Data Radicals logo

Gen AI at Work: Inside the Digital Bank Revolution

Dr. Geraldine Wong, Group Chief Data Officer, GXS Bank

Dr. Geraldine Wong

Dr. Geraldine Wong is the Group Chief Data Officer at GXS Bank, a digital bank based in Singapore backed by Grab and Singtel. She leads AI and data strategy efforts to enhance financial inclusion and customer experiences. A recognized tech leader, Dr. Wong has earned accolades like the SG100 Women in Tech award and is active in mentoring and advisory roles.

Dr. Geraldine Wong

Dr. Geraldine Wong

Group Chief Data Officer

GXS Bank

Satyen Sangani

As the Co-founder and CEO of Alation, Satyen lives his passion of empowering a curious and rational world by fundamentally improving the way data consumers, creators, and stewards find, understand, and trust data. Industry insiders call him a visionary entrepreneur. Those who meet him call him warm and down-to-earth. His kids call him “Dad.”

Satyen Sangani

Satyen Sangani

CEO & Co-Founder

Alation

0:00:03.4 Producer 1: Hey, Data Radicals. Welcome back to Season 3. To kick off the season, Satyen sits down with Geraldine Wong, Chief Data Officer at GXS Bank, a digital bank in Singapore. In this conversation, Geraldine reveals how she balances governance and privacy with AI innovation.

0:00:19.9 Producer 1: She also shares how AI can be a game-changer for banks, with fascinating use cases from chatbots to fraud detection. So whether you're an aspiring CDO or just curious about the future of AI, you're gonna wanna hear this.

0:00:30.9 Producer 2: This podcast is brought to you by Alation, a platform that delivers trusted data. AI creators know you can't have trusted AI without trusted data. Today, our customers use Alation to build game-changing AI solutions that streamline productivity and improve the customer experience. Learn more about Alation at alation.com.

0:00:58.2 Satyen Sangani: Our guest today on Data Radicals is special in multiple ways, but one of the ways in which this is special is that we're shooting live from Singapore. And I've got with me today, Dr. Geraldine Wong. Geraldine is the Chief Data Officer at GXS Bank, Singapore's first digital bank. With a PhD in statistics from the University of Adelaide, Dr. Wong has been recognized as one of the top global 100 innovators in data and analytics today. Dr. Wong focuses on leveraging AI and data strategies to drive financial inclusion and innovation in banking, reimagining the customer engagement in the digital economy. Dr. Wong, welcome to Data Radicals.

0:01:04.9 Geraldine Wong: Thank you.

0:01:10.4 Satyen Sangani: So, it's fun to be here in Singapore, and I just marvel at what GXS is and what it represents, because it's something that really doesn't have a similar sort of complement in the Western world. Tell us what GXS is and how you got started with it.

About GXS Bank

0:01:24.7 Geraldine Wong: So, just for the global listeners here, GXS Bank is one of the four digital banking licenses that was given by our regulators back in 2020. We have since launched our CASA or deposits in August '22, and then the FlexiLoan product or our unsecured lending in '23, and followed by the debit cards in November '23 as well. Our aim is really to make banking better and to actually encourage financial mobility and inclusion to the underserved population in Singapore. How this is different is that we are built on a consortium of our parents, Grab and Singtel. So, in your world, Uber would be the equivalent of Grab in Singapore, and Singtel being the largest info telecommunication company in Southeast Asia. So, thinking about that, you're sitting on a ton of data. For a data holder like me, it's pure joy.

0:01:42.3 Geraldine Wong: Lots of imagination that can come out of using that data to build the bank. And core to the bank's CVP is about leveraging the data assets from our shareholders, such as Grab and Singtel, to build better products, lower the cost of acquisition, as well as build better credit risk profiling as well. So, when we talk about the data vision, when I first started the data vision, it was about thinking, how do we bring in the data portion and component into the everyday products that we are creating? The touch points, where are we collecting some of this data, to then create that feedback loop again to help personalize some of the services, some products to our customers. Our data vision is actually built on three pillars. I would say, how do we delight the customers with the right risk management? How do we innovate? How do we foster innovation? How do we lower the operational costs and increase risk management with the use of data? So, broadly speaking, these are some of the three pillars that we are using data in GXS.

