The Roadblock Facing AI Use at Banks

Learn more about how community banks can leverage artificial intelligence and other topics at Bank Director’s Experience FinXTech May 13-15 in Tampa, Florida. 

The following piece appeared in FinXTech’s latest Intelligence Report, “Artificial Intelligence, A Real-World Approach” in a section titled “The Data Problem.” 

Most artificial intelligence applications that financial institutions are interested in will use
internal data — and that could be an unwelcome realization for institutions that don’t have formal data management initiatives.

“I think even before [banks and credit unions] start to think about the AI use cases and what [are] the right models and technologies to use, the first thing they should get control of iis the data itself, and [whether they] have all the right data in a usable format,” says Ashvin Parmar, who heads the financial services generative AI center of excellence at the technology consultancy Capgemini.

AI runs on a company’s data infrastructure: Computational power processes data via models and components that link to an individual application or use case through connections like application programming interfaces. If the data isn’t organized or clean, or if the technology underlying these connections, or rails, is “choppy” due to older, legacy or disparate systems, then it’s harder to run the technology and get good results, says Alexandra Mousavizadeh, cofounder and CEO of Evident, a research firm focused on AI adoption within the banking sector.

A data management strategy can help banks and credit unions understand what data they have and what they will need to acquire. An institution may realize that some of its identified use cases may not be able to move forward if it can’t validate the quality of the data, or if it’s not labeled and reviewed by subject matter experts.

Financial institutions will need to closely consider their data privacy practices and safeguards. AI models may need to train on or analyze an institution’s data, and organizations are responsible for their intellectual property, including the quality of the data, what information is in the data set and making sure that data doesn’t leave. Institutions will need to think about how they hide or remove personal identifying information and mask or anonymize the data.

They could consider using a data classification system sorted by sensitivity, public availability and market importance, says Daragh Morrissey, director of AI at Microsoft Worldwide Financial Services. Failing to manage this could result in long-term reputational risk and loss of customer trust, along with regulatory and compliance penalties.

But to manage this risk successfully, Morrissey points out that the relevant business line at the bank needs to be involved. The IT team won’t be able to look at loan data and ascertain what information is highly sensitive and what is public. Lenders and credit analysts will need to share their insights.

AI may also be able to assist here. Both Mousavizadeh and Parmar point out that institutions can apply AI to their data to clean it, standardize it and otherwise make it usable. AI may be able to make sense of a fragmented data set or sift through and strip out customer information that shouldn’t be inserted into a model. Parmar says an institution may even be able to juggle an AI project and a data management project in parallel. But good data management — and good data privacy and security protocols — are essential to truly leverage AI’s capabilities.

“The infrastructure is such a huge part of it; it’s impossible to delineate one from the other. But you could definitely do both at the same time. There are going to be some areas in community banks where you can start using AI on subsets of the data and build real capabilities,” Mousavizadeh says. “It’s just [that] the better the data is, the more you can get out of it.”