Interview with Young Pham, Chief Strategy Officer, CI&T

Why Dynamic AI Modeling Using Diverse Data Continues to be a Game Changer

Already an influential force in digital finance, the ever-expanding ability of artificial intelligence and big data models to analyze more diverse data sources—including social media, spending habits, transaction history, alternative data and more—in real time gives banks and fintech players the ability to assess risk through a more dynamic lens and optimize operational efficiency. Experts outline the expanding, powerful knock-on effects, including enhanced risk management through more accurate credit scoring and credit extension, better machine-learning-driven fraud detection and prevention, and the creation of richer customer profiles that enable "instant" insight-based decision making and customized product offerings based on individual behaviors.


Transcription:

Bailey Reutzel (00:12):

All right. We are going to get started here in the AI and cloud computing room. Just by a show of hands, who here thinks AI is evil? Okay. Okay, good, good, good. What about who thinks AI is going to become a God and rule all our lives? Okay, well, you did both, which is fine. It's fine. Okay, so we've got some levelheaded folks in here. That's cool. That's cool. I feel like most of the conversations, I'm going to start with grounding this conversation a little since, I don't know, four to five years ago with the advent of ChatGPT and all the generative ai, the conversations have been pretty black and white, especially with my friends who are not in the AI industry or in the banking industry, they tend to think either one way or the other. I think these conversations are becoming a little more nuanced, but I just sort of want to ground it in. What is one use case that you are seeing that's really working wonders?

Young Pham (01:16):

Right out of the gate, Bailey?

Bailey Reutzel (01:17):

I think so,

Young Pham (01:17):

Jumping right in. One of the big use cases we see, and I know we're talking about it specific to the areas of banking, but we actually even see a lot of it in terms of what's propelling the growth of digital, which is the use case behind how software development is accelerated with the use of ai. And so it's not just how the things that are impacting the running of the banks or things in the customer service piece, but the ability to regenerate code is actually building all these new capabilities. And that's the exciting part for AI that we're really seeing because it's a thing, particularly when we talk about the dynamic modeling, it's just going to get faster and it's just going to get better. And so you'll be able to generate more software, more AI, and then it'll just start going even faster.

Bailey Reutzel (02:12):

This does seem sort of evil though, just generating AI after AI, and I think who here watches sci-fi? Dystopic, sci-fi reads, dystopic sci-fi. I don't know why there are not more hands up. This is very confusing here in the middle of the room. I think that you guys should let me be clear. Definitely watch those for the worst case scenarios. And are you a software developer?

Young Pham (02:37):

I am a reform software developer. I'm definitely more on the business product side, and how we apply that into the world of financial services,

Bailey Reutzel (02:47):

Okay

Young Pham (02:47):

That's really what CI&T focuses on.

Bailey Reutzel (02:49):

Yeah, wonderful. I realize I didn't even introduce you. This is Young fam, he's with CI&T. It's up here on the screen. You guys know what's going on. I'm Bailey. I want to get a little more specific in terms of that software development use case you're talking about. Do you have any really specific, like this developer was coding up X, Y, Z banking app and it took him or her five months versus now two weeks or something.

Young Pham (03:20):

And talking about specific, and particularly in banking, we're currently working with it's large Latin America bank, and what we're doing is actually doing a core migration and taking existing COBAL code and then modernizing it and really doing it something that would've taken five years in 12 months. So you asked a question, we're seeing developers that are doing 50% greater productivity, but even organizationally now that's even a new capabilities. Think about being sort of the third rail on a train that you would electrocute yourself. You don't touch the bank core, but now you can. So now you have all these extended capabilities. Now that it's modernized to start working in other concepts. How do I display, how do I push that out? How do I merge all this data sets with it? And even starting finding other greater utilities to embed what they do as a bank into other parts of the environment.

Bailey Reutzel (04:22):

Is some of this interest in adopting ai, is that largely coming from your software development crew within a bank or the business side?

Young Pham (04:32):

No, I think it's across the board and I think there are conversations that we have all throughout the organizations and particularly with banks that we talk to. There's actually different types of conversations. And what I would say is every banking executive that we know talks about the idea of how it's going to transform their organization. They may not know exactly what that endpoint is, but it is, I may be able to change loan servicing. I don't know what that implies, but it's a fundamental shift to my business now on the other end where the people that are actually doing the day-to-day, we talked about software development, there's a little bit different conversation. It's about efficiency, it's about adaptability, it's about the change in their lifestyle and not just them, but also people in the underwriting world or that. So it's a very different conversation. You talked about the scariness of it all, and that's a big barrier. We do not want people to be feeling like they're going to be replaced, and it's the big balance particularly that we're having within organizations.

