Intercom revolutionizes customer service using AWS and Anthropic

Since its founding in 2011, Intercom has relied on the scalable infrastructure of AWS to support its hypergrowth trajectory. As AI emerged as a transformative force in customer service, the company recognized the need to develop a sophisticated solution that could automate support while maintaining high accuracy standards. The challenge was particularly complex, given the need to handle various types of customer inquiries and protect against AI hallucinations. 

The company needed a solution that could effectively handle the multiple components of the customer support process. This included assuring that a question is answerable, applying guidance that complies with customer support policies, and generating accurate responses. This in turn meant that Intercom needed to find the right AI partner to help develop an architecture that could scale reliably while maintaining the company’s high performance standards. 

Through extensive testing, Anthropic’s Claude emerged as the highest-performing large language model for Intercom’s needs. Intercom worked with Anthropic to develop Fin on AWS, using services including Amazon Bedrock, a fully managed service that offers a choice of high-performing foundation models, and Amazon SageMaker, which delivers an integrated experience for analytics and AI. The solution is made up of 15–20 different subcomponents, each paired with a specific large language model optimized for tasks like answer verification, policy compliance checking, and response generation.

The architecture lets Fin analyze customer inquiries, access knowledge repositories, fact-check answers, and verify compliance with policies and guidelines. The solution can also identify opportunities for further automation. “Fin will say, ‘Hey, here's knowledge I think you should add. Here’s an API that, if I had, I could resolve more,’” says Des Traynor, cofounder of Intercom.

The implementation has delivered significant results for Intercom and its customers. Within 30 days of deployment, customers achieve, on average, a resolution rate of 56 percent, with some customers reaching 80–90 percent. The success has led to rapid adoption, with over 4,000 customers implementing Fin since its March 2023 launch. The solution’s effectiveness is further validated by Anthropic becoming one of Fin’s largest customers, using it to handle Anthropic’s own customer support volume.

Looking ahead, Intercom will continue to innovate, exploring opportunities in proactive support, visual processing, and audio capabilities. “We want to keep pushing the boundaries,” says Traynor. “We want to always be on the edge, testing to see what’s possible using AWS along with Anthropic.”