Let's start from the beginning: Fyndoo’s team develops technology for loans. When someone applies for a loan, they must prove their ability to both secure and repay it by submitting various documents and going through different processes. In the Netherlands, we have many standards to which various market and government entities adhere, allowing us to extract data directly from source systems. While this approach works well, it’s not scalable since these standards are not universally used outside the Netherlands.
From OCR to GenAI Revolution
We began using Optical Character Recognition (OCR) four years ago with a simple idea: upload a PDF of a company’s annual report, automatically structure it, and have our system extract the necessary data to integrate directly into Fyndoo. This would enable short-term credit ratings, streamline credit decisions, and enhance customer management. Despite a promising start, we eventually stopped investing in OCR. While OCR excelled at extracting text from unstructured documents, it lacked the required "interpreting intelligence."
That's where GenAI came into play—a highly promising technology, though initially lacking clear application cases. GenAI uses data as a source, which saves a lot of time. Bringing stakeholders together to adopt standards is a huge challenge, but GenAI sidesteps this completely. Ultimately, we saw great potential in using a GPT model to analyse and interpret the documents. By integrating foundational models from suppliers like NVIDIA and OpenAI, and leveraging their ability to understand numbers and context in natural language, this technology became far more powerful than OCR. Recently, we organised a two-day hackathon with five colleagues and esteemed partners like Microsoft to use Microsoft Azure AI to increase accuracy and develop a launchable version of a financial statement spreading co-pilot.
"Personally, I think this is a wonderful development, though it's ironic: what was once structured gets put into an unstructured PDF, and we have to untangle it all over again. A bit of a shame, right?"
Next Steps
The hackathon was a resounding success, showcasing the immense potential of Generative AI in transforming financial processes. Over two days, our team, in collaboration with Microsoft and other key partners, developed a highly accurate and functional prototype of a financial statement spreading co-pilot using Microsoft Azure AI.
This new tool not only surpasses the capabilities of OCR but also introduces a level of interpretative intelligence that can significantly streamline credit decisions and customer management. The promising results have set the stage for further development, bringing us closer to launching a solution that will revolutionise how we assist our customers in achieving their financial goals. Stay tuned as we continue to innovate and refine this technology in the coming months.