Getty Images/iStockphoto

Build or buy fintech AI? To Mastercard, the answer is 'both'

Mastercard's operations CTO discusses the company's data principles, global contact center outlay, and its rubric for building vs. buying AI -- not just the generative kind.

Enterprise IT organizations have so many questions about whether to build or buy AI tools that create content, manage workflows, search and summarize, monitor network activity for security purposes, find sales and marketing opportunities … the list goes on and on. Calculating the costs of customizing a ready-made tool vs. coding it yourself becomes a difficult, multivariate math problem IT must solve with the finance team.

George Maddaloni, executive vice president and CTO of operations at Mastercard, is in such a position. Mastercard has built some of its own fintech AI to power its massive global B2B financial transaction network that covers 800 different products in 186 countries and 73 languages; other tech is bought. We discussed with Maddaloni how Mastercard makes these decisions and what large enterprise customers like his want to see from their cloud vendors.

Talk about the AI project Mastercard built to support its customer experience.

George Maddaloni: Mastercard is a big network provider. We serve B2B customers, banks and financial institutions. Through those customers, we also serve consumers around the world. We deal with tons of calls from end consumers that we handle on behalf of our customers.

The application of AI in this tool was around routing of these requests -- how we deal with them in the myriad ways they hit us. As they come through various channels, how do we get them routed to the right places and, therefore, make our contact centers much more effective? We built a routing engine, enabled by AI, which would enable us to get things to the right place and eliminate a lot of time that it would take for us to sort through handling inquiries. It gives back time to the customer to get an answer or solve an issue.

So it's more rules-based and machine learning AI than generative AI

Maddaloni: Generative AI was incorporated for language interpretation because we're dealing with emails in some cases, and we had to deal with multiple languages. There was a rules element and an AI model that was enabling those rules to be applied. And then [we mapped] all that into a contact center that has both new and legacy components. We patented this in terms of the application of both generative AI techniques and [other] AI techniques and applying it into an existing contact center infrastructure.

Describe the scale of your contact center operations.

Maddaloni: We have part in the United States and part in Asia. Whenever you have the scale of a Mastercard, technology is going to be in a state of 'modernization as you go.' There are a lot of different components depending on which service and which call center handles it.

What did your tech vendors not have that kept you on the build-not-buy path?

George Maddaloni, MicrosoftGeorge Maddaloni

Maddaloni: Every business will have its own nuances in terms of what they need to do with technology. We certainly still buy a lot of the capability off the shelf, no doubt. But the types of challenges and the types of inquiries we get are B2C -- in some cases, we get the B2B. Creating something that could interoperate between those really helps route [service] cases.

We have a certain number of languages we have to support. And we have this kind of language to the business, in terms of credit cards, debit cards, the kinds of transactions, sometimes loyalty program requests we get.

I would expect when you talk to anyone, they would say, "Yeah, we'd like to use these technologies and string them together."

But their actual application, especially when it comes to AI, we need to have control over … especially when you're dealing with a customer. You need that context, and there's just no way you're going to buy the context out of the box.

You patented this process. Is Mastercard going into the software business?

Maddaloni: We do patents around the business all the time to just make sure we're protecting the technology that we currently have. We've had a decades-long history of developing working with technology. Protecting our intellectual property is key. We've got a long-standing patent program. We're pleased that we were able to put an application in for this one as well, and we'll see what comes of it.

What do AI technology vendors need to develop to help B2B companies solve their customer service cases, which are less straightforward than B2C?

Maddaloni: You're right about the complexity of what we're trying to solve in the the B2B market. Hence figuring out how to route calls and error requests are key. How do you apply the language of the business into how cases can be handled? It becomes less about this customer and their transactional request at the time and more about how you are handling a case and the complexity associated with that business. And you need that context and information to be able to feed it back into the problem you're solving or the request you're going to fulfill.

Will generative AI make it easier for these more complex B2B questions to be solved with automation?

Maddaloni: I think that generative part of it is around getting the language piece right and the applicability -- how do I incorporate that into a product of some kind?

We used generative capabilities in a product we launched separately from the contact center called Decision Intelligence Pro, which goes after fraud prevention. Fraud prevention has always been AI-enabled. But we've been able to use generative AI techniques and the neural network to bring down the time it takes for us to detect fraud and detect more cases. We build into the [fraud] scoring service we offer our [banking and financial services] customers.

That's an example where we can begin to build products with generative AI techniques and make a service or product more effective.

What advice do you have for enterprises evaluating AI right now so they don't buy something they will regret later?

The other important thing we've that we've thought about from an AI perspective is using our Data & Tech Responsibility Principles in any new technology that we're bringing forward.
George MaddaloniExecutive vice president and CTO of operations, Mastercard

Maddaloni: I think the way we've approached technology decisions is to make sure that we keep it in a layered approach. To manage all these calls, you need a contact center platform, you need a case management platform and you need those capabilities. As we thought about the use of AI in that context, we didn't fully lock into one AI platform for an extended period. We have our data layer; we have our call routing layer and certainly our AI approach on top of it. It's really making sure we have this way to approach our architecture in order to avoid kind of an [AI vendor] lock-in that we were not able to move out of easily.

The other important thing we've that we've thought about from an AI perspective is using our data and tech responsibility principles in any new technology that we're bringing forward. One of the key principles is customer data privacy -- rules that we have to adhere to and make sure that we're fulfilling all of those obligations as well as eliminating bias in any of the data that we're consuming. All the data principles are key, especially those two.

Regarding generative AI, what do you want from tech vendors that you haven't seen yet?

Maddaloni: It's getting better and better over time. We've seen the results where vendors have their AI capabilities scored against benchmark tests and they're improving them over time. Certainly, performance is one thing. And No. 2 is controls -- making sure that we've got the ability to apply our data boundaries and data lineage techniques. As we run a global business, that is extremely important.

While we talk to a lot of vendors about what they might want to do with a subscription for a data SaaS model, we push hard on those data principles we have an obligation to fulfill to our customers. We make sure that they're incorporating them. Most of them are willing to listen to us and take that on board. And they're getting the same types of requests from other customers as well.

Editor's note: This Q&A was edited for clarity and brevity.

Don Fluckinger is a senior news writer for TechTarget Editorial. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.

Dig Deeper on Customer experience management

Content Management
Unified Communications
Data Management
Enterprise AI