An effective AI model starts with quality data, forcing financial institutions and fintechs to hone their data strategies for specific use cases.  

Seventy-nine percent of financial institutions say a lack of quality data is a major barrier to AI deployment, according to a March report by the Bank for International Settlements. Of those FIs, 96% cited inaccurate or unadaptable data as the top challenge, followed by unlabeled data at 94%.  

No matter how sophisticated the model is, inadequate data essentially renders AI useless, Richard Ullenius, vice president of banking and financial services at global tech company CSG, told FinAi News.  

“There’s no point throwing in a lot of sophisticated technology if [banks] don’t have the data in the right place to start off with.”

— Richard Ullenius, VP of banking and financial services, CSG

Keys to data quality include cultural alignment, case-by-case analysis and structure.  

1. Culture 

Ensuring that employees properly collect, report and store data is crucial to building a strong AI infrastructure, Will Rhoads, chief innovation officer at Brentwood, Tenn.-based Sonata Bank, told FinAi News.  

“Data quality, especially for community banks, is actually a cultural problem masquerading as a technical one,” he said. “You have so many data signals that live in a spreadsheet, in an email, in a team channel and in someone’s head. The process is not as formalized as you would find in a much larger organization.” 

Thus, banking leaders must establish clear guidelines on how data should be handled based on the specific use case and AI model, Rhoads said, emphasizing the need to examine everyday workflows when developing guidelines.  

“We look around at the questions our employees are already asking,” he said. “What are the reports people are requesting? Why are they requesting this information? … If we’re not going to question what our staff’s asking, we’re not actually solving the problem.” 

2. Department by department  

In most instances, data quality is not one-size-fits-all and should be treated differently across an organization, Sean Weadock, chief product and technology officer at digital banking platform Lumin Digital, told FinAi News.  

“You have to get the right people involved, whether it’s marketing, fraud or whatever your use case may be,” he said.  

For each use case, banks must gather insights from subject matter experts to determine which data points translate to better decisions, Weadock said.  

Banks then will better understand how to curate and structure data for each AI application and whether to augment AI models with third-party data, he said.  

Sonata’s Rhoads agreed.  

“We want the lending team to own loan data quality; we want the deposit ops team to own deposit data quality. The best ideas for automation and process improvement come from the people that complete that process every day.”

— Will Rhoads, chief innovation officer, Sonata Bank

3. Structure 

Structured data is pivotal for effective AI, especially for document-heavy workflows, Joshua Summers, chief executive of EnFi, an agentic AI platform for commercial lending, told FinAi News.  

When building AI agents for lending workflows, “we found there was a missing piece — the data is [terrible],” he said.  

“Every customer we go to, it’s a mess,” he said. “Where is the data? What is the data? A lot of it’s conflicting. How do you clean it up? How do you standardize it?” 

This prompted EnFi to focus on the institutional knowledge layer — the framework that connects enterprise systems and AI applications — to turn “unstructured, disparate data into something that’s structured, where we understand the people, the places, the things that are trapped in these documents,” Summers said. 

“We build relationships between them into a graph, and we expose that into our agents and into the humans that are using it,” he said.

Similarly, banking and payments platform i2c is building pipelines that funnel real-time data from legacy systems into AI models, CEO Amir Wain told FinAi News. 

“That data is often segregated in multiple different systems,” he said. “So, how do I put all of this together in real time and generate a comprehensive view?”

Further highlighting an emphasis on structure, Fargo, N.D.-based Bell Bank is building a “common layer” to align with its metadata repository, enabling the $14.8 billion bank to centralize and organize enterprise data and data policies, Director of AI and Data Michelle Mack told FinAi News. 

“There are lots of things happening in that space because with bad data, you’re going to get bad results, and the AI is just not worth it,” she said.

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.