The Business Need
A fast-paced news agency was looking to extract financial data from annual reports, statements, and related filings of bank firms. One of their requirements was to populate a standard template with financial data making necessary adjustments and transformations to adhere to the template. The extracted data would help the agency compile and analyze the data for publishing comprehensive reports for corporates.
As one of the leading providers of financial news, they were in need of an automated process crafted using a tool that promises to minimize human intervention as much as possible.
Challenges we faced
The formats and terminologies in which the necessary data points were found varied from report to report. They were not always found in the same place within the document under the same data fields.
Hence no rule-based script would be able to pick the information from the reports. So a machine learning approach was sought for in this complex process.
Also, to arrive at the data points the client needed, we had to make formula adjustments to the available data and then populate them to the database.
How we solved the problem
Annual reports of more than 1000 companies with entity names and year of the report were given by the client. Previous year details were also provided to make the necessary calculations.
We identified base data like company’s zip code, telephone number, address, branch, C-suite details, parent-subsidiary companies, subsidiary, debt, balance sheet information and extracted. Performance ratio indicators were then expertly calculated.
The captured financial data from the annual reports and statements were consequently made available in a web application so that the populated data can then be validated by the client along with the previous year’s USD value & local currency value using an automated component.
Once their validation process was done, the financial records that were generated by the analysts were sent for QC approval, and the signed-off data received from QC were updated in their system.
A machine learning solution was built that intelligently identified tables and sections within the reports were critical financial data was located. The output was projected on our homegrown Mojo tool that helped human intelligence to be added to the ML output and improved accuracy.
Results
With tools like Mobius’ Mojo interface at our disposal, we had the potential to enhance the user’s accessibility to unstructured data. The ML solution delivered for this client cutback manual assets by 30% while ensuring an accuracy of 95% consistently.