Share with your friends

A fantastic opportunity for the financial services industry lies in the fact that 80% of available data is unstructured and barely explored. Innovative banks and fintech companies are leading the way on text-data capitalization projects, adding digital value by processing a vast amount of information from diverse sources. Text applications are booming across multiple domains, including:

  • risk management for client onboarding
  • customer support
  • document management
  • automatic contract review

Why are text applications booming?

The text app explosion is due to an evolution in technology and the development of Deep Learning (DL). Deep Neural Networks (DNN) are replacing dictionary-based, rule-based and other conventional machine learning techniques because they can capture linguistic and other semantic features closer to human reasoning (I. Chalkidis, D. Kampas, 2018). Recurrent and other neural networks enhance machines' ability to grasp language and meaning more than ever before.

KPMG explores market needs to deliver high-end solutions for our clients. One of our passions is to ease the burden of the finance sector's day-to-day regulatory compliance.

Innovative banks and fintech companies are leading the way on text-data capitalization projects, adding digital value by processing a vast amount of information from diverse sources

Sven Muehlenbrock
Partner, KPMG Luxembourg

NLP for bank compliance: The KPMG Law Repository Use case

A bank's compliance setup shields them from damage resulting from non-compliance with standards applied to its various market activities. As a result, compliance officers must oversee a daunting amount of regulation from different regulatory sources. The day-to-day regulatory compliance function is resource-heavy, time-consuming and nests risks for a financial services organization. Compliance officers face the following challenges:

  • Keeping track of changes and amendments to a complex and dynamic regulatory environment.
  • Updating compliance policies and regulatory procedures.
  • Detecting and analyzing compliance risk issues.
  • Providing senior management with thorough compliance advice and recommendations.

With this in mind, KPMG is developing a cognitive tool that gathers sections of relevant law at European and national level. It's a legal knowledge base that uses high-quality semantic logic and state-of-the-art NLP tools. We scrape EUR-Lex and CSSF sources daily and insert annotated pieces of law into a semantic grammar that you can retrieve with a user-friendly and pared-down interface (Figure 1). Our tool, which is set to go live in November, aims to:

  • retrieve relevant regulation and quickly identify outdated information 
  • infer regulatory body legal advice on a broad variety of risk topics 
  • spot all requirements and prohibitions that apply to you from a deep pool of regulatory data
  • map requirements and prohibitions to each corresponding entity.

Combining semantic web and natural language processing technologies allows us to automatically identify, link, classify and tag all regulatory requirements that apply to a client's unique business. Instead of filtering legal documents by keywords or tags, our tool combines keywords, legal categories and entity tags in a single query.

A great promise

The solution will also allow users to select sections of the retrieved legislations and compile them rapidly and dynamically into a new document (Figure 2). As a result, the compliance function's research process will be reduced to only a few minutes, while boosting its quality.  

KPMG's project will build a semantic framework of CSSF circulars and Luxembourgish laws linked to their relevant EU legislation. This framework will allow a compliance officer to access a rundown of both layers of regulation, reducing the risk of missing significant compliance information. By building an AI-powered search engine for financial regulations, officers will only retrieve information that is tailored to their compliance framework.


Chalkidis I. & Kampas D., Deep Learning in Law: Early Adaptation and Legal Word Embeddings Trained on Large Corpora, Artificial Intelligence and Law, 2018