Data sharing at a digital bank

0:02:00.3 Satyen Sangani: Yeah, here's what's unique and interesting. I mean, you pointed out sort of the birth of it, but it's really the bank's entire business model is born out of essentially the data sharing that occurs from your shareholders, right? Otherwise, there are lots of digital banks globally and there's lots of ways to innovate in banking. But in this case, do you have full license to the data sets that both Singtel and Grab have to offer? Or how does one navigate that data sharing construct?

0:02:05.4 Geraldine Wong: I'm glad you brought it up 'cause it took six months of my time to come up with the agreement and then now we're operationalizing it, we're getting that flow. But if I were to rewind back to October 2020, where we drafted that first tripartite data sharing agreement, it was really very much a legal document, but in there it states down who actually gets consent, who is part of the customer, how is consent obtained in the customer journey, and then following which that this agreement has to dictate what are the right purposes, what are the data sets, who can access those data sets, how is those data sets being protected, how is consent being withdrawn? Because it's really important the ability to withdraw consent and the ability to give consent as well.

0:02:16.4 Geraldine Wong: And who is those stakeholders, who is the data owners? Those are some of the nuances that has to be put in place at the start. And then moving to the second stage, which is really about operationalizing the data sharing. If you think about the three organizations that we are talking about here, a unique person such as Satyen is known by different unique identifier. In a telco world, you're known by your ID or your telephone number. But if you think about a national identifier, one person could take multiple lines. So how do you pinpoint Satyen versus your son, for example? Because they might be belonging to your identifier.

0:02:27.5 Geraldine Wong: And then in the Grab world, it's purely based on mobile phone numbers, whereas in the banks world, it's all about national identifier. So you really have to apply that thinking to say, how do you triangulate what are the unique identifiers you use to triangulate all of this data to ensure that Satyen is Satyen when it comes to me? And at what granularity does this data come across? I mean, you could get the most raw form, or you could get it at the most aggregated format.

0:02:38.4 Geraldine Wong: And this depends on the comfort level of your data sharing partners. And also being able to articulate what's the value of this data that's being shared on an aggregated manner versus a raw level. And thinking about the frequency of the data that's being shared.

0:02:48.9 Geraldine Wong: You know, there are certain data points that will rarely be updated. For example, on the Grab side, right? The number of credit cards that you put on your platform, three versus five. I mean, it's not often that you're gonna update the number of credit cards there.

0:03:00.3 Geraldine Wong: So is there a need for it to be regularly sent to us on a daily basis? It could be only sent to us only when it's updated, right? So there are certain nuances when you operationalize such agreements that, I mean, the devil is always in the details. And what we place a strong emphasis on is that it needs to fit into that purpose. So we are very pedantic about the purpose that you set in there in the project sharing agreement.

0:03:05.2 Satyen Sangani: And so, but of course, with data, there's product development that's going to occur out of understanding people's profiles and what the world says about them and how they're described in the world. And so you don't always know what the purpose is gonna be. I guess maybe you can define it somewhat broadly, like we will be able to offer you new financial products and outcomes.

0:03:20.9 Satyen Sangani: But there's also some level of serendipity and some level of unpredictability in terms of what can and cannot be done. How do you navigate that as you move forward in the bank? And has that ever been a problem?

0:07:00.6 Geraldine Wong: If you're referring back to the data sharing side of things, you're right in the sense that you never know what you're missing out until you explore that data set as well. And we take two approaches to it. The first one being we work a lot with our stakeholders in Grab and Singtel, and they know their data sets the best to tell us because they have worked on topics such as their own credit risk profiling on the Grab and Singtel side, they know who are the best paymasters and who would default.

0:07:25.3 Geraldine Wong: They are the subject matter expertise of their data, so we rely on them for such information when they share it with us as well. The second way we go about it is that we are given very restricted access to their environment in a sandbox, in a very guarded sandbox environment and on-prem. So we are able to explore it for that purpose again. And then discussion-wise, post that discussion, say, hey, actually out of your thousands of data sets, actually I only need 100. And why? 'cause feature selection process tell me that this 100 is gonna be the highest signal for a credit risk profiling.

