AI is helping banks reduce compliance costs
(Maybe it can help your industry too)

AI’s ability to understand human language as well as numbers means that it is ideally suited for complex compliance monitoring tasks.

AI is already helping banks save billions in compliance costs

We’ve provided examples of what AI is and what it can do for all of us. There can be no doubt that AI has extraordinary potential for solving an almost infinite variety business, social, and simply human problems. But if we narrow down our field of view to just the professional challenges faced by Legal and Compliance Leaders in their daily work, we find that here too AI can be of service.

A lesson from the financial services industry

Consider for example the Financial Services industry, where AI is already being widely deployed. Perhaps no other industry faces more regulations or higher compliance costs than this one. And with good reason. Banks and insurance companies are the indispensable pillars of the world’s financial system, upon which we all depend. It is essential that government regulation of this sector be as thorough as possible.


It’s hard to put an exact dollar amount on the cost of compliance for the global financial industry. But according to a recent Federal Reserve report discussing the implications of AI for financial regulation, U.S banks alone are estimated to spend over $70 billion per year for regulatory compliance.


The idea of using information technology to reduce these costs is not new. But it has recently been given a big boost by a host of new technologies including AI, cloud computing, and blockchain. For a good short overview of this topic, see this report from the Institute of International Finance (IIF).


A new word has even been coined to describe the application of technology to compliance in the Financial industry: RegTech. IIF defines RegTech as “the use of new technologies to solve regulatory and compliance requirements more effectively and efficiently.”

In one example of RegTech cited by the Federal Reserve, AI is helping Citibank to pass the Fed’s capital adequacy stress tests. But capital adequacy requirements are not the only kind of regulation that AI is helping banks to meet. An even bigger area is monitoring of trading activities for misconduct and abuse. The Bank of England estimates that misconduct by traders has cost banks a global cumulative of $320 billion to date. This explains why banks are aggressively deploying machine learning to monitor the behavior of their traders and detect unusual behavior.

AI can understand the unstructured data needed for compliance monitoring

Unlike traditional statistical methods, AI today can understand language as well as numbers. For example, it can look at the huge volume of unstructured data that accompanies trading—text messages, emails, online calendars, even phone calls. Combining both structured and unstructured data and feeding it to AI, banks can decide in near real-time whether a trader’s unusual actions are merely a bold but legitimate new trading strategy, or warrant referral to the internal compliance team for further verification.

Financial industry RegTech based on AI is just getting started. But given the enormous costs of regulatory compliance in this sector, it’s a safe bet that it has a big future. Compliance decision makers in other industries should take note and learn what they can from the money industry’s pioneers. Your compliance costs may not be as high as Citibank’s, but they are almost certainly high enough to justify a close look at using AI and machine learning to lower them.

Reaching human parity

Much progress that has been made by computers in understanding the world and in fact, in many areas, computers have met and exceeded human capabilities.

AI is already closing in on human performance in basic language and image processing tasks:

  • 96% of image recognition parity with humans

  • 95% of speech recognition parity with humans

  • Continual progress in translation – Microsoft Translator can translate into 62 human languages and Klingon!

  • Almost 90% of reading parity with humans

reaching human parity 2016-17
reaching human parity 2018