June 17, 2020
A simple maxim guides how the BMO AI Capabilities Team works: increased model performance is related directly to increased revenue and decreased costs and enhancing customers’ banking experiences. Welcome to BMO’s culture, where the AI Capabilities Team has a prominent seat at the table and a license to pitch AI applications for the bank’s biggest value drivers.
“The fact that we’re partnering with the business teams tells you that BMO is directly tying AI to revenue and cost savings,” said Yevgeniy Vahlis, Head of Artificial Intelligence Capabilities at BMO Financial Group.
That opportunity to improve performance drives the bank to stay current with technical advances in model accuracy.
“Generally the way AI is talked about is pretty binary: are you doing AI or not? But there’s actually a big difference between a model that gives you 30% accuracy and one that gives you 80% accuracy. That’s a huge difference, and it’s something that’s not being talked about much when companies make the decision to have an AI solution,” said Vahlis. “This is the benefit of using cutting-edge research for industrial AI solutions: if you have the expertise to do that, your models can be significantly more performant or more accurate.”
BMO’s Platinum sponsorship of the Vector Institute offers opportunities to experiment with leading AI research through projects hosted by Vector’s researchers and Industry Innovation team. “These projects complement BMO’s internal capabilities, accelerating us in converting leading academic AI research into new models that we can use to support strategic value propositions with a diverse group of business stakeholders across the bank,” noted Sami Ahmed, Chief Digital, Data & Analytics Officer of Wealth Management. “We have seen positive traction based on our involvement across a number of Vector projects and it’s an opportunity to contribute back to the AI ecosystem in Canada.”.
One such project is Vector’s Natural Language Processing (NLP) Project, which involves multiple workstreams that focus on replicating state-of-the-art natural language processing (NLP) models and training them to perform domain-specific tasks related to participants’ business objectives.
In the NLP Project, Stella Wu, an Applied Machine Learning Researcher at BMO Financial Group, proposed and developed a financial version of BERT – one of the most advanced language representation models available. BERT refers to “bidirectional encoder representations from transformers”, an NLP technique released by Google in 2018, which was a breakthrough not only for its ability to understand word meanings, but their contexts as well.
Wu and Vector researchers used several online financial news sources to add over 182 million finance and market-related terms and their contexts to the data set. They then pre-trained the model with this enriched dataset, setting the stage to fine tune it to achieve specific tasks for BMO on analyzing market sentiment.
Supporting this work were the face-to-face conversations, advanced lectures, and weekly feedback that project participants received from Vector researchers and guest speakers. “I didn’t have any experience with NLP before,” Wu noted. “It was almost like a marathon lecture that gets you all the latest information. It just feels like when you’re in Vector doing the research, you’re more connected to the state of the art.”
“Access to Vector researchers significantly accelerates our progress,” said Vahlis. “Having the best researchers in Canada or Ontario available to us increases both the chances of success and the speed in which we can complete this.”
Apart from complementing internal efforts, the relationship with Vector provides another benefit to BMO: retention.
“It’s very hard to find people who have the research skills, a deep understanding of the science, and the ability to take that themselves and directly apply it to the business,” observed Vahlis. “They choose to stay at BMO. Retention is not a trivial matter, and working with Vector helps.”
Wu’s experience reinforces this. “I was fascinated by the Vector Institute. Before I went into the industry, I wanted to go work there. To me it’s a very inspiring place.”
Ultimately, though, the main value of that inspiration and research exposure comes from the real-world results that they create. “While the whole experience was very academic, the outcome was a practical benefit to the business,” concluded Wu.