Artificial Intelligence (AI) applications are becoming much more common everywhere, for consumers to research and explore online resources, for developers to more easily write code, and in customer service scenarios to (hopefully) improve responsiveness and enhance support solution development. AI also offers broad value within various areas of financial institutions’ (FIs) operations and client service departments. In fact, according to industry technology leader NVIDIA, “The financial services industry is undergoing a significant transformation with the adoption of AI technologies.” Finextra has featured extensive research on AI usage in financial institutions through numerous webinars and interviews, and readers can find additional insights from our most recent webinar on the topic by visiting our website.

Surveys show most financial service companies already seeing the promise of AI In the company’s fourth annual State of AI in Financial Services Report, which surveyed more than 400 global financial services professionals, from executives to “data scientists, developers, engineers, and IT specialists”, NVIDIA reports that “91% of financial services companies are either assessing AI or already using it in production. These firms are using AI to drive innovation, improve operational efficiency and enhance customer experiences.”

Finastra, a leading global provider of open finance and banking software, released its own report, Financial Services State of The Nation Survey 2023, that focused on recent trends in AI usage within the industry. Its results, from responding firms in nine countries (France, Germany, Saudi Arabia, UAE, Singapore, Hong Kong, and Vietnam, in addition to the U.S. and UK) showed that more than a third of financial institutions have improved or deployed AI technology within the past twelve months, and 83% of decision makers surveyed expressed their institutions’ interests “in Gen AI specifically”, with more than a quarter of all respondents having already implemented the technology in some form within their institutions.

Gen AI rapidly gaining traction in operational and service applications for FIs

Consistent with what Finextra has shared through our webinars, articles, and expert interviews, the key areas of focus for AI in financial services are not just in client service management, but also portfolio optimisation, risk management, fraud protection and more. Meanwhile, the NVIDIA report stated “generative AI (Gen AI) is quickly gaining popularity with organisations keen to uncover new efficiencies.”

The US Federal Reserve has conducted its own industry survey (on the usage, planned deployment, and potential risks of AI and its related machine learning modelling tools) in early 2021. However, though the results aren’t out yet, comments from firms and individuals replying to the Fed’s request for information seem to confirm the findings of NVIDIA and several of the major consulting firms: “Gen AI” is rapidly finding use cases to support its implementation in the financial and other industries, and via large language models – defined in the NVIDIA report as comprising “a class of deep learning architectures called transformer networks. A transformer model is a neural network that learns context and meaning by tracking relationships in sequential data” – and these are quickly coming into practice as well.

What’s here now, what’s on the horizon, and what bears close watching by FI managers?

Finextra asked experts on AI and machine learning from the financial services field to share their thoughts on a number of questions related to the technology and its current and future use cases within the industry. According to Adam Lieberman, Head of Artificial Intelligence & Machine Learning at open finance and banking systems provider Finastra, the rise of Chat GPT as a “mass consumption” natural language model in 2023 “marked the era of generative AI (Gen AI) democratisation”, and he’s bullish on the future of GenAI in numerous use cases for financial services going forward.

Lieberman highlighted three use cases in particular that he believes banks will focus on in 2024:

• Analysing data for ESG criteria classification,
• Collecting and processing data for Know your Customer (KYC) and Anti-Money Laundering (AML) purposes, and
• Strengthening their fraud detection systems

“Gen AI”, says Lieberman, “analyses customer behavior, discovers preferences, and has the power to unlock important insights, leading to more personalized user experiences and improved customer service.” Using examples from the industry, The Atlanta-based technology leader notes the increasingly aware and able response tools available. “We are already witnessing a shift from "traditional" chatbots to sophisticated ones that are highly trained in semantics and can provide knowledgeable responses.” As examples, he cites the technology’s capabilities to “not only generate immediate, precise, and natural-sounding answers, but it can also analyse a customer’s financial history and credit rating and help them source and choose the best loan option for their unique situation. “

John DesJardins, a Chief Technology Officer and solutions advisor with deep and varied experience working for some of the financial industry’s largest software providers, also sees greatly enhanced chatbots on the horizon. “Many clients, particularly younger ones, may prefer to interact via chat as that is easier to do while also working or while watching TV or surfing the web, etc.”, he says. DesJardins also envisions AI “assistants” helping staff to more quickly access key information for clients, and more “intelligent” robotic process automation (RPA) with AI added to the mix. “RPA may finally hit a level of usability with AI enhancements that will make it more valuable to enhance workflows and automate some decisions.” DesJardins feels efficiencies in time and focus will result. “This will be particularly relevant to routing clients more quickly to the right person to assist them and also in pulling together and even summarising critical context for that customer to improve the level of service given - and reduce frustrated customers due to waiting on hold, getting transferred, etc.”

Asked to name his top targets for AI deployment in financial service, DesJardins notes another strong use case being “Intelligent Advisors” that allow solutions to be “easier to build” by financial providers and “able to consider larger context windows and incorporate related data.”