Bailey Reutzel (05:48):

So there's two questions that I'll follow up with this with sort of the fear. I think you have seen it in some industries that shiny new toy becomes available. A tech entrepreneur with the shiny toy goes into let's say the media business and says, this is going to replace all your low level news reporters. You implement that ai and the AI kind of screws it up a little, which no one should be surprised about, but you do need some human intervention before they realize that they screwed up, they fire people. So I think you do have, it's not unfounded, this worry that AI could replace. So I guess how are you thinking about that and how are you talking to these decision makers about adopting AI without that maybe?

Young Pham (06:38):

Yeah, I think the area we spend a lot of time talking about is this idea of how change management is fundamentally different. So yes, things are more efficient, but that does not necessarily mean a reduction in headcount. It is can you do more? Can you retrain some aspects of the workload that they may make a better decision making? So when you look at the promise of ai, everyone looks at the idea that it's just on the efficiency end. The other big benefits are the fact that there's better decision making out there if you use it in the context of humans, that's a very important thing. The other things are how the customer experience gets better. It's also overall human productivity and that's a big benefit around it. And so I think that is the concern. I think organizations may sometimes jump to the first part, which is how can I take advantage of it and really be shortsighted on that versus also when I talk about the change management, if I want to benefit from all this, you're going to grow. You have this human capital and this workforce, how are you going to change your loan officers to think different? How are you going to change your commercial bankers to do different or your software engineers? And I think that's the exciting part that banks should really look to improve upon.

Bailey Reutzel (08:13):

And I think, I don't want to sound like a total AI skeptic. I do think there are some AI implementations that are evil, but I also use AI to help me be more productive of, I think the thing. So again, there's sort of this lack of nuance sometimes in what a large group of the population hears about ai. So I'm wondering just those decision makers that you're talking to about don't jump to cut your whole workforce, are you seeing that they're trying to get the whole, what is the word, the whole team involved down from the lowest level person to the highest level, and how are those conversations going?

Young Pham (08:55):

Yeah, and again, I talk about the areas beyond the change management. One of the big things we talk about is that there are really three principles that we look at when we say organizations you really have to think about. One is how do you implement an AI first mindset? So if you wake up in the morning, what's that thing that you do that helps enable your decision making that helps prioritize the things you need to do? And so it's a starting point. That's one of the basic principles that helps adoption, that makes it less scary. And then with that, we see a lot of people really embrace it. The other principle that we really talk about is adaptability. I think the reality is that people's jobs are going to change and there's no end point to it. So we talk about the creation of adaptable teams.

(09:47):

That's actually a new skill set because the more you're able to do, the more functionally your role changes. Am I creating new prompts? Am I doing things fundamentally different? And so that's a new skill. I think organizations are starting to learn with their people where it's don't be fearful. I'm going to teach you how to build adaptable cultures and adaptable teams. And then the last piece is the piece that actually is the easiest. It's the tools and the technology, which honestly I think everyone's here spending a lot of time and experimenting about. That'll continue to change. But to me, if you don't hit those other two areas, that's how you get everyone at every level to think about it.

Bailey Reutzel (10:28):

Yeah, it's interesting because I think I am using AI tools or even just think about any tool that you are using in general to sort of manage your time. I'm sort of using those in the hopes that nothing else fills that time that I've saved. I don't want to work as much, so I'm going to use these tools to help me not do that. So I guess, yeah, I worry that if I tell people that I'm using AI and saving time, that they will now fill that time with other stuff and that worries me.

Young Pham (11:03):

Yeah, I think that may be a lifestyle or just a prioritization issue. Bailey, I can't help you on that one.

Bailey Reutzel (11:08):

Yeah, fair enough. Okay, so we're talking about AI modeling. We're talking about AI modeling in terms of data, I think this is specifically I think a really great use case for AI because a human just can't work through the amount of data that an AI can or a machine can. So I guess dig in a little bit about what kinds of data banks are feeding that AI to help a use case.

