Navigating the Risks of AI in Finance: Data Governance and Management Are Critical

Regulators are cognizant of the disruptive impact and security threats posed by weak data governance (DG) and data management (DM) practices in the investment industry. Many investment firms are not developing comprehensive DG and DM frameworks that will keep pace with their ambitious plans to leverage new technologies like machine learning and artificial intelligence (AI). The industry must define legal and ethical uses of data and AI tools. A multidisciplinary dialogue between regulators and the financial industry at the national and international levels is needed to home in on legal and ethical standards.

Steps Toward Data Efficiency and Effectiveness

First, establish multiple and tangible goals in the short-, mid-, and long-term. Next, set an initial timeline that maps the effort in manageable phases: a few small pilot initiatives to start, for example. Without clear targets and deadlines, you’ll soon be back to your day-to-day jobs, with that outdated refrain from the business side, “The data governance and management thing is IT’s job, isn’t it?”

It is extremely important to begin with a clear vision that includes milestones with set dates. You can think about how to meet the deadlines along the way. As you are defining and establishing the DG and DM processes, you should think about future-proofing systems, processes, and results. Does a specific data definition, procedure, and policy for decision-making tie back to an overall company strategy? Do you have management commitment, team involvement, and clients?

As I pointed out in my first post on this topic, organizations having the most success with their DG and DM initiatives are those that take a T-shaped team approach. That is, a business-led, interdisciplinary technology team-enabled partnership that includes data science professionals. Setting realistic expectations and showing achievements will be essential disciplines, because DG and DM frameworks cannot be established overnight.

Why are DG and DM Important in Financial Services?

For investment professionals, turning data into complete, accurate, forward-looking, and actionable insights is more important than ever.

Ultimately, information asymmetry is a great source of profit in financial services. In many cases, AI-backed pattern recognition abilities make it possible to acquire insights from esoteric data. Historically, data were mainly structured and quantitative. Today, well-developed natural language processing (NLP) models deal with descriptive data as well, or data that is alphanumerical. Data and analytics are also of importance in ensuring regulatory compliance in the financial industry, one of the world’s most heavily regulated areas of business.

No matter how sophisticated your data and AI models are, in the end, being “human-meaningful” can significantly affect the users’ perception of usefulness of the data and models, independent of the actual objective results observed. The usefulness of the data and techniques that do not operate on “human-understandable” rationale are less likely to be correctly judged by the users and management teams. When intelligent humans see correlation without cause-and-effect links identified as patterns by AI-based models, they see the results as biased and avoid false decision-making based on the result.

Data- and AI-Driven Initiatives in Financial Services

As financial services are getting more and more data- and AI-driven, many plans, projects, and even problems come into play. That’s exactly where DG and DM come in.

Problem and goal definition is essential because not all problems suit AI approaches. Furthermore, the lack of significant levels of transparency, interpretability, and accountability could give rise to potential pro-cyclicality and systemic risk in the financial markets. This could also create incompatibilities with existing financial supervision, internal governance and control, as well as risk management frameworks, laws and regulations, and policymaking, which are promoting financial stability, market integrity, and sound competition while protecting financial services customers historically based on technology-neutral approaches.

Investment professionals often make decisions using data that is unavailable to the model or even a sixth sense based on his or her knowledge and experience; thus, strong feature capturing in AI modelling and human-in-the-loop design, namely, human oversight from the product design and throughout the lifecycle of the data and AI products as a safeguard, is essential.

Financial services providers and supervisors need to be technically capable of operating, inspecting data and AI-based systems, and intervening when required. Human involvements are essential for explainability, interpretability, auditability, traceability, and repeatability.

The Growing Risks

To properly leverage opportunities and mitigate risks of increased volumes and various types of data and newly available AI-backed data analytics and visualization, firms must develop their DG & DM frameworks and focus on improving controls and legal & ethical use of data and AI-aided tools.

The use of big data and AI techniques is not reserved for larger asset managers, banks, and brokerages that have the capacity and resources to heavily invest in tons of data and whizzy technologies. In fact, smaller firms have access to a limited number of data aggregators and distributors, who provide data access at reasonable prices, and a few dominant cloud service providers, who make common AI models accessible at low cost.

Like traditional non-AI algo trading and portfolio management models, the use of the same data and similar AI models by many financial service providers could potentially prompt herding behavior and one-way markets, which in turn may raise risks for liquidity and stability of the financial system, particularly in times of stress.

Even worse, the dynamic adaptive capacity of self-learning (e.g., reinforced learning) AI models can recognize mutual interdependencies and adapt to the behavior and actions of other market participants. This has the potential to create an unintended collusive outcome without any human intervention and perhaps without the user even being aware of it. Lack of proper convergence also increases the risk of illegal and unethical trading and banking practices. The use of identical or similar data and AI models amplifies associated risks given AI models’ ability to learn and dynamically adjust to evolving conditions in a fully autonomous way.

The scale of difficulty in explaining and reproducing the decision mechanism of AI models utilizing big data makes it challenging to mitigate these risks. Given today’s complexity and interconnectedness between geographies and asset classes, and even amongst factors/features captured, the use of big data and AI requires special care and attention. DG and DM frameworks will be an integral part of it.

The limited transparency, explainability, interpretability, auditability, traceability, and repeatability, of big data and AI-based models are key policy questions that remain to be resolved. Lack of them is incompatible with existing laws and regulations, internal governance, and risk management and control frameworks of financial services providers. It limits the ability of users to understand how their models interact with markets and contributes to potential market shocks. It can amplify systemic risks related to pro-cyclicality, convergence, decreased liquidity, and increased market volatility through simultaneous purchases and sales in large quantities, particularly when third party standardized data and AI models are used by most market participants.

Importantly, the inability of users to adjust their strategies in times of stress may lead to a much worse situation during periods of acute stress, aggravating flash crash type of events.

Big data-driven AI in financial services is a technology that augments human capabilities. We are living in countries governed by the rule of law, and only humans can adopt safeguards, make decisions, and take responsibility for the results.

References