Blog: CIBC Is Not Married to Homegrown AI Models — They Want the Best Capabilities for Their Clients

Terry Hickey, Chief Analytics Officer, CIBC

CIBC is using AI to improve the client experience, reduce risk, and increase efficiency. When deciding to use in-house vs. third party tools, they always run an assessment to determine what will best deliver against the objective and has the best capability. For example, to support recruiting, they looked at building an AI-enabled resume screening app but found several external vendors with strong existing solutions. Conversely, the bank talked to vendors about building CIBC’s Clientnomics model but opted to keep it in-house versus using an external cloud-based platform. A current area of interest is model management. In the future, Terry sees the focus shifting towards productionizing models and getting scale. Takeaways for startups — be the best tool for your stated capability and industry (vs. a tool that can do 5 things for 5 industries decently well) and be able to work on the cloud and on prem.

Amanda: Can you start off by describing your role at CIBC?

Terry: I lead the analytics team at CIBC — one of the top five banks in Canada. Our team helps the bank use analytics and AI to enhance the client experience, manage risk, and increase efficiencies. We set standards, select tools, and essentially enable the bank to perform better using data and analytics.

Amanda: So any kind of AI or big data initiatives fall under your team, or do product-specific teams have leads that will sometimes take those on?

Terry: It depends. Some teams have the skills to do more advanced analytics projects which might include predictive modeling or artificial intelligence, but those folks are mainly in the risk and audit areas, so a small subset of the organization.

Amanda: What’s been the strongest commercial application of AI deployed at CIBC?

Terry: Quite a few areas — from helping optimize investments in technology to finding ways to improve the client journey and experience. Using AI to answer questions like: Is the best time to call our client? How should we interact with them? Is it through email, a phone call, or meeting in-person at their home? What banking solutions best fit their needs and financial goals? A lot of this is about personalizing the experience for our clients.

We have models that can predict why someone’s going to contact us — like when they need a mortgage, to question a transaction, or need to raise an issue. With these models, we can proactively get ahead of these inquiries.

Amanda: Are most of these models built in-house or are you using outside vendors?

Terry: For all the use cases that we go after we create a business case and go to market to see if there are things out there that fit the bill. For example, resume screening. We could have built the resume screening application ourselves but when we went to market we found that there were multiple organizations with capabilities in that space. It made more sense for us to engage an external vendor where we can buy it for less, with on par functionality, than to leverage internal resources and build our own.

Amanda: That makes sense — how about on customer facing / revenue generating applications. Does CIBC build those in-house because they are thought to be differentiators?

Terry: It’s a mix which includes working with vendors. This week alone, we’ve had 3 vendors talk to us about how to create models that would allow us to better interact with our clients and ensure we’re helping meet their financial needs. We have a smart team inside of CIBC, but there are also a lot of really bright people out there that have great ideas and we would be foolish not to listen to those if they’re going to help us meet client needs.

Amanda: Can you give an example of a capability you looked at a third party for but ultimately decided to build in-house?

Terry: We have this thing called Clientnomics. It’s a system that enables us to truly understand who our clients are and better service them as a result of having all of that information at our fingertips. Clients today want personalization. They want you to know they emailed yesterday about topic X and then called in today about topic Y and that they have this amount of money in their bank account and respond accordingly.

We evaluated a few different external tools but didn’t find something that could do it at the scale we needed. Everyone we talked to wanted us to send them all of our data and they’d put it in the public cloud and come up with a model. We’re a bank that values our client’s data and understands their privacy concerns. That’s just not going to happen. People hold us to a higher standard than other organizations. That can limit the things that we can do with some of our partners.

Amanda: That’s really valuable feedback. What other advice do you have for AI startups that want to sell to CIBC?

Terry: I think AI is still so nascent that there are opportunities across the board whether it’s in client experience or risk or internal processes or fraud. Frankly, even building the models themselves — the tooling — can be AI-enabled, like leveraging auto machine learning to be able to improve our speed to market for our models.

Don’t bring me a tool that can do five different things for five different industries because that’s probably not going to solve the kind of complex problems that we have. What we want is a tool that is really, really good at resume screening, for example. If it can do A/B/C/D other things, that’s usually more dilutive to the core offering because that means that they’ve taken their eyes off of the initial use case. I want to partner with the best solutions for each problem.

One challenge we have today is with model management. It’s an interesting problem to find a company that has solved that. Track training data, track model versions, and track production.

Amanda: There’s actually a company called Algorithmia that does a lot of what you’re talking about. Worth looking into. Last question from me — when you think 5–10 years down the road, are there AI-related capabilities you’d like to have and don’t today?

Terry: The biggest thing will be seeing these models and applications in production. We’re evolving our strategy for AI and guiding principles, and seeing everything in action will help us build out our plan. One guiding principle we have is around impact, even if it’s only small in the beginning, every project we take on will make a difference in the long run so I’m looking forward to seeing what comes next.

Note: post represents personal opinions of the interviewee and does not necessarily reflect the opinions or positions of CIBC

Source: Artificial Intelligence on Medium

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top

Display your work in a bold & confident manner. Sometimes it’s easy for your creativity to stand out from the crowd.