Senior bank executives say artificial intelligence is now being deployed in meaningful use cases – but to realize its potential means figuring out how to scale it across the franchise.
A.I. technology is supposed to generate $1 trillion in incremental value to global finance and banking next year, say senior bankers in the technology space.
But such headline figures obscure the reality that there is little understanding of how to integrate it and scale applications using it.
Lisa Frazier, head of Wells Fargo’s innovation group in San Francisco, says holistic approaches are critical to unlocking technology’s value. “The organizations that embrace A.I. and big data at pace and at scale can systematically combine technologies to innovate with a real purpose,” she said, speaking at the annual conference of the Bankers Association for Finance and Trade (BAFT).
Many use cases…but no strategy?
Banks approach A.I. from the bottom up, however. They’re not fintechs creating brand new infrastructure.
Instead, a given department will have its own use case, from RPA to predictive analytics using machine learning, to cognition apps based on neural networks.
Even with a single business, such as transaction banking, there is a myriad of processes and regulations scattered around the globe, from customer onboarding to compliance to product design.
“How do you apply A.I. to get scale, while managing risk with adequate controls, and gain an advantage?” asked Madhav Goparaju, managing director and head of global advisory and solutions for Bank of America’s global transaction services, in Chicago.
Data, data everywhere
The foundation is digitization and data. Banks have been scrambling to build data lakes or move storage to the cloud where it can be accessed by multiple teams.
Even addressing the data itself is fraught, however. In a wholesale business, data comes from multiple sources, which need their own verification, compliance, and enrichment.
This is why many A.I. solutions have begun on the consumer side, and are now making their way into wholesale banking.
“A.I. isn’t new, but we now have access to huge amounts of compute and data to deploy into machine-learning models,” said Matt Fowler, vice president and head of ML, enterprise data and analytics at TD Bank Group, in Toronto. “That’s delivering insights into customer behavior.”
In consumer banking, the tech has been proven for small use cases, such as predictive modeling in mortgages or product recommendations. TD Bank, for example, is now getting into more complex uses, such as understanding transaction patterns in order to proactively warn customers of upcoming payments.
Governance: the differentiator
However, there is a vital step between organizing data and deploying it into A.I. solutions. Some call it modeling, others call it governance.
Marc Singer, senior partner at McKinsey in San Francisco, says there is a gulf between banks with or without a clear framework for managing risks. He notes that 87 percent of data-science projects don’t make it to production – but only 25 percent of banks have an enterprise-wide A.I. strategy.
“The difference is the organization and operating model that everyone understands and adheres to,” he said.
Wells Fargo’s Frazier says the roadblocks to realizing that vision are internal, rather than related to the tech itself. Legacy organizations and silos aren’t suited to the rapid pace set by big data and A.I. Banks are responding by moving more data storage to the cloud, modernizing infrastructure, and investing in A.I. tools to develop new models, but they must throw new resources at ensuring the integrity of the data itself.
“Without the intelligence it’s all just artificial,” Frazier said.
Frazier says a few years ago, after Wells Fargo experimented with A.I. in small pilots, it moved to apply it across the enterprise. “This is where we learned how hard it is to scale,” she said.
The bank had to dive deeper into people, processes, and thorny issues such as data protection and privacy – in a word, governance.
As banks restructure around this, they are still a ways off from solving outstanding issues including the ethics of A.I. and data ownership. Governance sometimes means not being able to do all the things a bank’s team would like. These institutions face reputational as well as regulatory risks.
Of course, such risks haven’t stopped big banks from engaging in all kinds of fraudulent or unethical activity.
But Fowler at TD Group acknowledged some regulators such as the Monetary Authority of Singapore have been proactive about defining ethical uses of A.I.
Deployments begin
Now that banks are finding A.I. can deliver on its original hype, they are going to use it more and more. Every global transaction bank is working on digital and analytics. The question is to what extent they are making this an end-to-end transformation.
The initial focus was on saving costs in areas such as fraud detection, complying with anti-money laundering laws, and bolstering cybersecurity. Now they are looking at the revenue-generating side, such as predicting companies’ liquidity and forex exposures.
As banks increasingly adopt data-hungry analytics, they are going to face new complications around staffing, skillsets, processes, and risk management.
Those institutions that turn their multitude of ground-up experiments into a cohesive strategy are better placed to succeed; those that ignore the unglamourous modeling and governance aspects are setting themselves up to fail.
