Chime Financial, a US fintech serving millions of banking members, deployed Anthropic's Claude via Amazon Bedrock to automate post-call summarization, saving more than 250,000 hours per year on call summaries while cutting average agent handling time by 18 seconds per call and lifting the customer service net promoter score by five points.
"Gone are the days of agents furiously typing notes while customers are on hold," says Alexa Schaeffer, strategic initiatives lead for the operations and member experience team at Chime. "Now our agents, empowered by AI assistance built on Amazon Bedrock, can focus on customers, empathizing with their issues and finding solutions as quickly as possible."
Before the deployment, customer service agents juggled multiple tasks during every call — listening, manual note-taking, and post-call summaries — which detracted from member experience across thousands of weekly interactions. "Our ability to be efficient means that we can provide our customers with a better experience," says Dennis Yang, generative AI product team lead at Chime.
Chime built an early AI call-summarization tool during an internal hackathon in May 2023, then tested the concept throughout 2023 against multiple foundation models on Amazon Bedrock before choosing Anthropic's Claude for its reasoning capability and long-context handling. The production system launched in February 2024 and is now available to agents on every call.
Chime routes calls through Amazon Transcribe, applies Amazon Bedrock Guardrails to redact identifying and sensitive information from each transcript, and sends the redacted text to Claude models in Amazon Bedrock for summarization, all consumed through ChimeCore, the company's internal technology platform. The architecture keeps sensitive data inside Chime's existing AWS environment.
The deployment produces call summaries in seconds for thousands of weekly customer calls, saving more than 250,000 hours per year (~$700,000 in annual efficiency gains). Agents use summaries for continuity on repeat callers, internal teams use them for service-improvement analytics, and partners receive post-call summaries on demand. "We didn't have to figure out our own process or dedicate engineering resources to building custom safeguards," says Mukund Sarma, product security lead at Chime. "We got enterprise-grade protection out of the box using Amazon Bedrock Guardrails, so we could move faster while maintaining our high security standards."
"We used Amazon Bedrock to go quickly from hackathon concept to production and scale AI across thousands of customer calls, delivering immediate business value with measurable cost savings and improved customer satisfaction," says Yang. "We're using AWS to propel our innovation in generative AI, and we're not just adapting to this technology — we're setting the standard for how AI can enhance customer experience in financial services."