0:07:56.9 Geraldine Wong: And going through with the stakeholders from both sides, showing them the model and the signals that it comes out from those features that we have explored in their sandbox, give them some form of assurance as well. Yeah.

0:08:10.2 Satyen Sangani: And it's funny 'cause inside of companies, as you well know, it's hard enough to understand the data sets that exist. And you've got this problem, which is you're looking at data sets inside of companies that themselves may not have well-described data. And so you have this like second level of indirection. And now you have this like one unique identifier that allows you to effectively... That you have to generate. It's not even implicitly a part of the data set that allows you to... So that's got to be a tremendous analytics challenge. I mean, you must spend an extraordinary amount of time just validating the data.

0:08:41.7 Geraldine Wong: Yes, it is. I would say that working with both partners who have varying amounts of description on their data and then having to bring it back to us and then putting it into Alation, for example, as a metadata or the data catalog is a challenge in itself. And every time the partners update on their site, imagine they're updating those features without telling us. Then when you find out as a post, right? Yeah.

0:09:03.9 Satyen Sangani: Right. Because they've updated the data pipeline you thought, this data thing is completely invisible. You have no idea where it started. And now all of a sudden...

0:09:10.7 Geraldine Wong: So that's why monitoring the data quality, the data trends is so important, right? If suddenly one day you're seeing the trends that looks like of a certain value and then suddenly it's moved up by a notch, then you need to have that alarm or alert to tell you that, hey, something is wrong there. That the data characteristic has actually changed.

0:09:26.2 Satyen Sangani: Have you ever discovered a problem before they do?

0:09:29.5 Geraldine Wong: Not yet, thankfully. Like I said, I think we have... The fact that we have actually access to their selected or small sandbox allows us to actually play around and be able to discover those anomalies first and then be able to check back with them. And I'm sure in both of our Singtel and Grab shareholders, they have their own challenges that they also are looking at. So it's a lot of interactions with them, a lot of discussions with them to be constantly being updated of what new changes are there as well.

0:09:53.8 Satyen Sangani: Interesting. And so you have done a lot of work, both in sort of traditional data pipeline development. You mentioned that you're building ML models and you've sort of done classic feature development. 

Gen AI at GXS

0:10:04.3 Satyen Sangani: You're now doing experimentation with Jet AI. Tell us a little bit about that 'cause there's still a lot of work being done there. How much have you done internally? How much have you done externally? How many things have led to success? How many things have led to less promising results?

0:10:18.3 Geraldine Wong: So, I mean, like we all know, traditional AI has always been there. We've been using it since the start of the bank. In the last 12 months or more than 12 months now, the whole GenAI came on board. And I would say in the last year, like many other organizations, we've been experimenting with GenAI prototyping and all that. But I think this year is when we get really serious about splitting them up into things, into projects that we think are worth building ourselves versus projects that we think that we should go procure as a software, 'cause it's not worth the time or the resources to devote on there. And it's not a strategic advantage that we wanna build. And I think what we have done last year is to set up a GenAI Steering Committee where the group CEO is on board. You've got senior stakeholders who can make decisions.

0:11:02.5 Geraldine Wong: So you've got risk compliance, all on those on the Steer Code to be able to steer and guide us and give us that oversight and to give us a guide towards, hey, where should we streamline? Where should we focus our projects on? So I can safely say that for the rest of this year, it's about these three projects that we're gonna work on. And these are very core to the bank's differentiation because it is core to the financial industry and what we are doing internally to create a better workflow and automation.

0:11:30.2 Satyen Sangani: Yeah. Are you able to share what those projects are? And how much are you able to share about what those projects are?

AI use case for banks: chatbots

0:11:34.9 Geraldine Wong: Okay. So the general one that we have, like many others, is about chatbots. But how do we use the chatbot to first help internal customer service agents to do their job better, to retrieve information better so that they can answer the customers quicker, right? And this reduces the number of time and also the interactions that they have with the customers. I think the challenge there is actually about the updates of documents and products 'cause as we are speaking, products are being released and the nuances of each product details is also being updated frequently. So for customer support agents, it's very important for them to get the up-to-date information about the products.

0:12:13.6 Satyen Sangani: Are you using off-the-shelf models and supplementing them with RAG? Or how are you developing these products?

