BMO uses artificial intelligence and machine learning to create innovative system for compliance monitoring
(Left to right) Anneke Deetlefs, data scientist; Taha Patoli, senior manager sales practice, legal and regulatory compliance; Conni Gibson, VP and chief compliance officer for personal and commercial Canada; Eric Morrow, director, data science - Machine Learning Centre of Excellence; Joanne Wingate, senior compliance officer.
Compliance monitoring is a critical and necessary task at any bank. Until recently, however, it has always been an activity that has required a great deal of manual effort. At Bank of Montreal, the last part of that statement isn’t quite as true as it used to be as the bank is using artificial intelligence and machine learning to automate compliance tasks.
It’s an improvement that has changed the way the bank operates, says Conni Gibson, BMO’s vice president and chief compliance officer for personal and commercial Canada.
“We do a lot of monitoring activities, of behaviours, of sales, of controls — are our compliance controls working? — and so forth. It was a lot of manual activities where we were going out to people asking them to send us spreadsheets, asking them to send us data. Now, we have the ability, through the machine that we’ve built, to access the entire bank’s data lake and pull all that data into us.
“It's not only compliance that looks at it. We’ve also partnered with our business on the front lines, the people who are actually selling things and actually serving customers. They can also look at it and have an insight into trending patterns. And then in phase two of the project, which we are launching now and we're very excited about, it's going to get into predictive behaviour.”
Although this is definitely a project that is under the authority and purview of the bank’s compliance officers, it wasn’t something they could create without assistance, says Eric Moss, deputy general counsel, senior vice president and chief compliance officer. It took collaboration with the bank’s technology experts to make the project a reality.
“We’ve worked with the machine learning group and other contacts before and some of our team were just thinking about the ways that we could do a better job and be more proactive in identifying things to make sure that our customers are treated fairly. This was a kind of a brainstorming thing: We have this broader data across the bank. [The team was] familiar with some of the machine learning capabilities that are out there, so they started a dialogue saying, ‘What can we do to pull together and utilize our data that we use in one context to help us serve and protect in another context?’”
It took around six months from the time the compliance team began brainstorming to the time the AI-based compliance system was deployed. Admittedly, the initial version of the solution was only able to look at one particular compliance issue in one particular customer protection area, but that was just the starting point. Today, Gibson says, it is examining four or five different areas, and she expects more will be added as the solution’s capabilities continue to scale. (While the bank is happy to discuss generalities of the solution, it is unable to talk about details, citing security concerns.)
Because the AI and machine learning capabilities were being applied to compliance tools, Moss says it was important that the compliance officers understand exactly what was going on. He also says it isn’t the type of black box operation that is often part of an AI implementation.
“We can absolutely articulate to a regulator ‘the data is coming from here, this is what’s being done with the data, it is being presented this way.’ I spent a significant amount of my career as a regulator, and being able to replicate and understand what the system has is obviously very important to the regulator. We have a firm understanding of all the components that go into it, what decisions or actions are being made and the output,” he says.
It’s the output that has made a difference to how the bank’s compliance officers do their job, according to Moss. He likens their pre-AI tasks to searching for needles in haystacks. Now, the AI solution hands them the equivalent of a large pile of needles. Being freed up from the more basic and time-consuming challenges means the compliance team is able to be engaged with more interesting and complex work—work that tends to challenge their existing skills.
Providing compliance employees with training and teaching them different skillsets is something Moss considers to be a core job for senior compliance officers. He adds that chief among the newly needed skills is the ability to react to real-time data.
“It’s a skillset to have someone be given an exception report that generates information for them to follow up on it on a very fast-paced basis, versus a skillset of someone who’s otherwise spending time researching and pulling information together to create the underlying information.”
While such a fast development and deployment cycle is somewhat unusual for a built-from-scratch project, Gibson says the people working on it made it possible. Excited people become project champions and cheerleaders, and Gibson says that’s exactly what any company needs if it wants to create and deploy any kind of successful project.
While such a fast development and deployment cycle is somewhat unusual for a built-from-scratch project, Gibson says the people working on it made it possible. Excited people become project champions and cheerleaders, and Gibson says that’s exactly what any company needs if it wants to create and deploy any kind of successful project.
“I can’t put my fingers on any challenges we encountered, and that’s partly because of the personalities involved; the folks on my team, they were just so engaged and so excited about it. And I think their excitement was contagious. The machine-learning folks here at BMO were also really excited because they saw the possibilities. They saw the precedent value. They were excited to build this and make it right.”
