Regulators: the need for Explainable and Auditable AI

In financial institutions the use of models which can be interpreted and verified by auditors, examiners, and regulators is an absolute necessity. The application of this model in a black box manner would not reduce but rather increase the number of false positives or miss actual suspicious activities, false negatives.

A growing worldwide problem

Compliance and regulatory (CoRe) risk has become one of the greatest challenges for financial institution executives and boards of directors. There are a growing number of overarching risks in today’s financial services environment: globalization and its impact on political, economic and operations processes; financial practices with significant built-in risk, such as sophisticated, just-in-time treasury and cash management; online banking and the risk of exposing customer information and accounts to unauthorized parties; risks created by outsourcing selected functions and tasks to third and fourth parties; and more.


Looked at in isolation, a regulation is a relatively simple affair, a legal document containing text that describes what needs to be done, by whom, when, and (sometimes) how. With some understanding of the underlying topic, a compliance officer can read the document; understand what is mandated and where it will affect his or her part of the organization.Then he or she can determine what is required in order to ensure compliance, what is required and how to demonstrate that compliance is met, not only to his management but also to the regulator. Of course things just aren’t that simple, this approach doesn’t scale easily and yet the scale and scope both of regulations and of the businesses of firms themselves continues to grow apace. In the real world, firms struggle to understand what legal and regulatory requirements they face everywhere they do business. Inevitably, they struggle to ensure compliance everywhere and are unable to demonstrate it to management and regulators, resulting in compliance failures, regulatory fines and, increasingly, personal legal sanctions for their management.

The limitation of automation

The problem is that, for each legal or regulatory text, someone has to read it, analyze it, understand the impact on their organization, and then undertake and manage whatever actions are needed to ensure compliance. This task is multiplied for each regulation issued by each regulator, in each jurisdiction and for every line of business. As markets, and ultimately firms, are evolving, they can end up having to comply with thousands of regulations from dozens of regulators. Even if this mammoth task is achieved that is not the end of it: Regulations change, their interpretation changes, and of course the firm itself changes. Firms have to keep up with all of this change. A medium-sized firm may have to scan hundreds of updates every week, identifying which ones affect regulations that contain requirements that affect them and then deciding what, if any, action is required in order to ensure continuing compliance. And the broader the business and product offering, the more complex the regulatory landscape they have to adhere to, becomes.This is a process that cries out for automation but both the regulations and the updates to them are in the form of unstructured documents that have to be read, interpreted and contextualized by skilled and experienced staff.

The promise of AI

Artificial Intelligence isn’t a single technology, it is a collection of related technologies, including Vision and Perception, Robotics, Speech, Natural Language Processing (NLP) and machine learning. Each of these technologies has specific uses and Natural Language Processing in particular is starting to come into widespread use in helping to analyze unstructured content such as laws and regulations. Together with machine learning, NLP solutions can read such documents and perform a range of tasks including: extracting metadata, identifying entities and relationships that are referred to, and the intent or purpose of specific parts of the document. For Regulatory Compliance this means that there is the promise that we can use NLP to:

Extract Metadata

this helps us to understand what the regulation is about by identifying financial products (e.g. loans or swaps), regulatory topics (e.g. anti-money laundering or market abuse) and business processes (e.g. trade settlement or customer due diligence). With this information, it becomes possible to determine whether the regulation is relevant, what parts of the organization are likely to be affected and who needs to review it.

Identify Identities and Relationships

Entities provide the ‘who’ in a regulatory document. Who is the document addressed to (perhaps the firm), by whom (a regulator) and who are the other actors (customers, other market participants, etc.)?

Natural Language Understanding

In Regulatory Compliance, it is vital to understand what requirements or obligations are contained in a law or regulation. These are the parts of the text that tell firms what they must do or must not do. NLP is able to help us to identify the requirements that are contained within a document and, using the entities and metadata, determine who they apply to and what products, topics and processes they refer to.

Cognitive Process Automation

Linking these processes to another AI technology, like machine learning, means that we can train systems to get better at these tasks, further increasing their utility.

Sensemaker – The Explainable AI Machine

As AI gets more and more complex there is an increasing need to understand the “how and why” behind every prediction a model makes. Quantiply, understands the need and importance for such transparency. We’ve carefully considered the needs of the key personas in the AML ecosystem, starting from the Regulators, to the Executive level, to the Case Analysts, all the way to the bank’s customers, to develop a principled approach to XAI. By keeping in mind, the wants and concerns of our audiences we’ve developed a Consistent, Repeatable, Auditable, Fair and Transparent (CRAFT) approach to XAI. Let’s dive in to understand why each attribute is important:


we take great measures to understand the quality of our predictions and tie it to the patterns the model is learning


If a prediction is made on Day 1 or Day 100 using the same data, it is extremely important that we see the same prediction being made by the model


By carefully defining reward-punishment policies we can ensure that our models are always held accountable


In AML, it is not only important to make accurate predictions but it’s important to make fair decisions


If a model makes fair and consistent predictions or even if its predictions are repeatable, none of this information can be verified unless there is transparency

Improved control efficiency

improved understanding of the model control framework will allow identification of potential duplications or inefficiencies

Fighting Financial Crime with Artificial Intelligence

Mitigate risk against damage to reputation, client trust, market share, and reduce false positives by more than 50%