0:12:16.4 Geraldine Wong: Yeah, precisely. That's what we are thinking. So that's the approach that we're taking as well. We are gonna leverage off on some of our hybrid of all these models that's out there. Whichever is performing is performing the best for this particular purpose, that's how we are using some of these models to do the chatbot. The second project is more from a fraud operation standpoint. In the banking industry, there's a lot of resource-intensive tasks that's happening in the back office, like operations. When onboarding, there's a lot of tasks being to check on customer applications and on transaction frauds, for example. So we're trying to leverage AI, both traditional AI as well as generative AI to shorten that process so that my hope is to be able to move the first line of defense to the second line of defense. So more resources being placed on the second line, where it's the checker.

0:13:01.9 Geraldine Wong: And imagine the GenAI being the first line, which is what we call maker. And that would create higher value order work for the second level of people.

0:13:13.4 Satyen Sangani: And is this sort of regulatory paperwork that needs to be generated in terms of fraud or is it?

0:13:16.5 Geraldine Wong: Yes, yes, yes.

0:13:17.3 Satyen Sangani: Got it.

0:13:18.6 Geraldine Wong: So generating of reports so that you check through what, and justify why did this person get flagged out? Yeah.

0:13:26.8 Satyen Sangani: Yeah. And as you develop these use cases, what are the problems that you have found? Like doing this work is pretty high-frequency. I mean, everybody's developing a chatbot, but what are the problems? I mean, obviously fidelity, but what have you done in order to increase fidelity? How do you know when you're using the right model? What are the different ways that you've sort of manipulated the models? And what have you learned through that process?

0:13:50.1 Geraldine Wong: I mean there’s a bunch of libraries that we use to determine some of the relevance on topic that the information is coming out with. There's a bunch of quantitative metrics that we actually take in. But I think the fact is we still need to adopt a qualitative approach and framework to testing this chatbot as well.

0:14:07.1 Geraldine Wong: And your compliance folks, your risk folks, are always gonna come up with questions that will be thrown to the chatbot to be able to test and see if they actually are addressing it well enough to the best of the business. And I think that's the challenge that I'm facing right now, which is how do we have a comprehensive enough testing framework that is able to address some of this needs of different departments and still make it to the customers in the front.

0:14:31.9 Satyen Sangani: Yeah. There's some interesting boundary conditions about, so with RAG and keeping the data up to date, is that a conventional data pipeline that you are now feeding into the GenAI model that sort of supplements and keeps it up to date? I mean, you mentioned this problem of sort of most recent product documentation. How do you do this work?

0:14:48.6 Geraldine Wong: Yeah. So, I mean, we're getting the refresh of the documents in general, but yeah, like you said, I think we need to continuously build the RAG upon the new refresh of the documents and be on top of things with some of this documentation and versioning, which to me is an SOP, is a process that we need to come up with. There's a technology that will help enable some of this. I think the technology is there, but the processes needs to be put in place.

0:15:13.7 Satyen Sangani: Yeah. How often are you changing the prompts and how often are you sort of revisiting and fine tuning the model?

0:15:18.5 Geraldine Wong: Right now it's mostly problem engineering and I think it's through user feedback, crowdsourcing through the users that are telling us whether it's not accurate and all that.

0:15:26.8 Satyen Sangani: And these are customer support agents within GXS.

0:15:30.2 Geraldine Wong: Correct. Yes, correct.

0:15:31.8 Satyen Sangani: So this is a friendly audience that you can rely on and trust.

0:15:32.0 Geraldine Wong: Yes. Yes, that's right. Yeah.

0:15:33.5 Satyen Sangani: Make sense. And it's a super interesting use case.

Supporting financial inclusion with alternative data sources

0:15:35 Satyen Sangani: So one of the interesting things about GXS is that you're essentially broadening the the umbrella of financial inclusion. Certainly, the United States has such similar problems. There are lots of communities that are underserved, that can't get credit because there is no credit history. And the consequence of that is there's this sort of bad, unvirtuous cycle, as it were, that the people that don't bank aren't banked and therefore there's just this inability to get out. But here, people are obviously using Grab every day. Most people will have a mobile phone in a country as sort of wealthy as Singapore. Tell us about sort of what you found in broadening inclusion. Is that something that hasn't gone as fast as you would expect?

