Vetting new technology and jumping through compliance hoops is nothing new for financial institutions — and implementing AI should be no different.

In theory.

As large language models (LLM) improve daily, financial institutions (FI) are finding new applications for them. The AI tech once known as the magic behind chatbots and streamlined call center experiences is now informing anti-money laundering solutions, code generation, debugging and credit decisions.

Banking executives were wary of implementing AI until recently. In fact, 73% of banking executives surveyed by consulting firm Gartner for a June 13 report said they are either planning to deploy generative AI or assessing the benefits and risks of doing so. In fact, in April just 7% of banking executives were not sold on AI, down from 47% in February. 

Now, the question is not whether banks should use AI, but “How do you implement AI responsibly across the entire organization?” Gartner analyst Moutusi Sau told Bank Automation News.

It starts with data

In the first quarter of this year and into the second, financial institutions, including Bank of America, Fifth Third and Discover Financial Services, upped their investments in AI. Tech providers, too, including Google, Microsoft and IBM, proved their commitment to AI, specifically through investment in OpenAI and Bard.

Each FI has its own AI strategy.

Discover Financial Services increased its total Q1 operating expenses 22% year over year to $1.38 billion, which included increased costs for employee compensation, business development, information processing and communications, professional fees, and premises and equipment, according to company’s quarterly earnings supplement.

Through its tech investment, compliance management continues to be top of mind, Executive Vice President and Chief Financial Officer John Greene said during the bank’s Q1 earnings call, specifically in machine learning and AI, he noted.

“We continue to invest in those aspects as well as technology to support our overall compliance management system,” Greene said.

Machine learning (ML), a subset of AI, is an area that Discover has invested in for many years, Arjun Kannan, director of data science research, told BAN, noting that the key to leveraging AI and data for Discover is the use of compliant data.

“Everything starts with data,” he said.

Static and supervised AI

To ensure that Discover‘s data is compliant and unbiased, the financial institution prioritizes using Fair Credit Reporting Act-compliant data, Kannan said, specifically, “static” and ”supervised” ML algorithms.

Static ML consists of models that are trained offline, according to the Google Developer website. Static differs from dynamic models, which are trained online.

For Discover, “’static’ means you’re not constantly training models with real-time data,” Kannan said. “Why that’s important is when it comes to compliance. It’s hard if you are constantly updating with real-time information and making decisions with rapidly updating models.”

The FI also looks to ”supervised” ML, he said, meaning that “we have a defined target we are tracking; you’re not letting AI loose, so to speak.”

AI decisions improve with rapidly changing models, however FIs must keep tabs on how fast AI can be trained. Also crucial is knowing the tech’s ability to monitor the efficacy of the decisioning, and how quickly it can be refined to reflect the impact of decisions, according to a report released in May by consulting firm Forrester.

“What these enterprises need is a platform that can simultaneously enable them to harness the power of AI while enhancing and governing it with well-proven and trusted human business expertise. The best automated decisions come from a combination of both,” according to the report.

Explainable AI

When using AI for decisioning, banks must combine their use of compliant data with thoughtful solutions that explain how decisions are made — a regulatory requirement within financial services, Kannan said.

“If you take an adverse credit decision against a consumer, you are expected to explain to the consumer why that was the decision,” he said.

Discover “connects the dots in terms of, how can we not … materially sacrifice performance and still achieve fair outcomes.” The FI balances performance, fairness and explainability, Kannan said.

“We have actually put our work in the public domain because we believe everyone will benefit by looking at this [research] and arrive at common standards,” Kannan noted.

Explainable decisioning was put to work in 2019, for example, when Goldman Sachs and Apple were accused of biases within their AI-driven credit decisioning algorithm, Gartner’s Sau told BAN. But when “they went back and looked at the model, they found that there was nothing they did wrong. It was just a pattern of unemployment that caused that issue.”

Leading AI decisioning platforms, according to the Forrester report, include SAS, FICO and IBM, based on their data-to-decision capabilities, use of analytics and machine learning and ability to monitor governance.

Connecting the dots

FIs are rushing to leverage AI within finance, and Discover is working to connect the dots between data, ML and credit decisioning to create an industry standard for explainability compliance, Kannan said.

“So, we use AI, sure, but the bigger application is really with machine learning, the static and supervised,” Kannan said.