Successful Integration of Generative AI in Banking Requires Both Vision and a Healthy Dose of Pragmatism

47

By Nigel Moden, Global and EMEIA Banking and Capital Markets Leader, EY

Across the European banking sector, leaders recognise the potential of generative artificial intelligence’s (GenAI’s) large language models (LLMs) to transform the industry—from customer support and experience to internal operations—and are actively assessing opportunities to realise productivity gains.

Yet, alongside the ambition to deliver truly transformational outcomes for customers, a healthy dose of pragmatism is required to succeed from the outset. Identifying and clearly defining early-phase deployments and ensuring that firms are focusing on use cases that not only build momentum but drive commercial results are key.

Establishing a clear and targeted approach to deploying generative AI (GenAI) in specific banking operations is critical to laying the foundations for a large-scale transformation of the banking industry in the long term.

Where are Europe’s banks on GenAI?

EY’s 2023 “EY European Financial Services AI Survey”—which canvassed the views of executives from 60 financial institutions across the region—evidenced the excitement, optimism and sense of opportunity shared by leaders sector-wide as they assessed the road ahead for artificial intelligence (AI) and GenAI adoption across their organisations.

Nearly two-thirds (61 percent) of European banks told us they had invested in GenAI applications in 2023, and 78 percent stated they planned to increase their spend over the year ahead.

Yet, financial-services executives—banking leaders among them—also highlighted the complex, multidimensional challenges that the sector is already navigating through GenAI adoption: maximising human capital and experience, enhancing business efficiency and mitigating emerging risks.

However, only 52 percent of European banking executives deemed their organisations to be on par with industry peers in their adoptions of AI. Nearly a third (30 percent) believed they were behind the curve, and one in ten (9 percent) had not established any AI-integration plans. This demonstrates the journey yet to be undertaken by the sector just to keep up.

Proactively pragmatic

One of the primary challenges banking leaders face in implementing GenAI is understanding what its true capabilities are. This means banks must build an understanding of AI’s potential applications; determine which ones can unlock real value, momentum and productivity gains; and fine-tune data and applications to achieve the differentiation that will underpin the next phase of GenAI integration across the sector.

The potential use cases of rapidly evolving models are vast and, typically, not built for specific applications; GenAI models require fine-tuning in most, if not all, scenarios.

This fine-tuning process can vary widely in terms of timelines. Some use cases can start delivering value incredibly quickly, while others may require longer phases of development and testing. The key differentiator is value—in the context of the business’s wider strategy, profit and loss (P&L) and commercial priorities, but also, crucially, in terms of customer impacts. Through these lenses, thousands of potential use cases can be reduced to a handful with the potential to be truly transformative. But firms must retain optimism and continue to sift out those handfuls of transformative use cases, as they could be the keys to finding a competitive edge in a crowded market. Several of the use cases being explored in the banking sector currently include detecting fraud, managing risks, automating processes and tailoring customer experiences.

The decision to invest in and scale a specific use case requires a strong degree of confidence in potential outcomes. Some use cases will take six months to derive value through implementation, while others may require patience for a multi-year time horizon.

Identifying these use cases and deploying the technology requires a deep understanding of both GenAI and the bank itself. The (lack of) availability of this talent is presenting a headwind to progress. EY’s 2023 “EY European Financial Services AI Survey” found that 30 percent of banking executives believed there was limited scope within their current workforces to implement GenAI. At the other end of the spectrum, just 4 percent considered themselves fully equipped from a personnel perspective, reinforcing the talent challenge looming over the sector through this next phase of transformation.

To address the shortfall, 22 percent of executives divulged that they already had AI training and upskilling in place for targeted groups, although 30 percent reported that training development was in its infancy or that they had only just started to consider what AI training was required. Training will be essential as technology evolves, both upskilling current employees and instructing the next generation of talent.

GenAI will never compensate for bad data

Alongside clearly defining use cases and upskilling talent, what else do banks need to be thinking about? With spending on GenAI technologies already widespread, the opportunity for differentiated advantages lies in the datasets fed into the technology.

Training private datasets within models is key to unlocking differentiation, but it needs to be achieved in a way that’s secure, understandable and usable. Of course, achieving this in the lab or through testing is just the first step to implementing models securely and privately across entire data networks.

These foundational steps will enable banks to leverage and benefit from GenAI as their integrations evolve.

As the key to differentiation, data is a bank’s most important asset. However, many banks have not yet fully built 360-degree pictures of what their data shows or consolidated and organised their data in ways that make it usable. GenAI does not—and never will—compensate for bad data; for some, the hard yards to opportunity will be achieved by digging into the data.

Continuous learning                                                                                                                    

Of course, fine-tuning data is only half the challenge. Fine-tuning GenAI models themselves is an ongoing process, and banks cannot delay getting started. While there may inevitably be a trade-off between investing in an existing model versus waiting for the next improved iteration—GPT-6 may perform better than a finely tuned GPT-4/5, for example—the biggest risk is effectively waiting until race day before starting to train.

Again, this requires strong human technical skills to manage. In 2023, two-thirds (68 percent) of financial-services firms expected up to a quarter of all roles to require AI training or upskilling in 2024, while nearly a fifth (17 percent) believed this figure could be as much as half. We continue to regard the lack of GenAI knowledge and experience as a significant commercial risk that will only gather momentum over time.

Action to support productivity gains through training and upskilling is not yet widespread. Last year, 35 percent of Europe’s financial-services firms had no plans in place to train their workforces in GenAI technologies. However, it was clear that some were taking a more focused approach; 12 percent had put training in place for targeted groups, with a further 10 percent developing plans.

It is crucial that firms actively consider the ways in which AI and GenAI technologies could change both the nature of work and how people learn and build experiences throughout their careers. At the same time, leaders must ensure that GenAI integration is not a “bolt-on” initiative, siloed from the wider goals and objectives of businesses, but fully integrated within banks’ wider strategies.

Next steps

Defining a clear approach to embedding and deploying generative AI across banking operations is the critical first step for leaders seeking to build confidence at scale. From there, understanding and identifying what’s possible and what will drive the most value is essential and will set the direction for future progress—and ultimately ensure that the datasets that provide the most differentiation are those that are developed and built upon.

AI is the key to creating more efficient and effective banking platforms and has the capacity to accelerate and transform operations. The banks that fully commit and dedicate resources to developing AI technologies will almost certainly create competitive advantages and be at the cutting edge of banking technology and the industry as a whole.