Reka La Roze, Change Manager at Unigestion, was faced with a challenge. The Private Equity team had struggled with poor-quality data from private equity quarterly reports for some time. Poor data meant delays and too many inaccuracies.
Unigestion was using an external vendor to extract data from quarterly reports. The asset management company has zero tolerance for errors, so the Unigestion team had to resort to doing manual data extraction themselves, on top of the vendor’s data extraction.
A decision was made to take action. It was time to modernise Unigestion’s Private Equity data process by implementing a new technology.
Automating the extraction of data from quarterly reports is difficult. The documents are highly complex, and each fund has a different way of laying out data. Using a traditional OCR to perform this task was not feasible since Unigestion was looking at the reports of thousands of different portfolio companies. OCR requires a customised template for each of potential document layout, so costs would quickly spiral out of control.
Most firms in the asset management industry extract data from quarterly reports manually or use an external vendor, at large expense. Manual data extraction can lead to good accuracy initially, but knowledge is lost because staff turnover is generally high. Having to continually retrain staff is time-consuming and often a reason for low-quality data.
Prior to meeting Evolution AI, Unigestion had experienced demos from several AI start-up companies, but their claims didn’t seem to match the reality. Reka La Roze found, “OCR software with some rules bolted on was closer to the truth.”
Unigestion came across Evolution AI and was impressed. The project team said “The interface is very intuitive. When we saw the demo we could picture our team using it without needing additional time for training.”
Evolution AI’s proposal was to provide artificial intelligence-based data extraction and a layer of human verification. Just two operators were needed to first train the AI model and validate its outputs.
Evolution AI’s software is able to deal with a wide range of document layouts and can learn from one or two examples. Importantly, artificial intelligence has the ability to generalise. Evolution AI algorithms can read and understand documents with layouts they haven’t seen before.
Another critical difference between traditional OCR and artificial intelligence is learning. True AI models learn after being corrected, increasing accuracy over time.
We really appreciate that Evolution AI were available practically 20 hrs a day 7 days a week — Reka La Roze, Change Manager at Unigestion.
Unigestion selected Evolution AI after a short trial, which demonstrated the accuracy of the artificial intelligence data extraction. The production project kicked off soon afterwards, and integration was completed within months. Both Evolution AI and Unigestion teams worked hand-in-hand to make this project successful. Rigorous testing proved that data accuracy had increased from less than 30% to over 99.5% with Evolution AI.
We saved 75% of our costs, the data quality is excellent, and time to completion was halved. It's impressive — Cyril Kirschwing, Head of Change and IT Solution Architecture at Unigestion.
The Unigestion Private Equity team recently presented the results to Unigestion’s Executive Committee, who were pleased with the outcomes. Unigestion’s work isn’t quite finished yet, though. Reka La Roze states, “We now have people queuing up to use the solution and I look forward to working with Evolution AI to implement this across the team.”