0:16:18.3 Geraldine Wong: So I think we do recognize that there is a segment of consumers that today do not have credit history or have very small credit history in Singapore. Actually, in Singapore, I think more than 90% would have a credit history in the credit bureau of Singapore. So the percentage that we're talking about is quite small. And with this group of segments, I think traditional banks will not be able to give them a loan because of the lack of this information. And so what we have done is actually to create a proprietary credit assessment tool, a model to be able to leverage off alternative data from Grab and Singtel.

0:16:50.2 Geraldine Wong: So if you think about your Singtel bill repayment behavior, your buy now pay later, or your financial wallet behavior from Grab, these are some of the proxies that we use to be able to credit risk score them. The other thing that we do is about risk-based pricing. So in our unsecured loan, FlexiLoan, the more data that you provide us from Grab and Singtel, the more preferential your rates will be. How the rates are differentiated is through the information that we get from you as a customer of both Grab and Singtel.

0:17:17.8 Geraldine Wong: So I think that's one part. I think on the second part is more from an acquisition standpoint. So when we are able to see that you have a certain preference on the Grab or Singtel platform of a digital product, for example, this is when we think that you have a higher propensity to spend or higher propensity to take up a certain type of consumer product from the bank. And that's when we do a more targeted and personalized targeting for you.

0:17:39.4 Satyen Sangani: So let's talk about the goal of GXS for a little bit. So one of the interesting things that I find as somebody who's studied development economics, this idea of financial inclusion, where a lot of what you're doing is giving the people who have otherwise or previously been lightly banked or completely unbanked the opportunity to participate in the financial system. Tell us a little about that. What are some of the products that you offer? How do you do that? Why are you able to do that in a differentiated way?

0:18:04.1 Geraldine Wong: Yeah. So we have two main products. One is the deposits account and the other is the unsecured loans. When I speak about the deposit account, it's a daily interest credit in account. And what this does is you can put a minimal sum of a $100 you let it roll over. So it's really good for students or people with very low savings to be able to accumulate wealth even as they go, but still having the freedom to take out the cash whenever they wanna use it for emergency needs. Our unsecured loans is basically what we term a FlexiLoan product. And this is putting the control into the hands of the consumer, giving them the ability to choose the tenure of their loan, which could range up from 2 to 16 months, taking a loan from as low as $200 for emergency need.

0:18:45.0 Geraldine Wong: Say for example, your kid needs a laptop, you could get a loan and pay it off in the next two months. And this is a quick and easy way of getting unsecured loans. Now, how do we use data then to help with these two products? In the acquisition, the customer acquisition, we actually leverage off some of the early adopters of the digital financial products within Grab and Singtel, looked at their behavior and say, hey, who are these people who would be the first ones to adopt a GXS account who would greatly benefit from using some of this deposits account? I think there's more impact from a FlexiLoan perspective where most of Singaporeans are actually, over 90%, I would say it's on the credit bureau of Singapore.

0:19:22.3 Geraldine Wong: But like you said earlier, there's a lot of people who might be no bureau or same bureau records. And today, these traditional banks do not serve them very well. What we're trying to do is to then say, hey, if you bring us alternative data sources, such as your Grab and Singtel data, we are able to risk based price you to give you the interest rates that are preferable for you. We are also able to ascertain or determine your credit risk profile. So we could offer them a credit limit of a very small amount of this non-bureau customers and through their repayment behavior on our platform, be able to then increase the limits to saying we learn your behavior enough.

0:19:58.0 Geraldine Wong: Now we are assured that you are a good paymaster, you repay us very well. We can increase that limit for you to further borrow as well. So I think this is building that whole trust with the customers that's important for us.

0:20:09.6 Satyen Sangani: And do every one of these micro-loans, I mean, 'cause they're effectively quite small, do every one of these micro-loans get reported back to the authority?

0:20:15.1 Geraldine Wong: The credit bureau.

0:20:16.6 Satyen Sangani: They do.

0:20:17.5 Geraldine Wong: Yes. So it's an obligation that we have to transmit or transfer all of these data points on a repayment behavior on a monthly basis as well.

