How a major fintech is using GenAI to revolutionize financial document validation

A major Brazilian fintech, specialized in digital automation, faced challenges in validating financial documents, dealing with high volumes of manual processing, high costs, and frequent errors. To solve this, BlueMetrics built a solution based on generative AI that automates the extraction, categorization, and processing of identification documents with high accuracy and scalability.

Overview

The client is a fintech with more than 20 years of market experience, offering innovative digital automation solutions for processes and documents. The company stands out for its full-service approach, supporting mid-size and large companies through their digital transformation. Its clients include some of the largest banks in the country.

In a market defined by intense competition and strict regulation, the client was looking for solutions that would increase operational efficiency, improve its own clients’ experience, and ensure compliance with financial industry standards.

**Market context:

• Growing demand for fast, secure digital processes.
• Need for compliance with financial regulations.
• Intense competition demanding more efficient processes.
• Client expectations for smoother digital experiences.

Problem: how to make document processing more efficient?

The traditional OCR solution the client used had limitations that directly affected operational efficiency and business growth. “We’re talking about hundreds of thousands of documents that have to be processed correctly every month. A textbook use case for generative AI and a great chance to do work that genuinely solves a clear client pain point.”

**Operational limitations:

• High volume of manual document processing.
• Frequent errors in data extraction and categorization.
• Bottlenecks that increased onboarding time.

**Business limitations:

• Difficulty scaling operations without expanding the back office.
• High processing cost per image.
• Client experience compromised during registration.

**Technology limitations:

• Limited, inflexible OCR technology.
• Need for a modern, scalable solution.

The solution: automation and greater accuracy through GenAI.

BlueMetrics built a multimodal generative AI solution to automate and optimize the processing of identification documents, with these highlights:

• **Automated processing of driver’s licenses (CNHs) and ID cards (RGs) in different formats and orientations.
• Precise data extraction, such as name, date of birth, and document number.
• **Automatic categorization of the extracted data, reducing manual work.

The solution uses generative AI to automatically correct image orientation, extract data with high accuracy, and categorize information, offering speed and reliability. It applies mainly to KYC (Know Your Customer), account opening, identity validation, and the automation of registration and contract processes.

According to Diórgenes Eugênio, Head of GenAI at BlueMetrics, “The project we developed with this client to extract information from identification documents went through several phases and approaches, since there are many ways to handle the extraction of organized text from images. The first major obstacle we found was that image orientation directly affected the models’ ability to identify and extract information. Once we spotted that obstacle, we explored a few solutions, including using generative AI models to determine how many degrees were needed to bring the image to a standard orientation. But the results were not satisfactory. So our final pipeline used Tesseract running alongside an API to detect when rotation was needed. After that, we evaluated several techniques for improving image quality and boosting character recognition. For the specific context of identification documents, not all of them delivered meaningful gains. The project had a major business impact, since the big difficulty with OCR solutions today is organizing the information, correlating the extracted text with what the document actually means. Beyond that advance, the architecture we created is built for constant evolution, since we keep getting cheaper and more capable models.”

Key features

The solution introduced several meaningful technical advantages that improved both process effectiveness and the client experience, and delivered concrete results for the operation as a whole.

**Smart pre-processing:

• Automatic detection of document orientation.
• Position correction and image optimization.

**Advanced extraction pipeline:

• Multimodal generative AI models for data extraction.
• Intelligent categorization and parallel processing.

**Solution highlights:

• **Adaptability: works with different documents and orientations.
• **Accuracy: fewer errors and greater precision in data extraction.
• **Scalability: cloud-native architecture to process large volumes of documents.

Results

Rolling out the solution brought a series of benefits for the client, translating into concrete, immediate financial results.

**Operational efficiency

• Lower operating costs and shorter average onboarding time;
• No more document-processing bottlenecks;
• High productivity with greater simultaneous processing capacity.

**Quality and accuracy

• High accuracy in data extraction and categorization;
• Significant drop in errors and rework.

**Business impact

• Greater scalability to handle demand peaks;
• Flexibility to process multiple document types;
• An evolving solution, aligned with future needs.

**Compliance and security

• Full traceability and compliance with financial regulations;
• Better detection of fraud attempts.

Conclusion

With the new generative AI solution, the client transformed its document validation processes, cementing its position as an innovation leader in the financial market. “The partnership with BlueMetrics showed how advanced technology and an agile implementation method can turn operational challenges into lasting competitive advantages.”