Young Pham (11:38):

And I'll go one further and start with one of the use cases. It's actually not even what they're feeding, but what they're creating. So with ai, the ability to do synthetic data to actually generate huge larger data sets so that I can do more testing and modeling around. So

Bailey Reutzel (11:56):

Sorry, say that again. Testing and what?

Young Pham (11:58):

Testing and more modeling around it. So if I want to create more loan parameters, how can I build a product and generate synthetic data so that I can feed it into that before it goes live? And so just the creation of vast amounts of data, we didn't have that capability before. So that's certainly a use case. I think people underestimate, it's always the aspect of, Hey, what do I have? There's also actually I can create new who whole data sets and change the way I do fraud or risk parameters around that. I can generate different risk scenarios that I wasn't able to do years ago and do it seamlessly. So at least on that end, the ability to take synthetic data and generate new scenarios, huge game changer for the ability, particularly around the data complexity. The other areas obviously are the existing data that you have being able to layer in the financial information and really make better decisions.

(13:01):

But the other thing which I think with the advent of more embedded, vetted finance is taking data at the different sources. Whereas a bank I'm already starting to push in. So if I'm lending on Autotrader or I'm doing auto lending and things like that, can I take information about the automobiles of social, all those other aspects to run into my credit decision. So it may not just be the information I have as a financial institution, but now that I'm embedding my services out there, I think it's a lot of traction that we're starting to see there where people are having some creativity to change the way that they're actually looking at how to acquire a customer.

Bailey Reutzel (13:47):

So well, let me do the fraud stuff first. I do think the fraud is a really nice use case. Can you give us a specific example? Are the banks they're modeling that kind of modeling those threat that vectors, and then are they actually in practice doing anything with their real systems? Does that make sense what I'm asking?

Young Pham (14:09):

Yeah, yeah, it does. And I would say there's different fraud models that historically banks have applied sampling projective based off of different based scenarios. But now what you're able to do is generate significantly larger amounts. So you're actually not necessarily running sample models off of it. You're actually generating enough data sets to where they can be more actualized. So it's a more accurate predictor. I think what is interesting, particularly in the case of fraud is that with more data historically where FIS have been maybe able to share that either through EWS or some of these other organizations, what do I make available now that I have much more available data and whether or not there's still data sharing between the different fis, particularly at a customer level to see if fraud happens. So I think that's probably, to me, I would restate what you've said is that's the area where I think governance and really understanding where you should be sharing data between FIS becomes a little bit. Is that something where it's helpful if we all do it so that we can more accurately predict fraud across the 4,000 plus banks and credit unions out there, or am I still trying to do it on my own because I have so much data now?

Bailey Reutzel (15:35):

Okay. So are you seeing banks largely want to do it on their own or is there some effort to share some of that data?

Young Pham (15:42):

I would ask the bankers. I'm always,

Bailey Reutzel (15:43):

Okay, okay.

Young Pham (15:43):

More of an optimist. I worked at Chase back in the day. I've worked at, we consult for other FIS now and there is a general inclination for more data sharing across the industry. We've seen that more now than what it was 10, 20 years ago. I'd like to see that trend continue because I think having access to that data more and it's going to be vital, you can do more basically off of that. And particularly if you can do that with each other as opposed to going through, and I'm sorry if the credit guys are out here, the experience, the TransUnion and those guys, but being able to share directly and start to move more towards that, whether it's in a data clean room or some of those other aspects, it's going to be much more healthier than working through the traditional data sets that are being provided through those platforms.

Bailey Reutzel (16:43):

Yeah, I see mean, is there sort of data security issues that arise in that scenario?

Young Pham (16:51):

Yeah, I mean there's always security, but I think that's by nature, pretty tackled it at this point. I think beyond the security anonymization of data is important and anonymization of data, particularly if you're making credit decisions and running scoring against that, that becomes the interesting use case there.

Bailey Reutzel (17:18):

Okay, cool. I'm glad you brought that up again, where I want to go next. More dystopia here for you. Yeah, so I think I'm a little bit skeptical about using different data than we already use for credit decisions. I think this becomes sort of like a China social credit system scenario or you could envision it becoming that way. So I guess the first question, well, yeah, the first question is what data are we thinking about using to make some of those credit decisions? And then we'll go from there.

