Michael Lawrence from RegTech Associates provides an overview of some of the key applications of AI in RegTech, with examples of exciting products to take note of.
In 2023, regulated firms face increasing regulatory pressure, an explosion of data, and, in some cases, sophisticated criminal networks that have become wise to old compliance processes. Simultaneously, Artificial Intelligence (AI) has experienced a step change in its capabilities. Now, AI is more than experimental, it is ready to be applied across a wide variety of use cases across a range of industries. Regulatory obligations are rife with diverse and complex challenges and incumbent compliance processes are often manual or otherwise outdated. RegTech stands to deliver significant benefits if it can harness emerging AI capabilities.
As an inherently forward-looking industry, RegTech was ahead of the curve with respect to AI adoption, and the many existing applications across products can develop further in light of recent technological developments. This blog provides an overview of some of the key applications of AI in RegTech, with examples of exciting products to take note of.
Figure 1 shows, for various categories of the RegTech Associates RegTech taxonomy, the number of RegTech products across the whole population of products in our RegTech data platform, Radar versus the number of products incorporating AI (AI Products).
There are some key categories within our taxonomy where AI features most heavily - largely within financial crime, cybersecurity, and conduct risk. We will focus on these and highlight a range of use cases and broader perspectives on these technologies.
As discussed in detail previously, AI is a wide term, encompassing various techniques such as machine learning (ML), deep learning (DL), and natural language processing (NLP). The five focus areas in this blog include products that exhibit these techniques:
ML: Unlike traditional AI systems which may rely on hard-coded rules, ML discovers patterns in data to generate insights or make predictions.
DL: While traditional ML models often require manual data preprocessing and feature engineering, DL models learn these features directly from data, typically needing vast amounts of data and computational power.
NLP: NLP strives to interpret the complexity of human language, such that systems can interact coherently with them. While earlier NLP systems were rule-based and leaned heavily on linguistic knowledge, modern NLP increasingly relies on ML and DL, leveraging large datasets to understand language patterns, semantics, and context.
Focusing on anti-money laundering, fraud detection, counter-terrorism financing legislation, and other facets of financial crime, this is the largest category tracked in Radar, with over 600 products, whilst also accounting for a disproportionately high amount of AI products. The use of AI in this category is driven by:
The need to detect complex patterns from high volumes of data;
The challenge of facing sophisticated criminal networks;
Significant regulatory scrutiny;
Competition and the need for product differentiation in a crowded market.
Firms subject to anti-money laundering and counter-terrorist financing regulations must perform checks on all customers to assess their risk of financial criminality, a process known as Know Your Customer.
Machine Learning
Sedicii's KYCexpert is a cloud-based onboarding solution, capturing documents and cross-referencing beneficial owners against databases, and leveraging ML to verify the authenticity of identity documents including live verifications.
Deep Learning
Nanonets offers an Optical character recognition (OCR) solution optimised for data extraction from various identity documents, powered by a DL model that has been pre-trained on millions of documents. Through an API, users can train OCR models with their own data without delving into code specifics or GPU concerns.
Natural Language Processing
Over Watch offers an identity-matching solution for CDD/KYC, using NLP to handle 18 languages, comprehending linguistic norms, alternative names, and international lexicons.
These products allow firms to screen all customer activity - from account set-up through to transaction activity to identify many types of fraud that the financial system is exposed to.
Fraudio's payment fraud detection tool harnesses 3rd generation machine learning, leveraging vast networked datasets containing billions of transactions. Through their API, Fraudio offers instant access to a multitude of supervised and unsupervised ML models, enabling real-time fraud assessment.
Feedzai's RiskOps platform approaches financial crime prevention by integrating identity management, real-time data analytics, and collaborative measures across teams. Incorporating DL from their research team, this platform digests a large number of diverse data points, and outputs a unified view of customer data, behaviours, and activities, supporting use cases ranging from transactional to behavioural screening.
Inscribe targets lenders and insurance companies, harnessing the combined prowess of NLP and computer vision to scan and verify the authenticity of documents like bank statements, pay stubs, and tax forms. With models trained on hundreds of millions of data points, Inscribe’s fraud detection software can detect signs of digital tampering, evidence of forgery, and the use of templates purchased online.
Regulated firms are obligated to monitor all transactions to identify suspicious criminal activity and must report this activity to Financial Intelligence Units to facilitate law enforcement investigations.
DX Compliance utilises ML to analyse patterns in customer transactions, aimed at reducing the typically high rate of false positives seen in AML cases. Their models assess risk in real-time, with multiple dedicated servers prioritising speed. The platform also provides test modes that allow for experiments with new scenarios and rules.
