Financial spreading analysis FOR MULTINATIONAL BANK
Multinational Japanese bank
Accurately extract a dozen of key elements from large number of public companies’ Annual and Quarterly Financial Reports (10Q)
Investment banks analyze quarterly financial reports (10Q) from a long list of public companies that they are interested in investing. The information that need to be extracted from these reports varies based on the business needs. Financial Spreading is a common use case where the overall auditing rating and financial balance sheets need to be identified and extracted to a database. For this particular requirement, foreign exchange swaps of varies types need to extracted from the reports, so that the business analysts can take the output data to plug into their risk assessment models to evaluate these companies.
Traditionally, these 10Q reports are analyzed by a team of junior analysts manually. They first find the relevant sections and paragraphs to copy and paste them to a set of new documents. They will take highlight important numbers that are corresponding to the foreign exchange interest swaps, etc., and then manually extract these numbers to spreadsheets. The spreadsheets are then used by the risk assessment models to produce the final results.
What We did
Singularity was given the opportunity to automate this process with AI / IDP solutions. It took one of our interns from Princeton University a few weeks to label over 2000 individual reports and build a viable machine learning model to extract the information from the summarized files.
We achieved over 98% accuracy for their testing files, which is more accurate than their past human extracted values, and a lot more consistent as well. Consistency is a business value that is above and beyond the cost-saving ROI as stated below.