Building a policy knowledge platform for AI-powered compliance

As a global Tier 1 bank, the client faces a highly complex and diverse regulatory landscape. Managing compliance across numerous jurisdictions, products and businesses requires navigating a vast and constantly-evolving body of regulatory obligations.

To address this challenge, the client sought to explore how the application of emerging AI technologies could enhance its compliance framework. The bank maintains an extensive repository of regulatory and compliance policy documents covering a wide range of business activity, and wanted to transform this knowledge into a scalable, intelligence-driven system capable of supporting AI-augmented compliance processes. The objective was not only to improve speed and accuracy of policy advice, but also to create a clear traceability from regulatory change through to policies, procedures, controls and evidence. 

Through an innovation project focused on solving this problem for a targeted domain, the client aimed to kick-start the general exploration of the solution space and develop a scalable methodology that could be applied more widely across the enterprise. Artefacts from the work are to be reused wherever possible to accelerate similar projects in other domains.

Drawing on Icon’s wealth of experience in AI-augmentation of processes within large enterprises, Icon undertook a nine month engagement to deliver the solution.

First, Icon conducted deep semantic modelling for the domain of policy expression, encoding complex normative information into an exploitable form. Building on this, Icon produced a graph-centric data model to support the efficient storage of structured data elements representing policy knowledge assets.

Icon then designed an ontology-based approach to break down raw, unstructured policy documents into structured, canonical knowledge assets, using LLM-backed transformation components. These structured knowledge assets were subsequently used to populate a knowledge graph, enabling complex cross-domain queries and supporting grounded, reliable and explainable Retrieval-Augmented Generation (RAG) for LLMs. Finally, the knowledge graph was leveraged to power various LLM-based policy advice and support agents, which made up the overall solution.

Icon delivered an architectural framework and data platform that transformed the client’s policy and regulatory information into a structured knowledge base, enabling more effective retrieval, and analysis by AI-powered tools and agents.

The solution delivered tangible improvements in the quality and consistency of policy documentation, whilst streamlining key processes including regulatory review, policy management, information retrieval, quality assessment, and workflow governance. It also enhanced the provision of policy advice, equipping compliance teams with faster access to relevant information and supporting responses to policy-related queries.

Importantly, the project achieved more than its immediate objectives. Designed
as a pathfinder initiative, the solution established a scalable foundation for future innovation across the organisation. It has since been used as a launchpad for several similar projects in other domains, providing the data model discipline, AI capability and engineering foundation, which would otherwise have been developed separately, incurring significantly additional costs for those projects.

The client engaged Icon due to our previous experience in their regulatory and compliance domain, along with our deep and comprehensive industry knowledge. Icon’s wide architecture expertise, particularly across data management and AIcentric solution design, was vital to delivering the project successfully.

Icon’s ability to apply semantic modelling, combined with practical, hands-on technology guidance, enabled the client’s team of engineers and regulatory domain SMEs to deliver a meaningful proof-of-concept. This provided a clear, solid foundation for future exploitation and development.