Canada’s Scotiabank is reaping the benefits of its AI investment across the bank, finding returns in the back office of its global banking marketing division, improved security on the front line, and better call center responses.

Phil Thomas, Scotiabank’s executive vice president of customer insights, data and analytics, said AI is paying off in multiple areas for the $926 billion Canadian bank, which reports $12.77 million in AI-related cost avoidance within operations.

“We made this organizational change in about six months,” Thomas told Bank Automation News. “We believe it’s an opportunity to look at how we can digitize some of our operational processes and very much in line with our technology growth strategy.”

The Toronto-based bank launched its global AI platform in November 2020, and it now has more than 500 people working with AI and data analytics. It’s leveraging that investment in AI throughout the organization, beginning with C. Mee, a personalized sales and marketing tool rolled out in February.

“We’re on a journey of becoming a very data-driven organization, and at the tip of the spear in terms of how we’re how we’re thinking about using AI to improve the customer experience, improve the employee experience, and then how we add value to ultimately to our shareholders,” Thomas said.

Automating contract analysis

Scotiabank’s most recent AI launch is AIDOX, an AI-based tool that automates the analysis of legal contracts in the global banking markets division.

Before AIDOX, a “subjective and careful review” of a legal contract would be performed by a person in a process that took approximately 1.5 hours, according to Pablo Vidales Calderon, who headed the AI team devoted to AIDOX. The new automation has cut that time by 85% to less than 15 minutes.

The system breaks down the analysis of the contracts into two parts — semantic verification and economic specifications.

“In every legal context, you will have a lot of semantics that are very relevant. So basically, if I were a lawyer, I will say the most important thing is the terms that you use in that contract,” Calderon told BAN. “As you can imagine, this is a very complex space because this is semantics and linguistics, but in this case, we have trained the specific permission for legal context.”

The solution was built in-house in Python, among other languages, Calderon said. His team trained a neural network to implement natural language processing using an approach to semantics called “word embedding.” This technological approach turns words into numerical values to help the system understand context and relationships.

For example, Calderon said the system can determine that the words “woman” and “crown” together could refer to a queen. The system works similarly with legal terms, and can, for instance, determine whether paragraphs are out of order, and even whether sentences and words are, as well.

AIDOX can also translate words into numerical representation and determine if the numbers in a contract make sense. This process uses what Calderon termed “simpler machine learning techniques based on a set of business rules.” For example, the system would flag $1.000 — which should be $1,000 — as a potential error.

Embedded functionality

Integration was also key to ensuring AIDOX’s success, Calderon added. Rather than roll out a completely new tool, the team embedded the AI functionality into the existing system used for manual contract evaluations.

“You can have the best neural network working perfectly, trained and everything, but if you don’t put it in your integrated process, it’s going to be difficult to gain that adoption and evolution of it,” Calderon said.

Scotiabank estimates AIDOX will save the company $1.7 million annually in legal contract analysis.

Since it’s AI-based, the system learns as it’s used — and it’s scalable, Calderon said. “We can translate the solution and it will scale very well to different use cases,” Calderon said. “Currently, we’re exploring the use of mortgages, which is highly intense in the use of documents and legal content.”

Automating for security

Scotiabank also leverages an AI toolset that uses advanced analytics with machine learning to prevent fraud, Calderon said. Called “watchdogs,” the tools monitor different onboarding processes for odd behaviors that might indicate a bot.

“We have trained these watchdogs to identify those customers that are fake,” Calderon said, adding that the watchdogs are also used to prevent phishing attacks.

Also: Read how automating AML paid off for Scotiabank.

The Canadian bank has a strong presence in Mexico, so the AI team is now training the watchdogs to speak Spanish.

“As you can imagine, this is highly reusable, but there are different aspects of fraud techniques in all countries,” Calderon said. “Behaviors are different. The nice thing of using behavior is that you learn from the data the behaviors that you need to look at” to train a watchdog for any geography or population, he added.

Automating the call center

In the call center, Scotiabank uses machine learning to assist with customer segmentation. As customers call in, the system identifies as much as it can about the customer in real time before assigning the customer to an agent. The contact center’s key performance indicators, including customer satisfaction and average handling time, have improved since the bank added the system, Calderon said.

“The good thing of having machine learning solutions in place is that, of course, you can adapt your learning,” Calderon said. “We’re not saying that we’re not doing mistakes when we’re deploying solutions, we’re saying that we’re learning from them and we will improve every day. That’s the power of it.”

Bank Automation Ignite, on April 13-14, is the event for inspiring automation initiatives and investment in financial services. At the virtual event, financial services professionals can discover new use cases and technologies that are accelerating automation in banking. Learn more and register at www.BankAutomationIgnite.com.