FNA uses DL, specifically neural networks, to analyse transactional data at a system level. Their technology allows for the discovery of hidden connections in large datasets, simulation of stress impacts on financial networks, and optimisation using proprietary algorithms.
Federated Learning
Federated learning is an ML approach where models are trained across multiple devices or servers, retaining data locally rather than centralising it. Consilient employs this method in the realm of anti-money laundering and financial crime prevention. Their system lets institutions share behavioural insights without directly sharing data, enhancing risk management. Models are trained locally at each organisation using a privacy-protected method, then aggregated on Consilient’s platform to produce an optimised ‘champion’ model.
With an increasing threat of cyber-attacks, and a greater awareness and appreciation of the value and sensitivity of personal data, there is an emphasis among firms to strengthen their defences. Breaches not only result in fines but can quickly become high profile, resulting in wider reputational damage. As a tech-native space, the potential for complex and robust AI adoption is high.
Cyber Security solutions allow regulated institutions to protect and defend their systems, networks, and data from criminals, who leverage digital technologies to implement advanced threats and targeted attacks.
Cyber GRX leverages ML to identify and prioritise third-party cyber risks. Their AIR Insights™ and Predictive Risk Profiles employ ML algorithms to automate the traditionally manual process of gauging inherent risk, sourcing data from their Global Risk Exchange.
Blue Hexagon utilises DL for continuous threat detection and response. Their platform collects and analyses raw data from cloud infrastructures, employing DL models to detect both known and unknown threats. This approach also offers predictive insights into potential threat behaviours.
Abnormal Security employs NLP and Natural Language Understanding (NLU) to analyse email content for fraudulent topics, sentiment, and communication patterns. The platform learns typical business communication behaviours and relationships, making it adept at spotting anomalies, especially from AI-generated attacks. Integration with various platforms like M365 and Okta further refines the detection process by understanding multifaceted behavioural signals.
Conduct
The Conduct category looks at how firms can comply with the regulatory requirements that ensure markets are run in a fair, efficient, and transparent manner. This means firms must be able to detect, prevent and monitor suspicious activity indicative of market manipulation and abuse. A recent wave of US fines for off-channel communications and poor recordkeeping practices highlights how seriously regulators view the need for investor protection and fair conduct. Specific challenges include:
Establishing and enforcing policies that govern which channels are to be considered legitimate for professional conversations;
Integrating with an expanding and rapidly evolving set of communication platforms and channels;
Automating the analysis of conversations, including the detection of suspicious interactions even among nuanced or coded terms.
5. Ecomms Surveillance
These products collect and use electronic comms data (Ecomms) such as email, chat, and voice messages, for the monitoring and surveillance of market abuse, suspicious trading, and transaction patterns including the identification of rogue traders.
Shield FC’s platform incorporates multiple AI algorithms to provide comprehensive surveillance of all communication data across different formats. Their system delves into context, semantics, sentiment, and intent to improve the understanding of each data point for potential compliance risks. The platform's ML capabilities adapt and improve based on usage and feedback.
Recordsure's ConversationReviewAI leverages deep neural networks for post-conversation assessments, designed specifically for in-depth reviews. By focusing on domain-specific language and segmentation, it aims to identify high-risk cases, in highly targeted reviews. The platform offers the ability to classify topics within customer interactions, providing a visual representation that aligns with review checklists.
The Fingerprint Supervision Platform continuously ingests and standardises communications across multiple channels. Using NLP and ML, it identifies risks based on predefined lexicons, generating and ranking alerts for compliance teams. All data is securely archived to meet global recordkeeping requirements.
There's a significant piece of the AI picture we've yet to delve into. Generative AI and Large Language Models (LLMs) represent a frontier of AI innovation, largely responsible for the renewed and growing interest in AI, carrying profound implications for RegTech and beyond. We expect the RegTech industry to be among the early adopters of this technology too and forecast that it will sustain the growth of RegTech for a considerable period.
About Michael Lawrence
With a background in economics and computer science, Michael has been immersed in RegTech since 2017. His experience spans financial services, including roles in credit risk—where he implemented an award-winning machine learning model—and data transformation, setting up global data pipelines. Michael has also contributed as a researcher, developing machine learning models for the UK Government. Today, he serves as a senior research analyst and product manager at RegTech Associates, creators of Radar, the online RegTech marketplace.
About RegTech Associates
Radar tracks over 1700 RegTech and RiskTech products, hundreds of which are already using AI techniques. If you want a closer look at the data, sign up for free today!
Have you got an exciting AI use case? If you would like to feature in our upcoming blogs, contact us at radar@rtassociates.co
Want to list or update your product on Radar? Here's how.
Curious about how AI and RegTech might reshape your business? Drop us a line at info@rtassociates.co
Every day we send out a free e-mail with the most important headlines of the last 24 hours.
Subscribe now