To increase speed, reliability and accuracy, CACEIS is leveraging on artificial intelligence (AI) to develop its machine reading capabilities.
Each year, CACEIS read through more than 13,300 fund prospectuses to extract information. A French fund prospectus contains an average of 20 pages, a Luxembourg prospectus between 150 and 200 pages. At CACEIS, for each prospectus, approximately 140 data points (information and management rules) must be identified and extracted to feed CACEIS’ reference systems and to perform the functions of depositary, fund administration and fund distribution, where applicable. This highly time-consuming activity has until now been carried out by several teams at CACEIS.
In order to increase speed, reliability and guarantee the completeness of the data extracted, the CACEIS Innovation Lab, which was set up in 2018 to develop innovation at CACEIS, proposed the use of AI to automate and digitise the reading of prospectuses. An AI-search engine will make it possible to centralise the reading and selection of data to be extracted.
The benefits of AI are multiple: continuous improvement of the tool as data is integrated, considerable productivity gains (only 3 minutes are needed to read 100 prospectuses compared to 20 man-days manually), reduced operational risk and improved quality controls.
A TOOL BY APRIL 2020
The project was launched in early 2019 with the collaboration of CACD2, the Crédit Agricole Group’s digital manufacturing arm. The tool currently being developed, SPA, or Smart Prospectus Analysis, was initially the subject of a POC (Proof of Concept) on the scope of some 50 prospectuses under French law. The POC proved to be conclusive.
In light of this success, next steps are an MVP (Minimum Viable Project) and a target date of April 2020 for the implementation of the tool to all fund prospectuses.