How Credit Unions Can Build a Foundation for AI Strategy

Artificial intelligence is rapidly spreading across every facet of the financial services landscape. Previous applications of AI in banking have been incremental and purpose-built for niche use cases such as loan decisioning or sentiment analysis. However, the rise of generative AI and Large Language Models (LLMs) has democratized the field, making AI capabilities cost-effective, easy to use and intuitive to apply across many domains.

In the past, when financial institutions would source third-party AI solutions, they only needed to account for them completing a handful of tasks. Today, with generative AI being so pervasive and accessible, credit unions must have policies and tools in place to control AI tools that can complete hundreds of disparate tasks.

The Challenges of Adopting AI

Although generative AI is incredibly versatile, AI assistants often lack specialized context. As a result, employees need to be trained to check all outputs for accuracy and bias.

Additionally, credit unions should seek to mitigate risks around data leakages. Generative AI tools like OpenAI’s ChatGPT are free and easy to access, and data entered into them by employees or vendors may be exposed in the event of a breach or extracted by malicious actors.

Speaking of malicious actors, AI is making fraud easier and faster. This is the same type of fraud that credit unions have managed adequately for decades, but members are less experienced. Credit unions are putting more resources into member education and process design, reducing vulnerability to common fraud methods.

Vendor Selection

The current market is flooded with many vendors offering different variations and productizations of LLMs. It can be extremely daunting (and resource-intensive) for credit unions to stay current on the vendor space. Institutions may be paralyzed by choice during the vendor selection process without proper expertise and guidance.

Understanding how these solutions are built can help narrow the vendor space to a more manageable size. There are only a few foundational AI models, and many of the thousands of vendor AI solutions out there are using one of those models with only superficial configurations. These vendor solutions are sharing more than 80% of their DNA with the foundational models they were built atop.

From a risk perspective, this makes filtering through vendors more manageable. Credit unions can better inform their decision-making around vendor selection by identifying the base model a vendor uses and the changes they’ve made.

Credit unions also need to pay attention to what their existing vendors are doing with generative AI – many vendors have added generative AI features, including NewGen, Salesforce, Alteryx and Adobe, to name just a few. Credit unions should know how their data is used and where it might end up – for all their new and incumbent vendors.

Safely Adopting AI

Credit unions still on the fence about implementing AI must establish a solid foundation for AI adoption. Creating a thorough AI policy that tackles risk management, data privacy and regulatory compliance is essential to help curb the potential for internal misuse and data leakage. A further step is implementing an internal LLM assistant via a secure vendor for proprietary data (e.g., Azure OpenAI, Microsoft Copilot).

AI will continue to disrupt the financial services landscape, and developing a strong strategy for adoption is crucial. Credit unions that adopt with a focus on enterprise value and measured risk mitigation will come out ahead.

Connor Heaton is the Director of Artificial Intelligence at SRM, an advisory firm serving financial institutions in North America and across the globe.