Scalable Infrastructure for AI in Banking: Bridging Cloud Computing and Regulatory Demands

Abstract

Over the past few years, the financial services industry has been buzzing with buzzwords around artificial intelligence (AI). Banks’ media releases profess their cover to be on the cutting edge of AI in order to make better use of customer data, understand customer needs and preferences, and deploy responsive products and services. AI’s applications in banking include management of risk (credit, market), compliance with AML regulations, cyber fraud detection, marketing communications, workforce optimization (scheduling of tellers), and treasury functions (fund and liquidity management). Most of these applications involve process automation, allowing for the work to be done more efficiently, but they do not fundamentally change the mode of operation of banks. The rise of better and noisier predictors in the history of financial markets has historically been met with restrictions and disqualifications so as to uphold a level playing field. As financial services institutions continue to explore ways to enhance their existing predictive models using ML technology, regulators and industry standards boards also take interest in the challenge of ensuring that ML models champion the same principles and objectives as their predecessors.

A long list of AI governance challenges stimulated the work described here. Their chronic presence as detected by too few FRTB, AML and other governance abuses would seem to call for new solution approaches. This work presents a high-level AI system framework and modular building blocks that map emerging self-regulation practices and tooling to assist efforts to mitigate governance risk in financial services. Focus is placed on the design of tooling for financial services ML models for autonomous, ongoing operations. A financial services firm’s AI systems ought to abide by a host of regulatory and ethical principles, requirements and obligations akin to the algorithmic proxies employed to execute trades.

To that end, building blocks are developed that are suggested for integration into the AI systems of banks and other financial institutions in order for them to assure compliance with regulation and upholding of ethics through self-regulated behaviour. Building blocks are defined broadly in that they aim at reusable, off-the-shelf automation scripts that allow for the independently and easily executable instantiation of governance processes tailored to model and regulation specifics [2]. The rooting of the tooling and best practices in the governance risks faced in the financial services industry, and in their consequences, might also provide a step towards prioritizing and demystifying the immensely daunting challenge of AI governance for firms across industries.