Complex Obligation Generation
Governance, Risk, and Compliance applications pertain to a level of thoughtful action more than any business function. Knowing what rules to enact and detract is very much needed for the efficient and secure functioning of any GRC related system.
This problem is redeemed by the creation of obligation rules, an intermediary set of rules which define what action should be taken or refrained from based on the internal and external compliance rules. This is more or less a summary of related and deviant rules. Every obligatory rule is more or less linked to another obligatory rules in an relational, both internal and external rules contain them.
California, USA
Client
Governance, Risk, and Compliance Industry
CLIENT
Our client, based in California, has been working in the field of providing AI-based GRC support tools for many major BFSI businesses for the past 3+ years. We were asked to develop an automated Obligation Generation Method based on the operations that were commenced by our Semantic Search Engine. Using the tags and relations derived from the previous search engine model, we were tasked to create a machine learning based obligation summarization tool which can convert obligation rules to a new obligation which has high traceability across all selected obligations and with no redundancy of the information.
CHALLENGE
Our model needs to tackle these challenges that were identified by our field analysis :
-
Veracity of the Obligations - The generated rule summary from the bookmarked related-rules, across regulators’ law and standards, can then be shared across other compliance bodies for peer review and correction for maintaining the integrity of the obligations without any negative effects.
-
Context and Concept Accuracy - Keeping the main intent and key concepts from all selected obligations and then creating a new obligation is a quite complex process due to the size and the crossing of various obligation logics.
SOLUTION
With the challenges in mind, we have created flexible solutions to deal with them:
-
An ensemble machine learning model, i.e., which contains several base models (such as concept extraction, duplicate removal of concept & contexts using encoded representation of each rules, chronology and context based content binding) is used for the obligation summarization.
-
After going through each steps of the ensemble learning model; building the concepts, remove redundancy, identify concepts and etc, we apply a generative based business language rewriting of the final summary to generate multiple variations of obligation summaries.
-
The obligations are then checked for traceability (positive relations) and overlap (redundant relations) and the most context-correct and relevant obligations are selected for display with their related themes and relevancy scores.
RESULT
-
The generated obligations can direct the compliance agents to the right direction of decision; what is needed to be done versus what is needed to be averted while applying the various rules.
-
All obligations are centralized and do not deviate in drastic degrees which could affect our decision making processes.
-
By displaying the obligations with a relevancy score, the source of rules, the country jurisdiction and the themes present we can easily identify where these obligations have to be introduced.