0:20:24.3 Satyen Sangani: So you start at a loan of $200. And what are the interest rates on these loans?

:20:29.5 Geraldine Wong: 2.98% is the lowest.

0:20:35.0 Satyen Sangani: That's incredible. I mean, that's incredible pricing. So if you think about that, I mean, in the United States for a similar sort of unsecured lending that you might be able to do with a credit card, you're talking about a minimum of 12% but probably close to 15% or 17% or 18%. And so then you take this $200 loan out, you obviously make a couple of payments. You give the money back. What is the growth that you've seen? I mean, what's the opportunity for somebody who otherwise was unbanked if they have the right income, can they loan thousands, tens of thousands?

0:21:03.3 Geraldine Wong: This is still dependent on their income as well. So the regulators have also a certain guidelines on what is the multiplier effect on the loan limit that they can take based on their income levels on the annual. So if you think about the gig workers, they don't really have a constant income coming in. Their annual income might be very low, and they don't hit those guidelines that the regulators have given. And this is where we, see the right to play, being able to give them such an unsecured loan.

Leveraging Gen AI at a digital bank

0:21:33.2 Satyen Sangani: Fascinating. So you've done this interesting work on Generative AI and you now have a couple of models that, that you are deploying. As you've done this work and you think about sort of future applications, what trends do you see and where do you see sort of both the use of the technology going and where the impact of these models will have on the banking industry?

0:21:52.3 Geraldine Wong: Yeah, I think just like earlier at the, conference, I think a lot of people were saying about how GenAI has actually made it easier for many people in the organization to adopt some of this AI into their day-to-day needs. And I see that there's gonna be an acceleration and an amplification of the AI in many aspects of banking. How I see it is about looking at it in three forms, task level, workforce impact, as well as enterprise level. And today we see pockets of task level impact. So meaning within a business unit of say finance or procurement in their day-to-day tasks, the adoption of AI is there. But I think what we haven't seen is this whole... Well, workforce impact whereby it requires a cross-functional team, a workflow if you may. If you think about a procurement for example, how do you automate that whole procurement with just AI, with just GenAI.

0:22:46.2 Geraldine Wong: And that impact there is far greater than the individual task level ones task level, meaning information retrieval, summarization, creating a deck. I mean, those are not, we call it, I mean, those are not seriously the big fish there. The enterprise level impact that I like to see is also about how do we create new business revenues from GenAI today. We haven't seen much of... Well at least I haven't seen much of how GenAI has created new products, new banking products. If you think about AI being part of a product manager, product creation. So you go to different segments of your consumers, see what their pain points are, do the summarization and then say, Hey AI, can you create a new product that would match the needs of 50% of my segments in the consumer business? And the AI quickly generates some AI product, a banking product for you with such features. And you iterate and iterate and iterate. This is some of the tools that a product manager could really leverage on to create a new product from scratch.

0:23:39.4 Satyen Sangani: Yeah. Because all of those call logs and all of the behavioral data is available. And the model may spot, even if there's 400 garbage suggestions, there might be just one good one, which itself could land a financially viable product.

0:23:53.4 Geraldine Wong: Yeah. But what I'm also saying is also like you need to know your customer segment. So for example, if you had 10,000 banking customers today, what have they said about... What are the profiles look like? And you know, their profiles, what have they taken from you? What kind of financial products have they taken from you? Have they been sticky with you? If they are not so sticky, there must be a gap that you are not solving, the problem statements are still there that exist today that you're not solving.

0:24:16.3 Geraldine Wong: And there's both a quantitative and a qualitative approach. I think what you mentioned earlier was a qualitative approach where you are getting feedback from consumers, both from maybe from an ad review, from a focus groups, et cetera. But there's also behavior from all of the apps that the consumer is actually interacting with. And those behavior could also fit into this AI, I call it a generated product, that you create a product as well and keeps iterating. And of course it relies also on the product managers to be able to filter out what are the key features that they wanna see. Also speak to the business product people to see which are viable and financially commercially viable products that would make sense to the company as well.

0:24:52.4 Satyen Sangani: Well it would also require these AI models to understand structured data in these segments, which right now is sort of an iffy thing. I mean, mostly it's trained on unstructured data. Let's rewind a little bit. So how did you get to this point of GXS? I mean, you obviously started in the academy, you were trained as a statistician.

