Artificial intelligence is moving from science fiction to practical reality fast.
AI — technology that teaches machines to learn so they can perform cognitive tasks and interact with people — is suddenly accessible to many companies. Costs associated with the advanced computing and data-storage hardware behind AI are plummeting. A growing number of vendors also offer AI tools such as robotic processing automation that can be configured without the help of a rocket scientist.
So this is clearly an area more banks will need to pay attention to going forward.
Already some AI pioneers have emerged in the financial industry just over the past year: Bank of New York Mellon‘s use of robotic process automation in trade settlement and other back-office operations; Nasdaq‘s search for signs of market tampering with an assist from AI; UBS’ initiative to answer basic customer-service questions through Amazon’s virtual assistant, Alexa; and USAA‘s development of its own virtual assistant.
Most large banks are considering using AI wherever mundane or repetitive tasks could be offloaded to a computer fairly easily.
What It Can Do
Here are some examples of where AI could make the biggest difference.
• Customer conversations. Chatbots, natural language processing and speech processing could all be used to improve social interactions. In addition to USAA, Bank of America, Capital One Financial, Barclays and BBVA are experimenting with AI-powered virtual assistants.
“The vision that excites me is the one where we have seamless interactions, where I’m interacting with people, with the bank, with systems in the bank, and at the end of the day what the bank is giving me is exactly what I want,” said Marco Bressan, chief data scientist at BBVA. “We shouldn’t have a fixed idea of what the customer wants. There are some customers that the less they see their banks the better, as long as their money is well taken care of. Other customers want to see their bank every day. We have to serve both. And communicating with each of those from an AI perspective is very different. One has to do with full automation, and the other has to do with a smart interface.”
• Automated investment advice. AI is used to help investment advisers and robo-advisers make better recommendations to customers. Australia’s ANZ Group has been using IBM’s Watson in its wealth management division for three years. Watson can read and understand unstructured data found in contracts and other documents, comb through millions of data points in seconds, and learn how to draw conclusions from the data. It can assess a new customer’s financial situation more quickly and comprehensively than a human being, and it never forgets anything.
BlackRock uses AI to improve investment decision-making. The startup Kensho combines big data and machine-learning techniques to analyze how real-world events affect markets.
• Faster, better underwriting. BBVA uses artificial intelligence to improve its risk scoring of small and midsize businesses. “We realized we could update data in real time and integrate it with what the risk analysts were doing to have a much deeper understanding of their own portfolio,” Bressan said.
Some online lenders use AI to speed up their process. The software can look at hundreds or thousands of attributes, such as personal financial data and transaction data, to determine creditworthiness in a split second. The system learns as it goes — when a lender gets payment information on loans, that information gets fed back into the system, so its knowledge evolves.
However, some people question whether AI programs can be trusted to make sound, unbiased lending decisions.
• Streamlined operations. BNY Mellon, Deutsche Bank and others are using bots in their back offices to automate repetitive tasks like data lookups.
• Assisted account opening. Account origination can be a slow, cumbersome process. Some banks are experimenting with robotically automating some elements, such as data verifications.
• Fraud detection. Card issuers and payment processors like PayPal use AI to compare current card transactions to the user’s past behavior as well as to general profiles of fraud behavior. Human analysts teach the model to discern the difference between legal and fraudulent transactions.
• General efficiency. “The financial industry is an enormous percentage of the GDP,” said Robin Hanson, an associate professor at George Mason University. “A lot of it is due to various regulations and rules about who has to do what and how. It’s entrenched in regulatory practices, and it’s really hard to innovate in finance because you run into some of these obstacles.”
For example, Hanson wanted to sell some books at a convention. To do so, he had to apply for a tax ID, pay a fee and cover $25 in sales taxes. That required him to go to his bank to get a cashier’s check, for which he had to pay a $5 transaction fee and postage. “That’s an enormously expensive, awkward process,” he said. “If we had an efficient financial system, that would cost pennies.”
As AI is used to improve the speed and efficiency of tasks now performed by humans, there are potential unintended consequences. For one, people in lower-paying jobs in operations, branches, compliance and customer service are likely to lose those jobs.
“Bank executives say they’re going to take those people and put them into high-tech, high-pay jobs to help us code, help us do this, help us do that. It’s just not going to happen,” said Christine Duhaime, a lawyer in Canada with a practice in anti-money-laundering, counterterrorist financing and foreign asset recovery and the founder of the Digital Finance Institute. However, “the bank may end up with the same number of employees,” as it sheds customer-facing jobs and hires trained software developers to code.
There are also privacy concerns around the use of AI in financial services. “From a consumer protection point of view, there are concerns people need to take into account when it comes to AI, machine learning and algorithmic decision-making,” said Steve Ehrlich, an associate at Spitzberg Partners, a boutique corporate advisory and investment firm in New York. “Say a company wants to look at your social media or your search engine history to determine your creditworthiness. They go into Facebook and find a picture of you that you didn’t upload. It’s a picture of you at a bachelor party or gambling at a casino. That data gets fed into the algorithm. For one, they should tell you they were going to be taking that information.”
There is also the chance that bots and AI engines could run amok and make poor lending decisions, or commit an operations error that a human with common sense could have averted.
What Banks Can Do
These caveats aside, banks’ wisest course is to prepare to be part of the revolution.
One thing they can do is create an internal center of excellence where a group of people become experts and help bring AI to other parts of the company. They could test technology and use cases and guide the business units in their adoption of bots and AI. Citigroup and BBVA are among the banks doing this. BNY Mellon has a robotics process automation team that partners with businesses and has come up with eight pilots, including settlement and data reconciliation.
Banks also can try to encourage people to embrace AI — even if their jobs are at risk. It helps to communicate that there could be some benefit to them. “People in operations and data analysts don’t want to be doing this work anyway — swivel-chair work, mindless copying and pasting and keying in data,” said Adam Devine, head of marketing at WorkFusion, a robotics process automation software provider that competes with Blue Prism and Automation Anywhere.
David Weiss, senior analyst at Aite Group, also sees the trend as an eventual positive for employees. “I personally argue for human augmentation — go after the peak human problems first,” he said. “There, you’re not going to cut jobs, you’re just going to make people more functional, and leverage their inorganic intelligence more.”
But there’s no question the workplace will change and people will have to adapt.