The banking industry is awash in giddy talk about how artificial intelligence and machine learning will shake up how business is done. Yet on many levels, the promise remains unfulfilled, the revolution yet unrealized. The giddy sent back to committee.
On many levels, machine learning and artificial intelligence have served banking well. Automated underwriting engines leverage this technology to drive risk-based processing of loans. Meanwhile, automated valuation models have provided property valuation estimates, says Atul Varshneya, vice president of artificial intelligence at Tavant, a digital products company.
And yet, he notes, gaps persist: “First, I don’t know if our industry has really pushed to find creative ways to leverage AI and machine learning. Second, the mortgage process has always been seen as a factory—and although we’re always thinking about how to improve the factory, we’re not thinking about radically changing it.”
So what else (gulp) should bankers sweat?
“Third,” Varshneya contends, “we live in a highly regulated industry and need to always be aware of how our technology solutions can drive our industry’s primary goal of enabling the borrower.”
Structuring data to make it matter
Ah, but some things already work well in the AI landscape.
One bright spot centers on monitoring communications to improve internal processes, said Tim Estes, founder and president of Digital Reasoning.
“Machine learning and natural language processing algorithms enable banks to read volumes of unstructured data,” Estes says. That includes text from internal chats and emails, attached files and even phone conversation audio. “The technology can then understand nuances in human language and extract relevant insights to pinpoint instances of crime or misconduct.”
In the same way, he adds, “This technology can identify patterns in human communication to spot crime as well as pinpoint the instances where a sales team member has the opportunity to upsell or cross-sell a customer.”
Yet amidst this, you’ll find one theme that refuses to die: AI will displace and replace humans. “One of the biggest challenges is overcoming the assumption that technology is in competition with employees,” Estes says. “The truth is that it does the opposite.”
As AI enables, assists and augments employees, a small proof-of-concept project can serve as the turning point to calm anxious staff. “There’s nothing like watching a machine start to make predictions beyond what you thought it could and realizing that expertise has actually been transferred effectively so it can be scaled—and human time recovered for other things.”
Too slow … or no?
So why has AI been slow to catch on? Well, it has been slow—and then it hasn’t, said Avi Pollock, President, Grapevine6, an enterprise social and digital sales engagement platform.
“Almost every major financial services organization is using AI in some capacity, even if it's only in test or pilot mode,” Pollock says. “The reason widespread rollouts are few and far between is because it’s still difficult for people to put their faith in the machine.” (Not that those same nervous Nellies and Nickys don’t rely on smartphones to navigate countless aspects of their lives.)
“There is an expectation from people that the machine needs to be 100 percent accurate, 100 percent of the time, and tolerance for less than that is unacceptable,” Pollock notes.
That raises a supreme irony. “We have no issue putting our faith in people, who make errors all the time,” he says. “We don't have that expectation for perfection. As organizations continue to deploy AI engines in various capacities and see results relative to their current benchmarks and performance—instead of a benchmark of perfection—they’ll learn to trust the machine. Then we will see more widespread adoption.”
Hans Godfrey, COO of Agorai, an AI company that advises major banks, brokerages and retailers, offers this analogy: AI adoption compares to those of steam power, electricity, computer tech and the internet. Electrical service, for example, began in the U.S. in the mid-1880’s but “full” electrification was not achieved until 1950.
In similar fashion, the resources that propel AI have needed time to ripen. “Though research through the mid-1950s and ’60s seemed promising and received funding from various governments, the technology of the time was hampered by several technical obstacles.” Computing power proved scarce, as did large volumes of relevant, quality data.
“Now that those factors are much less of a barrier, AI-enabled solutions are primed to achieve their potential,” Godfrey says. “We’re reaching a tipping point because AI-enabled applications change the paradigm for developing solutions. Instead of writing application code that performs a specific function using hard rules, models are trained based on data that enables machine learning to develop the rules and take appropriate actions.”
Banking AI in action: an example
Sameer Gupta, financial services advanced analytics leader at EY, describes one of their successful projects: “A bank received more than 500,000 customer complaints annually. Its existing process for identifying sales practice issues relied heavily on key-word filtering in customer call transcripts with a manually crafted lexicon—then a team of hundreds of employees performed manual monitoring, review and validation.”
In response, EY “recorded customer calls [that] were converted into transcripts,” Gupta says. Extensive pre-processing of the transcribed calls decoded misspellings, acronyms and jargon. One team layered multiple natural language processing techniques—including semantic pattern matching and sentiment analysis—to identify and extract customer signals. Those signals informed the development of linear and non-linear machine learning models to predict the risk of sales practice issues.
In other words: guesswork out, best work in.
“The bank achieved a more than 90 percent capture rate of sales practice issues, reducing the risk of regulatory matters requiring immediate attention,” he says. “False positives declined by more than 30 percent and manual reviews dropped 80 percent.”
‘We need another five to seven years’
Though their workings might as well lurk behind the curtain for many, advanced technologies are indeed used frequently, says Leilani Doyle, senior vice president of product management at U.S. Dataworks, an integrated receivables technology provider in the banking industry.
“If you look at the fraud prevention areas of any credit card-issuing business, there is most certainly a blend of data,” Doyle says. That breaks down to patterns of behavior based on cardholder demographics, locations relative to handheld devices and social media accounts.
While that might sound like the makings of an exciting foundation, some bankers still see shaky ground.
“Many decision makers don't understand the technology well enough to feel comfortable making a buying decision” on the right tech, Doyle says. “We need about another five to seven years of continued real-life experience to get to the point where leadership can feel justified in adopting some form of AI.”
Alas, even the savviest executives can’t learn as fast as the machines can. But those who pick up the pace of adoption may find themselves ready, if you will, to giddy up.
Jeanne Pinder is the founder of ClearHealthCosts.com, an award-winning startup bringing transparency to the health care marketplace. She was an editor, reporter and human resources executive at The New York Times for close to 25 years.