Young Pham (17:55):

Yeah, I would bucket it in a couple areas. I think consumer data is in a very interesting one because the established ability to run credit decisions around there, I think people are starting to layer in things like social information and other things, not specific to possibly a lending decision or a credit card decision, but still looking at general preference for marketing and things like that. So in that areas, we're seeing that being applied there. In the small business world, that has always been the ability to bring in other sets of data, particularly for decisioning because it's less standardized. And so when we look at all the fis out there, and there are specialty niche, whether you're working with cash and carry small businesses or if you're working with ag for sure, the credit criteria is fundamentally different. So bringing in information, we work with an nefi that's really tied to agricultural lending. That's a lot of the small businesses and now seasonality, crop forecast, those things actually feed into that to make a better credit decision than just, Hey, how's your revenue been the last three or four years? So those are some of the cool things we're starting to see on that end. And I see it much more in the small business and commercial side now.

Bailey Reutzel (19:16):

I see. Yeah. I don't know if any of you could possibly see a dystopic, stoic example out there with that. I'm hearing you talk about crops. If the crops, if we're looking at weather forecasting data to potentially lend out money to a farmer and the weather forecast is bad, they're not getting a loan now, it's worrisome, I guess is what I'm saying.

Young Pham (19:41):

I wouldn't say they wouldn't get a loan. I think they would just price the risk accordingly.

Bailey Reutzel (19:47):

I see.

Young Pham (19:47):

Into their rate.

Bailey Reutzel (19:48):

Yeah,

Young Pham (19:48):

Or maybe they don't get the loan, but

Bailey Reutzel (19:51):

Yeah, but evil,

Young Pham (19:53):

You're so negative. You're so negative.

Bailey Reutzel (19:55):

I know, I really do. I

Young Pham (19:56):

Power for good not to end the world.

Bailey Reutzel (19:58):

Yes, I know. And see, I agree with that. I'm always the skeptic. I have to bring these questions up so that we can talk through them so it doesn't happen. That's the idea here. Yeah, I think because the other thing is there is talk about bringing sort of social data into credit scoring. So I'm thinking latest TikTok posts, I don't know, some of y'all's Facebook posts, I might want to check that, right? If I'm on TikTok being like yolo, I just bought a $200 set of nails with my bank account at zero, I don't want the bank to see that personally. So I guess how are you thinking about bringing in social data as well?

Young Pham (20:40):

Yeah, and I would look at it a couple of key functions. I think I am not a regulatory expert, but FCRA would probably kill me if I said go put social media into the ability to apply credit decisions in there. But what's interesting, if you look at concepts like buy now, pay later, which may not be a full credit underwriting but may be tied to, again, embedded finance or how we apply payments into a retail case study or a retail use case or something like that, that FIS do power into. That may be a very interesting thing to be able to look at someone's social media feed and say, Hey, I've been going on vacation to wherever that I can't afford over the last six months or so. Right.

Bailey Reutzel (21:35):

Yeah, it gets a bit big brothery in ways. Yeah. Okay. I'll change gears. We don't have to keep going negative. I'm sorry. Yeah, I think the idea of using big data, it appeals to me less in this sort of credit sense. It appeals to me more in the fact that you could maybe make new products that are very customized to specific segments of your audience. In this way, I think we are expanding the opportunities for more people, whereas some of the credit stuff feels like limiting opportunities. So I guess talk to us about that. How have you seen banks or FIS utilize this big data and these AI running these big data models to create new products?

Young Pham (22:20):

Yeah, again, I talk about the small business and commercial use cases, guys. I think that's been a very underserved segment in banking for years. And because there's a lot of businesses that sort of come up. We haven't applied it in AI context, but we work with an FI that really focuses on small businesses, but they have a subsegment that they market to women entrepreneurs. And what's interesting when you go through that is that there are less of a credit risk, but there's not maybe as established history. So creating new products may not be a securitized loan or a different credit extension or other things that may be very unique to that audience, I think has really high potential based off of information that, and I talk about entrepreneurs because there's not a lot of revenue history, there's not a lot of other information that's available. So you may scour social media, you may see that this person is presenting at Finovate or some of these other events. So you start to make the distinction that, hey, maybe there's something there and that's going to be interesting. That's a non-conforming product that would be very unique to build out, but that could really own a market segment. And so that's a good example. I think I'm excited if anyone wants to do that. I'm a big proponent of that.

