Automating Bank Check Verification with Vision AI and Generative AI

Customer Overview

Our client, a leading technology company, provides digital solutions to banks and financial institutions across the US and Canada. Check verification is critical in check processing, as both receiving and paying banks must verify checks before sending digital copies for clearing. Banks approached our client to automate check verification, which involved multiple manual steps, like checking account and routing numbers, validating signatures, ensuring funds availability, and managing potential returns. The manual processes were prone to delays, errors, missed details, and exposure to fraudulent checks. The goal was to develop a solution to improve check verification speed and accuracy, while mitigating risks.

Project Overview

The client sought to address the inefficiencies and risks inherent in the manual check verification process. They wanted to automate every stage of check verification, from reading handwritten or printed data to detecting forgery and validating authenticity. The goal was to provide banks with a scalable, reliable, and compliant system that could handle high volumes of checks efficiently and improve overall operational accuracy and risk management. They partnered with TenUp to build an automated solution that streamlines the check verification process, minimizes human intervention, and ensures faster, more reliable validation.

Challenges

Building a system that could handle check validation, fraud checks, and reporting with high accuracy and compliance, while operating seamlessly within existing banking infrastructures.

• Managing wide variations in check layouts, fonts, and handwriting styles without loss of accuracy.
• Extracting and validating multiple data points, including account details, signature, date, and amount, while ensuring the written and numeric values matched.
• Detecting tampering, overwriting, or forged signatures in scanned or digital check images.
• Generating reports that summarize check verification results, highlight exceptions, and indicate which items require manual intervention.
• Ensuring seamless integration with legacy banking systems while maintaining data security and auditability.

Solution

Combining Vision AI, OCR, and Generative AI, we developed a system that automated check verification, fraud detection, and reporting with high accuracy and compliance.

• Trained two Vision AI models (an Object Detection Model and an OCR Model) to adapt to diverse check formats, fonts, and handwriting styles without compromising accuracy.
• Configured the Object Detection Model to identify and create bounding boxes around key data fields, like account number, signature, date, bank details, and amount. Used the OCR Model to extract the information from the detected bounding boxes and validated it by calling third-party and internal APIs.
• Created specialized utilities within the system to compare extracted data against maintained databases of key criteria like signature specimens, bank details, and routing numbers. The system also verified that the amount written in numbers and words matched.
• For accurate authentication of signatures on checks, we implemented an AI-based signature verification solution using Deep Learning to extract and compare signature feature embeddings against stored customer references.
• Implemented a Generative AI module to generate summary reports with checklist-based outputs, indicating validation results for signature match, funds availability, fraud detection, bank name & bank branch, and other key details. It also flags exceptions requiring manual intervention, providing an actionable overview for faster decision-making and compliance.
• Built seamless integration layers to ensure compatibility with existing banking systems while maintaining data security and auditability.

Benefits

The AI workflow automation solution we developed provided the following advantages to our client:

• Up to 70% faster check verification through automation of validation, fraud detection, and reporting workflows.
• Nearly 60% reduction in manual effort and operational costs while maintaining full compliance and audit readiness.
• AI-powered validation enhanced fraud detection by over 90%, minimizing missed anomalies and false approvals.
• Fully compatible with legacy and modern banking systems, ensuring smooth adoption without infrastructure changes.

Technology

• DETR (Detection Transformer)
• DocTR (Document Image Transformer)
• Embedding Models via Ollama
• Mistral via Ollama
• Python
• FastAPI

Industry

• Fintech