AI Cannot Solve a Data Problem | Rohit Chauhan, Former EVP at Mastercard

0:05 Welcome everyone to CEOs magazine series of one-on-one interviews with cos data leaders and key influencers. I'm your host Robert Lutton, vice president sandal consultants and I'm coming to you today from Toronto Canada working with a co magazine. Today I have the pleasure of introducing Roett Chahan for who formerly served for approximately 20 years as EVP of fraud and AI at Mastercard where he led a global strategy of execution for next generation fraud detection platforms powered by AI graph intelligence and advanced analytics.

0:43 This will be one of three special interviews starting today. And today we'll be talking about connected intelligence. How graph and AI are redefining fraud identity and trust. Welcome ro. >> Thank you. Thank you for having me. >> Pleasure is all ours. We really do appreciate you given some time today. I wonder if we could start off with something that you know has been maybe on the minds of many people in the financial industry. For years, the conversation around fraud has focused on building better AI models.

1:19 Do you believe the industry has reached a point where data architecture is becoming the bigger differentiator? Robin, it's true. It's very true. And and the reason is that the fraud of today is much different than fraud of 20 years ago. And fraud is getting more and more sophisticated. And to fight this fraud, institutions have to leverage every piece of data that they have. And to be able to orchestrate that in a seamless fashion can only get solved through an architectural solution. And hence it is the architecture that actually is the center stage of these discussions. right now >> I can see and you're quite right you know the differentiation over the past 20 years has been amazing about you know how the fraudsters are are out there so we certainly look forward to hearing more about this now I' I've noticed you've spent almost two decades building AI systems that have protected billions of payment transactions looking back has the fraud problem fundamentally changed or Has our architecture simply failed to keep up?

2:39 It's wrong to say that the model failed because model is going to be as good as the data it sees. >> Right? >> So the responsibility now is the onus is on institutions to make sure that when the models are being built they get they provide the perfect data set or as complete of a data set as possible. And it is not the models that are failing. It is our ability to provide the right data set in the right fashion which is what is I feel the industry is lacking right now >> and your premise is that hopefully architecture will will address that.

3:26 so we know that banks have invested billions in better fraud models. Why do you believe the biggest limitation today isn't the model itself but the architecture underneath it? So Robert, what happens is that if the model starts degrading and or is not performing as as good as it has performed in the past, the first knee-jerk reaction is that let's apply even more sophisticated sophisticated AI to see if we can improve the model.

4:01 But if the underlying data remains the same then the model leveraging much more sophisticated AI is going to be only marginally better. The way to get a you know a leap forward a huge step forward is more on how the data what kind of data is presented to the model because partial picture of the issue is going to even just create a partial solution using AI. AI cannot solve the data problem. The data problem has to be solved independently and then only AI can be leveraged to create a solution that will be robust and can stand fix. We we we hear that a lot fix the data problems first and then AI will work magic on top of it. So that's great to hear that alignment does that.

4:58 ro when you talk about you know how AI is helping banks detect fraud faster but it's also the reverse is true that it's helping fraudsters move faster so when adversaries are now operating at machine speed in your opinion what has to change in the way banks think about their data architecture so Robert you have to understand that EI in the past was very expensive to operate and to even access. But now it is readily available to everybody and hence even you know small operators leveraging AI can start launching very sophisticated fraud attacks on institutions and institutions are only playing defense. So it's it's a asymmetric war.

6:00 Fraudsters are constantly bombarding with fraud attacks hoping that one fraud will get through and they are ahead of the curve. And for institutions to win they have to stop every fraud. Right. >> And the only way the only advantage institutions have over fraudsters because the technology is exactly the same on both sides is that if they sit on a superior data set, they have a better chance of stopping fraud. And hence I'm going back to the same thing that the architecture of how the data is is connected and stored and is made available to the model becomes like a non-negotiable in actually winning this war.

6:47 Well, you know, we win the battle. I don't know whether we win the war war. I certainly hope we do because I mean the the way that these fraudsters and adversaries are are out there hitting these banks and these corporations and it's not just financial institutions that are suffering the same issue. I wonder if you can maybe just expand upon a thought that you had there. You often say that fraud isn't a transactional problem, it's a network problem. What are the traditional fraud systems missing when they evaluate individual events instead of connected patterns across entities, devices, accounts and behaviors? What's your thoughts on that?

7:30 So Robert like in the past fraudsters will you know steal a credit card data set or whatever the case might be and then make big transactions in buying huge things a Rolex watch or whatever so that they can just get away with their prize of of stealing >> right >> however the the the kind of fraud that happens nowadays the transaction look extremely kosher in isolation you look at it and it'll be like a grocery bill for you know 127 hours and 13 sadness >> right >> and like you look at it and it doesn't look you know it looks perfectly kosher in isolation but once you start looking at the overarching picture then only you start revealing a different truth that kind of emerges to identify that this really is very suspicious. For example, if you know you make a a grocery transactions in New York City at like at 1:00 in the afternoon only to discover that a coffee was bought in San Francisco 30 minutes ago, >> right?

8:39 >> You just know that it it's just not possible for somebody's not right there. Yeah. >> Somebody to kind of just move from San Francisco to to New York like and then and and be, you know, make two transaction present. These are not online transactions. your card present transaction and boom, the context just told you that this is so in the past isolated transaction were enough to to to to weed out fraud because they would just look suspicious because they would do things that would just allow them to just run away with as much as they could in a single transactions. But now I a a transaction by itself may look all kosher, all n you know all perfect but it's only the context that allows you to to surface you know a different reality that may be that may be the truth. you know, Rohead, thank you for explaining that because I was wondering how one could, you know, move from the individual event to the connected patterns and and you know, when you when it's explained like that, how can two close transactions maybe separated apart by distance or location, whatever, or similar type of things that it looks for in a connected patterns. it begins to make sense about how you know it can detect these fraud patterns. So if you're advising chief data officers or chief AI officers, what's the one architectural shift they should make to stay ahead over the next 3 to 5 years?

10:16 So it is it it is clear that AI is is a monumentally powerful technology that can do a lot of good for an institution. However, the strength of your AI will be dependent on the foundational strength of your data architecture. If you sit on a weak data architecture, your AI will not be able to you know to leverage the tech the the tech you will institutions will not be able to leverage the technology and its benefit to its maximum if they are operating out of an inefficient data infrastructure and hence it investing in this infrastructure sooner is going to have big payouts into the future.

11:12 >> you heard it here first, the foundational data architecture. Nice little nugget there. Rohit, we appreciate you providing your time today and we look forward to our next interview on architecting AI that banks can trust. >> Thank you. Thank you, Robert. Thank you for having me. >> You're welcome. And for our listeners, please visit cdmagazine.te for additional interviews.