Push past proof-of-concept fatigue to untap gen AI’s power

Banks are encountering challenges as they implement the powerful technology in the real world. We offer advice on overcoming hurdles.

This content was originally featured in a BAI by ProSight article in January 2025.

With its promise of innovation and net new revenues, generative artificial intelligence initially seemed a bit like a banking paradise. Instead, early use cases have emphasized productivity—and fallen far short of expectations.

So far, few proofs of concept have made it into production, leading to what can only be described as proof-of-concept fatigue.

What’s causing this stagnation? According to a March 2024 report from McKinsey, 70% of financial institutions use a centrally led AI model, which aims to reduce friction between business units but can slow execution due to required input from these units.

How can banks course-correct and pivot generative AI toward profitability?

The answer is deceptively simple: focus. Amid the frenzy surrounding generative AI, focus is key to operationalizing the technology and enabling the rollout of a consistent flow of high-impact use cases. This approach balances the risks of generative AI with its potential for innovation. It starts with data, then builds an enabling infrastructure and governance organized for the long term. Most importantly, it never loses the human element.

Taking steps to operationalize generative AI

Despite banking leaders’ optimism about generative AI’s revenue opportunities, most outcomes to date have been underwhelming. Organizations face challenges, from inadequate technology infrastructure to talent shortages. Concerns about product stability are persistent, especially given the almost daily release of product updates. In the US, the lack of regulation adds to bank leaders’ apprehension about the safe and secure use of the technology.

While models come and go, generative AI only makes sense when it starts with outcomes and relevant data. The early focus on use cases and experimentation with proofs of concept is shifting as banks struggle with how best to put gen AI into production—and extract business value.

Here are four key aspects to operationalizing generative AI that are paving the way for a continual flow of high-impact use cases:

1. Start with data. When it comes to the availability of data, banks have made great strides in everything from reporting and governance to more informed decision-making. Achieving data maturity, however, remains a struggle—and the lack of progress is an impediment to generative AI adoption. Recent research from Alkami Technology found that only 9% of financial institutions could be categorized as “data-first organizations,” which fully embrace the data-driven mindset and use this data for nearly every decision. The survey also found that while many organizations (39%) were in the “innovation-ready” stage of digital advancement, 14% are still slower in adopting advanced capabilities and rely heavily on third-party digital providers for data.
   
   The upshot? Building a data foundation is the first step to operationalizing generative AI. It ensures your business eliminates silos, it ensures quality, and it creates discipline across the organization.
   
   
2. Create an enabling infrastructure. It takes a range of foundational capabilities to scale generative AI adoption across the enterprise. Large language model (LLM) adapters, connectors and prompts are a good start. Ally Financial provides a forward-thinking example of the high-impact use cases that can benefit a bank’s bottom line; it has rolled out a cloud-based platform that, over time, will provide access to multiple LLMs, providing users with the flexibility to query the model of their choosing.
   
   
3. Set up governance that’s organized for the long term. Governance is about tracking and measuring. But it’s also about developing the support functions that ensure an organization is ready to manage—and adapt to—the accelerating speed of change. For instance, a financial institution may create a dedicated data governance committee that’s responsible for tracking data quality and compliance. Investing in an innovation hub for AI and machine learning (ML) can serve as a dedicated governance space for fostering more creativity and design thinking that can provide a framework for implementing use cases.
   
   For example, we partnered with a payment card provider that co-invested in a shared AI/ML innovation hub and is already reaping the benefits. The hub’s approach to design thinking provides the company with a framework to prioritize and implement use cases.
   
   
4. Keep it human. The financial services sector can expect AI to have an impact on tellers, traders and advisors, to name a few. However, the goal of generative AI is to augment human capabilities, not replace them. Emphasizing the technology’s human aspects allows a more nuanced approach to enterprise change than many headlines would have us believe.
   
   Take software code development, for example. For experienced, full-stack, high-end developers, generative AI tools may offer modest to no benefits. But the same tools could be enormously beneficial for entry-level developers just beginning their careers and still finding guidance helpful for tweaking code and writing syntax.

Although we’re in the early days of generative AI, a down-to-earth, pragmatic approach will get financial institutions on the road to real outcomes. By prioritizing these four principles, banks can overcome proof-of-concept fatigue and achieve sustainable innovation. Let’s focus—and think bigger to fully harness the power of generative AI.