Sun Finance automates ID extraction and fraud detection with generative AI on AWS
TL;DR
Sun Finance built an AI-powered identity verification pipeline on Amazon Bedrock, Textract and Rekognition. Extraction accuracy rose from 79.7% to 90.8%, per-document costs dropped 91%, and processing time fell from up to 20 hours to under 5 seconds. Combining specialized OCR with LLM structuring outperformed using either tool alone, and a serverless fraud-detection layer uses vector similarity search.
Nauti's Take
Nauti finds this case impressive: 91% lower costs and processing time from 20 hours to under 5 seconds are hard, measurable wins you rarely see documented this clearly. Risk: 90.8% accuracy sounds great, but the remaining 9.2% errors are in KYC territory, where every single case matters for AML and identity fraud.
A blueprint for finance firms with high document volume — compliance teams still need to nail thresholds and escalation paths.
Sources