Bailey Reutzel (23:52):

Yeah. Yeah, that'd be awesome. I mean, do you see this eventually in the future becoming sort of customized by the individual? It seems kind of wild, but you could see a future whereby I am marketed to very specifically,

Young Pham (24:09):

From a marketing context, it's starting to happen now or products as well. So from a context of configured products, I think that future is really, really, really now. I think more and personalized experiences in general are going to be pretty frequent throughout. And so we see it on the servicing side, we see it in the other areas, particularly on the marketing side. I think specific products, I think, again, to my point earlier, credit products are hard. So I think it may be interesting when you start to look at banks and how they specialize in the segments they work in or if they're working with different distribution channels. I'm a big proponent also of amid finance. Then you have the ability to customize specific to that use case. So it may be in an automobile marketplace, it may be in insurance where people are buying a house and be able to display through, so you're able to create a different term for that specific product. And so I think those products where there's a little bit more flexibility than pure credit or residential mortgages.

Bailey Reutzel (25:24):

Yeah. Cool. We have probably four minutes now if anybody has a question. We do have a mic runner. We have a question here up front. Yeah, we'll take that.

Young Pham (25:38):

And please don't ask about the end of the world.

Bailey Reutzel (25:41):

Got that covered Subject matter expert, please be positive.

Audience Member 1 (25:45):

So actually well along those lines, I was just curious. I know you're talking about the dystopian potential for this technology.

Bailey Reutzel (25:53):

He's going for it.

Audience Member 1 (25:54):

But the other flip side of that is that we've had in our past, particularly in this country, a lot of issues like redlining, things that happen via human forms of keeping people out of finance. And I think it's important to weigh the pros and cons of how can we use this technology to make sure that it is for good and honestly the dystopian future. In other words, we're getting more data, more data via ai. What can humans do to make sure we sort of evolve? Right?

Bailey Reutzel (26:23):

Yeah, no, I was just going to say I should have asked that question. It's a really great, great question.

Young Pham (26:29):

I think ethics and AI are fundamental, and so organizations that we work with, that's an underlying thing that we talk through of the ethics of what's good or bad, how does it infect employee, how does it infect the customer? How do you make clear ethical decisions, particularly for running simulations also. And that's actually a legal requirement as well, having transparency and fair decisions in that. I think what I don't know, and what's interesting is that particularly for banks, is this self-regulated or at some point the regulators go, actually, we're going to make that determination. And I think more than any other industry, this is the one that other than healthcare life sciences, this is the one where the regulators are actually going to make everyone move in the same direction. And so that's the only thing I would say. I think there's a genuine willingness, but the area that's to be determined is whether or not the Fed makes a decision or

Bailey Reutzel (27:45):

I think this gets into, I'm going to ask the last question. Sorry. I think this gets into governance a little bit in terms of, wait, did I lose the plot of my question? Nope. It's not governance, it's AI explainability. I think it gets into that a little bit because the idea is that with these huge vast data sets, these LLMs, it is going to be hard for humans, probably impossible for humans to really be able to understand how the AI made that decision. And so if that is the case, are we sure that they are making the right decision without a bunch of human bias that went in And then I don't know any other AI bias. So I think, how do you think about that as well?

Young Pham (28:28):

Right now, there are platform tools that we're seeing, particularly in financial services, one on the marketing end. So for dynamic experiences, being able to report that and create traceability in that. There are some cool tools that we've seen out that that's a new product because of the fact that the experience is dynamic. I think it's starting to, it will take a bigger place also in the origination process as those decisions. If they become way more dynamic and we move more towards an agentic process, then yes, there's going to be traceability built into that. I think that's the only way that you can provide transparency, but that's something I think a lot of organizations don't think about. They just think about, Hey, I'm going to roll this out and it's going to make things faster. Well, actually now there's a new condition, which whether it's the customer or the regulators come in and says, I need to be able to see how you made those decisions, which I think a lot of organizations are not thinking about it.

Bailey Reutzel (29:31):

Yeah, that's fair. Okay, I think that's all the time we have. Thank you so much young fam, you went through the ringer with me up here and with our audience questions, so give him a round of applause and we'll see you back here soon.

Young Pham (29:42):

Thanks everyone.