Geraldine’s career journey

0:25:08.8 Geraldine Wong: I'm trained as a statistician in climate change. So predicting droughts in Australia, very different to... I wanted to save the world with environmental research, really. And moved over to Europe to do the same thing on the extreme side of things, extreme weather events.

0:25:23.4 Satyen Sangani: Where these models are, by the way, having massively strong predictive impact.

0:25:27.8 Geraldine Wong: Yes. Global climate models, regional climate models, they all rely on huge amount of processing. And I mean 10 years ago, and now it's a different story with all the processing power and compute resource as well. And then 10 years ago, that's when big data became a thing, at least in Singapore and Telco's, banks were just hopping onto the bandwagon. I think the issues that they had then was having a lot of data, but in silos.

0:25:50.4 Geraldine Wong: And the first organization that I joined was called DataSpark, and they were, a startup company spun out of Singtel, where Singtel had a ton of location data from telcos, from the signals, the network signals, so how we use our mobile phones. And that was a monetization product that we created geolocation. So something like, I think in Telefonica in Spain, that's what they created as well. Similar to that. And then moving on to doing some consulting work in, again, in the data science area.

0:26:18.3 Satyen Sangani: And that data would be sold to like, that's foot traffic data that would allow it to be sold to retailers primarily. Is that who would be buying it?

0:26:24.6 Geraldine Wong: So we don't sell the data, we sell the insights and the software piece.

0:26:26.4 Satyen Sangani: Ah, okay.

0:26:27.4 Geraldine Wong: And the consulting piece that goes with it.

0:26:29.8 Satyen Sangani: Okay, excellent.

0:26:30.5 Geraldine Wong: Yeah. So because the data is proprietary to the telco company, you've gotten consent from the consumers.

0:26:35.6 Satyen Sangani: Right. So you can sell an aggregate, but you can't sell the specific information.

0:26:39.2 Geraldine Wong: That's right. You're right. And real estate planning, shopping malls even crowd control as well for major events. Or even where you should place investments on your real estate, more traffic like your shop locations. And I think there was also another use case on network optimization as well from a Telco's perspective. Imagine if you had to do a migration from 3G to 4G, 4G to 5G, where do you wanna place some of the cell towers? And then I moved into consulting because I wanted to broaden my scope of using data for different verticals. So beyond telco, where else could I use data science and data? And I think at that point, at that stage, many organizations were going through that journey of discovering what data they had. Some maybe haven't been collected, their data or even in the worst form quality wasn't there, the data quality wasn't there. But what we did was to help build prototypes in government in the transportation, e-commerce for example.

0:27:32.3 Geraldine Wong: And then I moved back into the client side, which is the consumer side to Singtel consumer. And that was purely by chance because I wanted the ownership to again, build something from scratch. And at that time I think it was very exciting because we were able to put most of the data into a central data lake and leverage off open source tools. So you have Spark Computing, Python, et cetera. Some organizations was still on that stage and be able to also work with insurance companies as well to help leverage some of this data for their usage.

How do you grow from data scientist to Chief Data Officer?

0:27:58.8 Satyen Sangani: So you have an interesting background in that you've gone from being a sort of practicing data scientist to being a chief data officer, which require sort of different sets of skills because one is fundamentally quantitative and analytical and the other one is more, I mean probably stakeholder management and policy management and people management are, I would imagine much more of your job. How did you make that transition? I mean, there's probably a lot of people listening to this podcast who would like to do similar, to maybe think about or consider similar transitions in their careers.

0:28:27.4 Geraldine Wong: I've gained many lessons and experiences from many of these roles that I've gone through. From the consulting role, I actually gained the experience of presenting and relating some of this value of data to senior stakeholders. You are put in front of senior stakeholders to be able to sell the value and the strategy. And that is a skill that the clarity of thought, the thought process that you wanna take them through. It's very important and it's a very structured manner. And I can tell you like when you are put through the fire you will know how to get there. Because even through interviews, like even when I interview someone now, sometimes I feel that it's lacking when they go through the interview process with me. That, that train of thought, that thought process as to how do you articulate to a senior stakeholder.

