長い道のりの第一歩

Early June 2024, AWS invited Celent to join them for an analyst day that, to no one’s surprise, was dominated by the application of AI, and particularly Gen AI, in financial services. The overall message is that companies across all industries are pressing ahead with developing and scaling Gen AI tools and embedding AI, in general, more deeply into their businesses. This mirrors a lot of conversations that we have had recently, be it face-to-face at our sell side round table, through conversations with industry leaders, or through surveys (FIs / Vendors) we conducted.

Although I heard from many interesting solutions providers and projects, one conversation stuck in my mind in particular: talking to Nasdaq’s Head of Regulatory Strategy and Innovation, Tony Sio. Using Amazon Bedrock, Nasdaq saw an opportunity to use Gen AI to reduce the repetitive tasks of market surveillance teams. This would alleviate some of the continuously growing strain on the teams and empower companies to meet their regulatory obligations.

In this instance, rather than the surveillance analysts finding and reading through all pertinent information, the gen AI model will search, screen, and synthesize documents, filings, and news sentiment on companies, sectors, and peers and present them to the analyst, highlighting possible aggravating or mitigating factors. The analyst will then decide on whether to investigate further. According to Nasdaq’s tests, and despite RPA and ML being applied to this area for some time, the inclusion of Gen AI reduced the time analysts require to complete the task by a minimum of 30 % but have examples of much greater reductions.

Solving a business problem with Gen AI is not that groundbreaking and has been done a few times now. However, what I thought was more interesting about this application was that it was integrated into one of the most sensitive areas of the capital markets technology stack. Compliance investigations must be extremely transparent, tightly guarded, and accurate, making the process resistant to change. Regulators want insight and assurances that nothing is missed, want to be able to hold people accountable when it is, and make sure that no one that is being surveilled knows how the system works. As a result, this layer of the tech stack is often isolated and heavily scrutinized by risk and legal departments as well as regulators.

In a previous conversation with a Gen AI leader at a Tier 1 FI, they made a poignant observation. Each technology has its flaws that need to be overcome. For Blockchain, that is, for the benefit to be realized, you require succeeding in creating a network effect; in generative AI, there are inherent risks that cannot completely be removed. These risks, namely accuracy, explainability, and hallucinations, are not unique to compliance. These are especially acute in regulatory and compliance functions, given the high cost and public nature of failure in this department. Even the most prominent of banks are not immune, as JPMorgan, Goldman Sachs and Citibank found out in recent years. As these risks cannot be completely mitigated through technology, the implementation issues are not purely technological, but as Sio confirmed, more on the operational and risk side. The biggest hurdle was integrating a system with risks so deep into the surveillance process. So much so that to get approval from sec ops, Gen AI’s ability to make decisions and decide on what pertinent information is had to be reduced to zero, making the risk functions comfortable with its integration.

Although the impact of the current Gen AI app is already significant, limiting the models in this way has the effect of limiting the impact that Gen AI is having on today’s business. However, this initial integration is a necessary first step in a long process for any organization to harness the full potential of Gen AI. Sio sees more opportunities to enhance productivity in the future once the regulatory framework and risk departments are more comfortable with the fallacies of this technology, even in critical areas such as market surveillance.

Celent is in full agreement with this and is encouraged to see Gen AI being utilized across the regulatory tech stack so early in the Gen AI ecosystem development. Our recent survey of FIs showed that operational buy-in was the main success factor when implementing Gen AI for capital markets participants, and Celent sees the operational risk department as instrumental in that process. Deep collaboration between risk, technology, and business will be necessary to make the technology a success. With technology, such as Gen AI, which has some flaws but lots of promise, small steps towards a medium-term goal, like Nasdaq is taking here, will be necessary to reap even larger long-term benefits.