Citing advances in the latest Large Language Models (LLMs) and more fine-tuning functionality now available, DesJardins says “usability and interactivity” will be enhanced to help in analysing spending patterns in a contextual way for all types of customers, from retail up through mid-sized business and including multinational corporate clients, with specific content tailored as needed. He notes that one application for corporate customers might involve helping finance teams “identify unusual expense patterns or cash outflows”, or if aiming at consumers, “identify(ing) subscriptions to cancel, or other opportunities to better manage their spending, (find) unexpected subscriptions or other expenses, etc.”

“The key”, says DesJardins, “is advances in natural language processing (NLP) combined with other traditional AI that can spot trends and anomalies, which unlocks a more complete picture with the contextual descriptive information.” Adding in numerical data and the interactivity that the latest Gen AI brings, DesJardins points out, enables another valuable use case developing “what-if” scenarios for financial planning or other company representatives around tax, investment or insurance pricing and planning exercises.” Hot items for Lieberman include Agentic - AI systems that can pursue complex goals with limited direct supervision – which he asserts will be an area of likely expansion, and to him, “fascinating. This is a new layer of AI that enables us to use large language models and generative AI systems to perform tasks and interact with external environments, rather than just retrieve information. For instance, it can assist us in filing a support ticket, reporting a problem, checking a current stock price, or interacting with already existing APIs (application programming interfaces).”

How can FIs select and prioritise the right (and most secure) AI solutions for their situations?

What should financial institutions aim for in positioning their own Gen AI initiatives? Lieberman says to start with what he calls “a base education layer.” All functions of the organization should be made aware of relevant use cases for their implementation of AI, including the proper technology required. Then, he says the next step is to develop or modify existing AI governance policies to include Gen AI. “Responsible AI is an important concept that should be orchestrated across the organization”, says Lieberman, and “we need to empower developers with the proper tools, infrastructure, and architecture to develop and maintain the lifecycle of generative model solutions.”

What about potential pitfalls of the emerging technology? How can FIs guard against security and data privacy mishaps? Both Lieberman and DesJardins state that ensuring client information will not be misused or improperly shared is job one. Of course, it’s also imperative to recognise and prepare for other challenges surrounding Gen AI, say the two experts, with Lieberman pointing out that “blurred lines of intellectual property (IP), have emerged[..]it's worth noting that some generative AI systems are still in the beta stage and may produce inaccurate or nonsensical answers, resulting in 'hallucinations.' To balance risk and innovation, companies must implement meticulous risk mitigation strategies to prepare for any potential scenario.”

Beyond security risks, providing a positive client experience, and minimizing potential frustration points for customers (“they should always have a quick path to speak directly to a person”) must also be top priorities, emphasized DesJardins, and financial institutions should take pains to ensure “clients don't begin to feel like they and their data are being monetized and that how their data is used is transparent.” Also, he asserted, there are use cases where Gen AI’s large language models may not fit as well, such as low-latency applications like payments monitoring, fraud detection, or algorithmic trading. In such instances, DesJardins says, banks should “use other types of AI and machine learning” tools as a better fit for these specific purposes. With these caveats, he also cautions that the total cost of ownership (TCO) of “computationally expensive” Gen AI processing might exceed cost projections as applications grow in number and scope, and the current lack of transparency and “explainability” around the “black box nature” of AI must be improved, noting that clients and regulators will need assurances. “Today”, DesJardins asserts, “even data scientists struggle to explain how these models work in ways that lay persons can understand.”

Successful financial services AI deployment? Look both outside and in for answers, say experts

What is the outlook for Gen AI in the financial world now, and what are some best practices to make AI work in financial institutions’ and their customers’ best interests? As far as the former, Lieberman says “Gen AI is here to stay, and banks need to minimise the risks and capitalise on the opportunities it offers. In 2024 we will see significant investments in upskilling programs to increase the adoption of AI tools to safely augment human productivity.” DesJardins concurs, reiterating the need for building greater knowledge at all levels of firms servicing multiple types of customers. “First and foremost, leaders should educate themselves on Generative AI trends and how they relate to their area of financial services.

For small or medium sized financial institutions, they may want to wait for their current software providers to offer enhanced products and services related to recent advances in AI.” DesJardins recommends a mix of outsourced and internal expertise to bring it all together. “They (financial institutions) may also want to hire some expertise in house that can work with vendors to implement AI-based capabilities, and (for) larger institutions, they may want to invest in pilots around Gen AI and new use-cases and also do client surveys and other research.”

As with many technologies on the ‘frontier’ of development and adoption, DesJardins comments, “all involved should take what they hear, see, and read with a grain of salt, (with) clear and achievable goals in mind for AI projects and initiatives. They should balance their AI investments with solid thinking around (its) benefits and ensure they focus on solid business cases and ROI. Doing so, he says, “will also ensure they have the answers (when asked) for their investors.”