The problem

The operational risk and mitigation team of a large US bank is very distributed across the globe, and issues arising within the bank have global implications. A major problem that exists today is that these are rarely socialized and resolved uniquely across different lines of businesses and geographical regions. Financial institutions are very regulatory, and with such processes depending on huge amounts of textual data, many processes are manual and repetitive across teams.

The center of excellence team of the bank wanted to see if Cognitive BI could help solve this problem in two phases. First, to develop a model that would group similar issues across different lines of businesses, thereby enabling the bank to effortlessly tackle the issue of interest with the associated remediation steps; and second, a user-friendly interface to enable the business users to narrow down a specific issue of interest.

The cognitive solution

Though it was not a difficult problem to solve, it was a hard one to orchestrate. The data was readily available; however, putting in all into one place was a challenge, since the accesses and controls within the organization were strictly regulated. The entire problem was tackled in three stages:

1. Data cleaning and text feature generation

• Create a central data repository by lines of business groups on the various issues and resolutions for the same
• Store unstructured, preprocessed text to generate features – tokenization, word removal stopping, lemmatization, TF/IDF

2. Weighted similarity scoring and validation

• Group similar issue category together and assign similarity scores (Sine, Jaccard)
• Developed a similarity scoring engine that maps issues with similar issues and their associated mitigation steps
• An unsupervised clustering model to group issues.
• Use a weighted similarity score based on compliance levels and number of regulatory rules to be followed by the category of issue

3. Consumption and scale

• Access-based login dashboard to enable consumption and keeping the organization’s access control intact
• Dynamically identify similar issues based on the text of the current issue in real-time
• A constantly updated database of over 200 clusters of issues spanning over 20 lines of businesses

Key insights

The entire analysis proved to be a major headway towards the bank’s vision for 2022 to resolving existing operational risk issues.

• Scale down 1000+ issues to mere 200 unique issues, enabling the business to reduce the number of maintained mitigation systems and address issues in a standardized manner.
• The use of the dashboard to find relevant issues drastically brought down the time to identify and mitigate an issue from 8-10 days to less than 6 hours on average.

This has brought a great deal of traction between the different lines of businesses, and offered an opportunity to enrich the data for newer insights and extend the analysis to look at operational losses and regulatory risks.