As one of Canada’s Big Five banks, Bank of Nova Scotia takes a data, analytics and AI approach aimed at better understanding and serving customers, said Grace Lee, chief data and analytics officer. Her mission is to drive business growth, customer experience and operational efficiencies through the use of AI, machine learning and data-driven insights at the bank better known as Scotiabank.
The stakes are high when it comes to customer retention: Scotiabank has more than 10 million retail, small business and commercial customers in Canada and 10 million retail and commercial customers in Latin America, the Caribbean and Central America. The bank has approximately 90,000 employees and approximately $1.2 trillion in assets.
Scotiabank’s two AI application areas
In recent years, Scotiabank has adopted an AI strategy that focuses heavily on last-mile execution, Lee said. “Where we’ve seen other organizations fail to take advantage of AI and machine learning at times is that it doesn’t necessarily always lead to practical results,” she said. “So you’ll find that sometimes we call it ‘blue-collar’ AI or analytics, but it’s really about making sure we see that [AI] Models through all the way from inception to [deployment into] Production.”
[ Lisez la version française : « Comment la Banque Scotia met l’IA en œuvre pour améliorer l’expérience client » ]
And that means that AI is embedded directly into existing processes and offers real benefits to stakeholders, e.g. For example, providing timely advice and personalized offers to customers, creating a degree of efficiencies so employees can better serve customers, or allowing the bank to better predict when customers might be in need, Lee said. “There’s a lot more we can do to actively monitor and truly understand our customers’ behavior and therefore their needs and preferences,” she said.
However, AI isn’t just helping Scotiabank grow and evolve the customer experience by “knowing better,” but being able to “do it better,” Lee said. It also gives the bank the ability to “apply AI to automation, whether that’s in a chatbot or any other intelligent automation that we would have across our portfolio,” she said.
When it comes to implementation, it’s important for AI teams to recognize that while AI has traditionally meant artificial intelligence, Scotiabank and other organizations, particularly in the banking industry, are increasingly referring to it as “augmented intelligence,” Lee said. Because it really has to be embedded in existing processes so that it benefits the customers and employees of the bank.
“There’s very little we really want to do that would be fully automated without some degree of augmentation and human oversight,” she said. “So I think that’s a really big lesson that we learned early on when we tried a little bit more with the artificial and not a lot with the augmented. We found that despite being a very sophisticated model, the receptivity and impact it had didn’t really do much for our customers or our employees. So co-creation is super important.”
AI use cases at Scotiabank
Scotiabank is working to provide Natural Language Processing (NLP) to provide an enhanced customer experience. In the first phase of the project, the bank is building a chatbot to handle basic FAQs, Lee said. It aims to “address common questions from customers [about] products and prices, [for example,] which are routed to a live agent who can be answered via an AI-guided user interface,” she said. “We want to offer our customers a better conversational experience so they don’t have to wait minutes or long on the phone to reach an agent when their question or request is relatively simple.”
If the chatbot proves effective, not only would it result in a better customer experience, but it would also make the bank work more efficiently by allowing their customer service representatives or other advisors to work on issues that need to be handled by humans.
Scotiabank is using AI to improve customer experiences in several other ways, Lee said.
One of them is the Global AI Platform launched in November 2020. The platform is the infrastructure that enables the bank to offer its customers faster insights and better advice, using machine learning to anticipate and understand their needs. “We have an on-premises component and we have a cloud component that is growing rapidly. And that’s where we actually do the analytics work and store the supporting data [our] AI solutions,” said Lee.
In January 2021, Scotiabank launched another AI initiative, the Strategic Operating Framework for Insights and Analytics (SOFIA), an AI tool designed to help the bank better understand which retail and commercial customers are being impacted by economic uncertainty are and how they can be operated Predicting cash flow.
Then, in February 2021, Scotiabank launched C.MEE. C.MEE uses AI and Big Data to further improve the customer experience. Using the global AI platform, C.MEE analyzes data across all customer touchpoints to identify the most relevant advice it can give a given customer, and then delivers it through their preferred channels.
By taking signals from client activity, C.MEE continually learns and understands more about their financial behavior and where they are in their lives, improving the relevance of advice, Lee said.
Across all of these projects, “AI is driving more efficiency and better insights and information through our workforce base, ensuring that no matter how much someone decides to use a supported channel or not, they have a much more customized, personalized and relevant channel receives series of offers or services.”
Organizational structure is key to adopting AI
One of the main reasons Lee said Scotiabank’s AI strategy is working is the bank’s organizational structure, with key data and analytics leaders reporting to a common executive.
The bank also has a dedicated CIO aligned with this function who is responsible for the global data and analytics platform. This person also serves as the bank’s liaison with other CIOs throughout the organization. So if the bank needs to integrate AI into different technologies or processes, there’s someone who can act as an “interpreter,” Lee said.
This dedicated CIO “would also connect the legacy systems that we might continue to see across the bank with our more modern hybrid infrastructure and capabilities that would come alongside an AI engine or model,” she said. This person “also helps set those requirements in a way that balances the old and the new, and ensures we’re making the right tradeoffs to have some impact for our customers and our employees.”
Scotiabank’s three-legged stool of data, analytics and technology for AI
This three-legged stool of data, analytics, and technology was critical to the bank’s adoption of AI, Lee said. “It’s less a question of ability and more a question of the operating model, but making sure we’re practical but also ambitious and ambitious has helped us a lot [that AI is] Getting integrated into those technology teams and making sure we have the right data pipelines to make it sustainable,” she said. “We built ours [AI] models in a way that respects both. It really is a real partnership between these three groups.”
Because Lee’s team requires such a large amount of data to create these AI models and AI-based processes, this “handshake” between data and analytics is extremely important to ensure that when the team sees it from an AI perspective, Modeling needs are joined at the waist with data partners and prioritized which data pipelines need to be built. These teams work together to ensure analytics teams across the bank have access to high-quality, well-managed data, Lee said.
“We’ve stumbled a few times in our past trying to do AI without that partnership with data,” she said. “From a data availability perspective, we could potentially collect enough data to build the model in the first place. But in terms of maintaining it and being able to use it for ongoing process automation or marketing automation or whatever, that became such a resource-intensive, difficult and error-prone process.”
Scotiabank learned that lesson the hard way: through some early failures. What started out as a great idea that Lee’s team believed a model could be built on proved untenable from a sustainability and execution perspective. But “by working better with data and technology, analytical models suddenly become not only buildable, but sustainable,” she said.