At a glance
Industry
Banking and Capital Markets
Location
Global
Challenge
Minimize the time, cost and resources needed to perform key information extraction from large quantities of complex documentation across printed, graphical and handwritten formats.
Success Highlights
- 96% accuracy in key information extraction from over 20 types of documents
- 98% faster document processing time—from average 30 minutes down to 30 seconds
- Up to 98% reduction in average cost per document processed—from $5 down to $0.10
Our approach
We worked with the bank’s AI center of excellence to build, test and refine a solution based on optical character recognition (OCR), machine learning (ML) and gen AI. All of these technologies have matured significantly in recent times, making them ideal partners for an automated solution.
Drawing on specialist Cognizant AI expertise gained in academia and across multiple industries, we were able to develop a minimum viable product (MVP) in just four weeks. We then refined the solution until it performed accurately enough to be deployed into production.
The key capabilities of the solution are:
- OCR text extraction: We used Microsoft Azure AI Digital Intelligence as the OCR engine to scan structured and unstructured documents and turn them into digitized text. It employs deep learning to mimic human visual recognition and excels at recognizing printed and handwritten text—including from low-quality images and scans. Its API also made it ideal for integration into the process flow of this solution.
- ML-based document classification: A classification stage enables relevant documents to be identified and irrelevant information to be filtered out, making the KIE process even more efficient. We devised and trained an ML algorithm to classify documents that had been digitized using OCR, so only relevant information would be put forward for gen AI analysis.
- Analysis with gen AI: We instructed Microsoft Azure OpenAI GPT-4o to analyze the digitized text from relevant documents and extract key information from them. It identifies entities and relationships between them, enabling it to extract the fields required for the relevant business process. In addition, we used GPT-4o’s multimodal capabilities to identify and extract relevant images, including signatures, logos and seals.
- Back-office integration: The working solution runs on Microsoft Azure, our client’s strategic cloud platform, and integrates with relevant back-office systems also running in Azure.
- Human in the loop: While the solution automates the extraction of key information, all extracted fields are reviewed by a bank employee to ensure they are accurate.
Business outcomes
The solution is currently deployed in over 20 different use cases across the bank, including customer onboarding, mortgage applications, fraud claims processing and assignment of customer power of attorney. Its wide applicability means it can be applied to further use cases in the future. Since going live, the solution has delivered significant benefits for the bank and its customers in terms of process acceleration, cost reduction, optimized workloads and customer satisfaction.
Key benefits achieved to date include:
- 96% accuracy in key information extraction: The solution is capable of extracting relevant information from tens of thousands of documents with 96% accuracy, and without requiring help from SMEs when dealing with complex legal or technical documents.
- 98% faster KIE process: The average time to process a single document before the solution was 30 minutes. The solution takes an average of 30 seconds to perform the same task—from digitizing the selected documents with OCR, to extracting the key fields using Gen AI and populating a form in the relevant back-office system.
- Up to 98% reduction in document processing costs: The solution has the potential to reduce the cost of processing a single document from an average of $5 to just $0.10 or less. The savings potential across the bank’s business divisions on a worldwide basis is significant.
- Optimized workloads: Subject matter experts in legal and technical matters no longer have to be called in to help identify the relevant information in a complex document. This streamlines their workload and frees up more time for their primary job role.
- Enhance customer experience: Key customer-facing processes can now be completed much faster, delivering a significantly-improved customer experience.
