response latency became a key factor for its phone solution.

To streamline Dasher support, DoorDash’s contact center team wanted to use generative AI for self-service within Amazon Connect. The team sought a solution it could roll out quickly and at scale while maintaining a high standard for issue resolution and customer satisfaction. The team chose Amazon Bedrock, a fully managed service that offers a choice of high-performance foundation models from leading AI companies through a single API, to serve as the base of its solution.

The AWS GenAIIC invested in and worked with DoorDash to design, build, and deploy a custom generative AI solution to enhance its existing voice AI assistant. Over 8 weeks, teams from DoorDash and the GenAIIC iterated over design and implementation and developed a reference architecture that was suitable for production A/B testing. The Amazon Bedrock–based solution reduced generative AI application development time by 50 percent.

To build its solution, DoorDash chose to use Anthropic's Claude models in Amazon Bedrock. Using Amazon Bedrock, DoorDash quickly got access to all the models it needed. Claude was instrumental to the project because it has the capability to mitigate hallucinations, prompt injection events, and detect abusive language. With the release of Claude 3 Haiku, DoorDash achieved the accuracy and speed it needed for its voice application, achieving a response latency of 2.5 seconds or less. DoorDash does not provide any personally identifiable information to be accessed via these generative AI solutions. In addition, Amazon Bedrock enables data security through encryption and ensures data from DoorDash or its customers are only used in the DoorDash application.

“Employing AI-generated responses for phone support presented unique challenges that demanded innovative strategies to enhance response times and answer quality, providing Dashers with best-in-class support. Amazon Bedrock proved to be a perfect fit for our requirements, allowing us to concentrate on refining the solution's finer details," says Vraj Shah, Lead Project Engineer at DoorDash.  

Access to a deep knowledge base was another important factor for the self-service solution. The more in-depth and diverse the knowledge base, the more effective the solution. 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.

To index existing content, it used Knowledge Bases for Amazon Bedrock, a fully managed capability that helps organizations implement the entire RAG workflow from ingestion to retrieval and prompt augmentation without having to build custom integrations to data sources and manage data flows. Knowledge Bases for Amazon Bedrock took care of the backend work for DoorDash.

DoorDash also increased the capacity to test its solution with help from the GenAIIC. Previously, the team had to pull contact center agents off help queues to complete a small number of manual test cases per hour. Using Amazon SageMaker, which developers use to build, train, and deploy machine learning models, DoorDash built a test and evaluation framework that quickly drew insights from A/B testing and evaluated key success metrics at scale. This framework helps DoorDash to complete thousands of automated tests per hour—a 50x increase in capacity—and semantically evaluates responses against ground-truth data.

By automatically handling common Dasher inquiries, DoorDash has improved self-service workflows and increased issue resolution speeds, which helps to accelerate delivery time and enhance Dasher productivity and satisfaction. Addressing routine questions with generative AI has also freed up DoorDash’s live agents to solve higher-complexity issues that benefit from human troubleshooting.