Banks increasingly tout the potential for machine learning and artificial intelligence to transform the industry, but how many are deploying the technologies in a material way?

A survey out this week found 46% of financial institutions queried have deployed ML in multiple areas and find it “core” to business, 44% have deployed ML “in pockets,” and 10% are “experimenting and investing” in infrastructure and talent. But the survey did note that it included only data-science practitioners and C-level data-science decisionmakers from financial institutions that are currently using ML or intend to in the future.

Because ML-powered capabilities like image processing, natural language processing, and machine translation are “largely based on open-source libraries, and can be deployed relatively cheaply in the cloud, the barriers to entry have fallen dramatically,” according to Refinitiv, the surveyor. Refinitiv predicted a “flurry of commercial and product innovation from organizations of all sizes” with benefits beyond automation.

But barriers remain.

Josh Sutton, CEO of Agorai, a provider of AI tools, said the term “AI” is used too loosely and that only 10% to 20% of FIs deploy AI in a material way. Still, he said many banks are looking at pursuing AI implementations or pilots.

“What it amounts to is, how do you use data to run your business more intelligently, and how do you use algorithms and different approaches to analyzing and interrogating data to identify things that are not necessarily 100% linear in the way that they follow through?” he asked.

Also see: What Dimon’s Shareholder Letter Reveals About JPMorgan’s AI Approach

Sutton said banks that deploy AI are seeing positive returns from cost savings. By using robotic process automation with AI, the banks find a “dramatically increase” in efficiency on tasks like KYC activities, account opening or loan approvals.

“What AI is enabling via automation is a number of those things that used to take hours, days, or even weeks, to be done instantaneously,” he said.

Sutton said largely only top-tier banks so far have used individual-level data with even basic ML algorithms to offer “mass personalization” of banking services. Agorai mostly works with midsize to large banks, but Sutton said the company is starting to build and deploy products more applicable to regional and local banks.

“Typically, they have a problem right now, which is their budget constraints, so they’re looking for turnkey products, rather than bespoke solutions,” he said.

THE NEW FRONTIER

Sutton said natural language processing is still something of a new frontier for banks exploring AI, even though applications like chatbots aren’t so new.

“Natural language tools in general, right now, are very good at enacting specific commands,” he said. “They’re very bad with things that are vague.”

Asked where banks get stuck when it comes to implementing AI solutions, Sutton said it’s largely a management issue.

“Very similar to how the internet enabled a different way of banking, AI enables, across the board, a different way of engaging with your consumers as to what can be done,” he said. “Banks aren’t changing how they work to capitalize on what is possible today because that’s a big change from a cultural point of view.”

Sutton said it is hard for banks to find quality AI professionals to hire.

“More so than any other time in history, I’m seeing a lot of great point solutions be built by innovators and entrepreneurs and I think that there’s a real opportunity for banks to leverage those solutions,” he said. “Probably not coincidentally, at the same time, the venture arms of a lot of banks are more robust than they’ve ever been before.”

Arvind Purushotham, Managing Director & Global Head of Venture Investing for Citi Ventures, recently told Bank Innovation there are successful examples of ML being applied at Citi, including through its venture arm. For instance, the bank is an investor in Anaconda, a Python data science platform used for machine learning applications.

“We invested in Anaconda because we think that broad capability can be brought into Citi and embedded in many different places and that our data scientists, engineers, and analysts can leverage those relationships to actually use machine learning for their own purpose,” Purushotham said, adding the bank is “still in the early phases of that.”

ML/AI ADOPTION

To get over the ML and AI adoption hump, Purushotham said banks, first and foremost, need access to good data.

“It needs to be prepared in a way that you can actually make sense of it all,” he said. “Then you need to be able to apply these machine learning technology models to develop a capability that is then understandable, and you ought to be able to apply a model to come up with results and then be able to explain those results.”

Poor data quality was identified as the biggest barrier to ML by 43% of respondents to the Refinitiv survey. Another 38% cited a lack of data availability while 33% blamed difficulty in hiring data-scientist talent.