Ramp built an AI agent using LLMs, RAG, and embeddings to automatically fix incorrect merchant classifications for credit card transactions. The system resolves customer requests in under 10 seconds instead of hours, handling nearly 100% of requests compared to the previous 1.5-3% manual resolution rate. The agent uses multimodal context including transaction data, receipt images, and user memos, with careful guardrails to prevent hallucinations and inappropriate actions.
Table of contents
BackgroundProblem: Using User Input to Fix MatchesApproachCommunication to UsersEvaluationImpactExtensionsFinal ThoughtsSort: