How Sun Finance and AWS accelerated document verification and reduced fraud risk

Sun Finance moved identity verification to automated mode using AWS services and generative AI. The company built a pipeline that extracts data from documents faster, processes applications more cheaply, and helps catch suspicious matches in near-real-time.

How the Pipeline Works

The solution is built on dividing tasks among multiple tools. Amazon Textract handles OCR and extracts text from identity documents, Amazon Bedrock structures the output and normalizes fields, and Amazon Rekognition supports visual verification and anti-fraud scenarios. This approach proved more practical than trying to solve everything with a single model: a specialized service performs extraction, while the LLM takes on interpretation, format correction, and assembling the final structure for internal systems.

Architecture speed is equally critical. Sun Finance built a serverless IDV pipeline where processing is triggered by events and does not require constant infrastructure for peak loads. This is especially useful for lending and fintech processes where document flow is unstable: it can be minimal at night and spike sharply during issuance hours. In this scheme, cost control is easier, and response time does not depend on manual queues of operators who previously could delay verification for many hours.

• OCR extracts raw text and fields from the document
• LLM brings data into a unified structure and removes noise
• Rekognition adds visual signals for identity verification
• Vector search helps find suspiciously similar applications

For anti-fraud in this scheme, it is important not only to read the document but also to compare it with already processed cases. Vector search allows storing application embeddings and quickly finding close matches by image, text, or combination of features. If the system detects overly similar documents, repeating patterns, or atypical proximity between different applications, it can send them for additional review. This provides a more flexible layer of protection than simple hard rules.

Why It Worked

The key finding is that combining OCR and LLM produces better results than using each component independently. According to Sun Finance, extraction accuracy increased from 79.7% to 90.8%. For such scenarios, this is not a cosmetic improvement: every few percentage points of accuracy directly impact the volume of manual rework, the number of form errors, and the speed of application decision-making.

The LLM here does not replace specialized recognition but complements it: it understands field context, aligns names, eliminates typical OCR artifacts, and prepares data for further automation.

The economic impact is equally notable. The cost per document processing, according to AWS, dropped by 91%, and processing time fell from a maximum of 20 hours to less than 5 seconds. For fintech, this means several things at once: lower operational costs, fewer rejections due to long wait times, and a better chance to verify a customer before they abandon the application.

In parallel, the vector-based anti-fraud system allows searching for repeated or overly similar applications without heavy server overhead. This makes verification not only faster but also more resistant to attempts to bypass basic rules.

What This Means

The Sun Finance case shows that generative AI in document processing works best not as a universal replacement for the entire stack, but as a layer on top of already strong specialized services. For banks, MFOs, and other companies with KYC processes, this sends a clear signal: if OCR, structuring, and anti-fraud are properly distributed across different components, you can simultaneously improve accuracy, reduce response time to seconds, and significantly lower the cost per verification.