© 2019 SourceMedia. All rights reserved.

How BNY Mellon is going further on AI

While several traditional banks are testing artificial intelligence, BNY Mellon says it's going further than most, committing to the technology by involving it in multiple aspects of its business.

"We’re not experimenting with AI, we’re really doing it," said Roman Regelman, senior executive vice president and head of digital for the bank.

Among other things, BNY Mellon has deployed 300 bots using robotics process automation software from Blue Prism. While some purists do not consider them full-fledge AI, bots are widely seen as a low-IQ form of AI that automate simple tasks previously performed by human beings. At BNY Mellon, the bots are deployed across businesses and functions and execute about five million processes.

“These robots are doing manual work,” Regelman said. “They’re doing stuff that humans can do, but don’t like to do, and often don’t do that well. That allows the humans we have, the employees, to spend time on more value-added activities, like spending time with clients and spending time on more complicated cases.”

Analysts have taken note. “BNY Mellon is a more visible part of the big-bank pack who three or four years ago were just reaching the point of 100 bots each of varying sophistication in proofs of concept, alpha and beta tests,” said David Weiss, principal analyst at Market Structure Metrics.

As it deploys AI and bots, Regelman said BNY Mellon is making sure that the technology and humans work together.

“The future of AI is not AI alone, but what we call AI and human intelligence,” he said. “It’s not about robots replacing people, and it’s not about people fighting with the machines. It’s them working together as artificial intelligence plus human intelligence. When they work together, we have something very different that fundamentally unlocks something neither can do.”

Where BNY Mellon uses AI

In one example, BNY Mellon has been extensively using AI in clearing U.S. Treasuries.

Every day, roughly $60 billion to $70 billion worth of U.S. Treasury transactions fail, according to Regelman, because something doesn’t match. Market participants learn what failed and they rectify that.

BNY Mellon

The bank has implemented an AI-based U.S. Treasury settlement fails prediction model that detects trends in the types of trades that usually fail and finds the signals or combinations of factors that could spell possible failure: amount, time of day, and so on. It notifies staff that a transaction is likely to fail before it does.

“That allows staff to work with clients and manage their liquidity,” Regelman said. “That’s an example where everyone wins. That’s computers working with people.”

Another example involves emails. The operations team receives one million emails a year on so-called client inquiries.

“Quite often they’re not inquiries at all, they’re just instructions,” Regelman said. “Managing a million emails a year is a lot, and a lot of this work is not value-add. So we deployed an algorithm to look at these emails and classify them.”

The AI system successfully categorizes 90% of them.

“This way the humans can focus on specific emails, which go to the right people,” he said.

The same program can now resolve some email queries, such as, "Did you get my earlier email?"

“Ultimately the humans that work on this have better jobs, the clients get more personal service because we focus on specific cases, and it’s AI working together with humans,” Regelman said.

A third example is contracts. The 235-year-old bank’s oldest client contract will be 100 years old next year.

“As you can imagine, it wasn’t digital,” Regelman said.

The bank has digitized all its contracts by scanning them and extracting all the relevant data. It uses machine learning to understand the terms and AI to understand inconsistencies between contracts and between clauses.

“Did this eliminate lawyers? Of course not,” Regelman said. “Now lawyers can focus on resolving inconsistencies with clients, as opposed to spending time trying to find old contract. It’s pretty good accuracy and a better job for the lawyers.”

The bank also uses AI to detect fraud; to automate the classification and extraction of key data from unstructured client wire transfer instructions received in email attachments; and to automate the classification, compliance verification, examination and payment processing of trade finance documents such as letters of credit.

Regelman noted out that computer programs can be infused with the biases of their human creators.

“Responsible usage of AI is a really big deal, because as the world becomes more dependent on data and computers, we need to make sure it’s responsible,” he said.

This is a common theme in the banking industry these days.

Bank of America was the driving force behind the Council on the Responsible Use of Artificial Intelligence at Harvard University. McKinsey and Accenture have built practices around the responsible use of AI in financial services.

BNY Mellon also has a diverse digital team, Regelman said.

“Our head of AI is an African American who was not born in the U.S., that’s just one example,” Regelman said. “When someone like that is working on AI, he’ll probably have less of a bias. Or he may have a different bias than the rest of the company does cumulatively.”

Those 300 bots

BNY Mellon's bots are typically doing high-volume, repetitive, manual and rule-based tasks that require processing across different systems, including feeding and retrieving data to and from different programs.

Bots download transfer agency data several times in a day and investigate and resolve commonly occurring accounting breaks.

Bots extract, review and enter tax information from W8 and W9 forms during client onboarding. They automate the data entry and processing of corporate action elections received from clients. They produce monthly statements for clients including reports for audit purposes.

They process client instructions on core securities servicing platforms and create audit reports. They automate the extraction and consolidation of client reporting across platforms to enable direct access by clients.

They repair incorrectly formatted manual wire transfer instructions, which leads to faster processing. And they automate the reconciliation of investment managers’ monthly statements against internal and external records.

“AI is one of the most exciting opportunities we see in front of us,” Regelman said. “But it’s not AI alone, it’s AI with humans, and it’s not AI as a side show. It’s just a fundamental way of working and of unlocking the intelligence of people around the world.”

For reprint and licensing requests for this article, click here.