7 Key Pillars of Modern Risk Management That Are Shaping the Future World of CFOs, CFO News, ETCFO

By: Rajosik Banerjee

In a new era of unprecedented customer experiences, the demand for faster, high-quality financial analysis and forecasting has increased dramatically. Financial institutions, globally as well as in India, are using supervised and unsupervised artificial intelligence (AI) and machine learning (ML) models to manage risk or calculate prices.

However, they always run a risk that a wrong model specification or wrong use of a model will lead to a decision, with material consequences, such as financial and reputational losses, commonly referred to as Model Risk.

Model risk management

To measure and mitigate model risk, institutions should now consider implementing extensive and complex model risk management (MRM) structures. Additionally, with improved customer service and growing cost pressures, they must also consider AI and ML as possible solutions to improve cost and operational efficiency. The benefits of adopting AI/ML models are too powerful to ignore. But at the same time, it is important to have international regulatory guidance on how to manage the specific risks arising from these models.

Let’s look at the seven key pillars that can enable an existing MRM framework to be enhanced based on the specific requirements of AI/ML algorithms.

1. Establish a definition of AI/ML models – The obvious benefits of AI in finance, banking, and business analytics are easy to assess, but difficult to define. Broadly, it is the theory and technology around the development of computer systems or intelligent intelligence algorithms that are not explicitly programmed, but “self-learn” to perform tasks, which would otherwise require human intelligence.

In addition to the new models being implemented, the more widespread use of the AI/ML application may also lead to the replacement of manual processes or simple AI/ML models. Banks should seek to establish an “enterprise-wide definition” of what these models include, beyond its traditional manifestation. Therefore, the inventory of models needs to be expanded to include them.

2. Update Model Prioritization Definition– The prioritization parameters of the models, in particular the criteria of materiality, criticality and uncertainty, might need to be improved to correctly identify the risks inherent in any AI/ML model. Due to the wide range of possible uses, it may be necessary to extend the prioritization of models so that specific AI/ML-like risks, such as reputation or social impact, are properly considered. A new approach is relevant.

For example, the “risk of harming the client”, such as the danger arising from discrimination, must be taken into account. The level of risk can be assessed in terms of likelihood and severity of harm, or a combination of both.

3. Establish an appropriate risk appetite– Traditional risk appetite statements will not work for ML models. Since there is a lack of regulatory guidance in this area, banks should leverage their peer networks to gather industry insights and draft an initial risk appetite statement and thresholds. associates.

4. Identify responsibility– Given that several independent risk management functions will be involved – MRM, Compliance, Data Management and Control, Operational Risk Management (ORM) teams; it is imperative that a cross-functional governance framework be established with clear definitions of roles and responsibilities.

5. Investing in Skills Upgrading – Banks should develop skills internally or call on external experts. External subject matter experts (SMEs) can help them benchmark against peer banks on risk management, controls and governance framework improvements, as well as state-of-the-art model development and validation techniques for ‘AI/ML.

6. Improve the compensating control framework – Existing risk and control frameworks including MRM, data management (including privacy), compliance and operational risk management (IT risk, information security, third party, cyber) do not do not explicitly address the risks as envisaged in the AI/ML models.

They need to be improved by designing additional compensating controls, improvising the existing data management framework, creating better control frameworks around compliance and operational risk, and conducting company-wide training programs. company.

seven. Develop additional tests and procedures for AI/ML models – There are some key elements that need to be specifically tested during the life cycle of the model, including during their design, implementation, operation and validation phases.

New validation approaches address a wide range of elements such as input data, parameter selection, model calibration and improving interpretability and bias removal, developing framework for ongoing monitoring, model implementation and design of challenger models as an effective alternative. Additionally, the concepts of adversarial debiasing, disparate impact removal, calibrated equalized odds processing, fair meta-classifier, and optimized pre-processing are gaining prominence in the context of AI models.

The path to follow

There is a considerable difference in the approach that banks take to manage risks related to bias, interpretability and other challenges. Although some of the global banks are already validating their ML models and some are even investing in AI/ML centers of excellence, others are still at a very nascent stage.

Going forward, it is imperative that CFOs of banks globally as well as in India develop a meaningful understanding of the technology, including its existing and potential uses within their organizations, and have a firm grasp on the implications of AI/ML from a risk perspective. .

Over the next few years, regulatory scrutiny of these models is expected to intensify as banks begin to add more and more AI/ML models to their inventories. International guidance or standards in this area will be helpful in setting the minimum benchmark for MRM practices in all jurisdictions and defining the future of business.

7 Key Pillars of Modern Risk Management That Define the Future World of CFOs

About the Author: Rajosik Banerjee is Partner and Head of Financial Risk Management at KPMG India

Disclaimer: The opinions expressed are those of the authors alone and ETCFO.com does not necessarily endorse them. ETCFO.com will not be responsible for any damage caused to any person / organization directly or indirectly.