Sun Finance automates ID extraction and fraud detection with generative AI on AWS

2026-04-30 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, long

Summary

Sun Finance, a Latvian fintech operating across nine countries, partnered with the AWS Generative AI Innovation Center to automate identity document (ID) extraction and fraud detection. Facing a 60% manual review rate for 80,000 monthly microloan applications, primarily due to OCR errors and fraud, the company sought a new solution. Within 35 business days of technical handover, a new pipeline leveraging Amazon Bedrock (Claude Sonnet 4, Titan Multimodal Embeddings), Amazon Textract, and Amazon Rekognition was live. This solution improved ID extraction accuracy from 79.7% to 90.8%, reduced per-document costs by 91%, and cut processing time from up to 20 hours to under 5 seconds. The fraud detection system achieved 81% accuracy, combining visual pattern detection and background similarity analysis.

Key takeaway

For AI Engineers building document processing or fraud detection systems, prioritize a multi-component architecture. Your team should combine specialized OCR services like Amazon Textract with LLMs for structuring to overcome PII limitations and improve accuracy. Implement parallel fraud detection methods, such as visual analysis and vector similarity search, to enhance overall detection rates and processing speed. This approach can drastically reduce manual review and operational costs.

Key insights

Combining specialized OCR with LLM structuring significantly enhances ID extraction accuracy and efficiency.

Principles

• Separate OCR from LLM structuring for PII-sensitive documents.
• Multi-tier OCR provides resilience against diverse document quality.
• Fraud detection benefits from multiple, complementary methods.

Method

The ID extraction pipeline uses Amazon Textract for primary OCR, Amazon Rekognition as a fallback, and Claude Sonnet 4 via Amazon Bedrock for structuring and validation. Fraud detection runs visual pattern analysis and background similarity search in parallel.

In practice

• Use Amazon Textract for raw text extraction from IDs.
• Employ Claude Sonnet 4 for intelligent text structuring.
• Implement Amazon S3 Vectors for fraud pattern similarity search.

Topics

• Generative AI
• Identity Verification
• Fraud Detection
• Amazon Bedrock
• Amazon Textract

Best for: AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• Rocket Close transforms mortgage document processing with Amazon Bedrock and Amazon Textract — Generative AI, combined with OCR, can automate complex document processing with high accuracy and speed.
• A Fraudster’s Paradise — Generative AI significantly amplifies the scale and sophistication of online fraud, impacting both professional criminals and casual cheaters.
• HybridCodeAuthorship: A Benchmark Dataset for Line-Level Code Authorship Detection — The increasing hybrid nature of codebases necessitates fine-grained, line-level AI authorship detection benchmarks.
• Intelligence-driven message defense and insights using Amazon Bedrock — Generative AI on Amazon Bedrock effectively detects disguised contact information and extracts sentiment from communications, surpassing regex limitations.
• On the troubling rise of generative AI suspicion in academic publishing — Overextended suspicion of GenAI in academic publishing leads to false positives and an "AI witch hunt."
• Why Commonwealth Bank’s $1 billion suspected loan fraud should change how we bank and do business — AI-generated documents are facilitating large-scale financial fraud, prompting increased security measures and regulatory scrutiny.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.