0:29:08.3 Geraldine Wong: I think getting to GXS the first, I would say four to six months, it was very much an individual contributor role. 'Cause I had no team back then. A lot of the discussions were with engineering team on creating sandboxes, on creating pipelines all the way up to this data sharing agreement that I was telling you about, which is talking to legal, talking to your compliance folks, et cetera. So I would say the transition really happened also within this four years of when I started all the way up to when I'm, actually right now. Because even in that first six months, there were days that I would speak to engineers in setting up sandbox and the next day I would go and talk to the board about what we are doing in the space of data and AI. There's a lot about understanding the board and what they also are interested in. And I think it's really comes through practice. I would say you have to do it before you get, you learn from the mistakes as well.

0:30:02.3 Satyen Sangani: But you also have to try to learn these skills.

0:30:04.3 Geraldine Wong: That's right.

0:30:04.8 Satyen Sangani: They're not gonna just come to you.

0:30:05.9 Geraldine Wong: That's right. And I think I came in with a lot of assumptions that people are gonna adopt data as it is in my previous organization. They're gonna understand what data means. But this is far from true. I was one of the first few members who started in this organization, as new people came in, I realized that I had to constantly be advocating and evangelizing because the level for which they came in was from their previous organization. And GXS is a very new organization, it was brand new. If you think about it and people come in with their preconceived assumptions of what data means for them based on their previous experiences. And it was like starting from scratch again and saying, Hey, hey, no, no, this is how I can help you. This is how we do things. This is what the tools are there for you to enable yourself to help. So there was a lot of selling, a lot of evangelizing.

0:30:53.8 Satyen Sangani: Which is shocking even in the context of a digital bank where one would think that it's essentially so much more tech forward than a traditional bank.

0:31:00.6 Geraldine Wong: It is very tech forward. I think the fact that I was one of the first few staff that was being hired actually shows a lot. So I think the first three people that were hired, the CPO, the CTO and the CDO, so it tells you how much a focus it was placed in the bank itself, but you can't assume the same for the rest of the organization as well. And so when new people came in, it was educating them again, what they think about data, how can you help them, how can you partner with them? And I think this is, it was a very important but necessary step. And I am glad that we took that step because now we have no lack of use cases that people come knocking on our doors.

Playing data offense vs defense at GXS

0:31:34.0 Satyen Sangani: So the original CDO role in financial services was primarily a defensive role. A compliance oriented role, and just making sure that you had data transparency and adequacy. How much of your time today is spent on sort of what I'll call defensive compliance oriented risk oriented initiatives versus the analytical work that obviously is fundamental to your original background?

0:31:57.0 Geraldine Wong: So I would say it's a balancing act that I have between the data privacy, data governance as well as the data science and GenAI. And one of the first things that we actually did as a bank to build the trust and to foster that credibility with our customers, was in Singapore we have a data protection trust mark. So this is a certification given by the government or IMDA to ensure that all your data, all your robust practices on data governance, data privacy was there. So we wanna get it certified so that people would trust us with their data as well. And on the offensive side, I would say that, like I said earlier, I had to advocate to many of our users and our stakeholders on, on how data can be used.

0:32:37.1 Geraldine Wong: And again, when we use data from data sharing partners, it needs to come with a right amount of consent. So you see there's a balancing act here. I go into the discussion, going to advocate saying, Hey, you know, we can leverage off our data shareholders on their data, but when you put in your request, you need to have clear purposes, clear consent, and who can use it as well. So there's this balancing of education on using it, but also guard rails around it as well.

0:33:06.1 Satyen Sangani: Yeah. So there's the motivation, but then the business opportunity comes with sort of, if you will, the rules to surround it. There are some data scientist, some data professionals who might wanna do what you've done, which is sort of grow to be CDO. What advice would you give to those individuals?

0:33:21.1 Geraldine Wong: I think read broadly, read very broadly about all the data related articles and issues that comes with it. Depending on the role of the chief data officer, I think you need to be able to understand the nuances, the challenges that comes with it. I think it's also a self-reflection on whether you are able to navigate those challenges within the organisation that you are at as well. Never be afraid to take on more scope. We spoke about GenAI. And this is additional scope that we talked about as well.

0:33:50.8 Geraldine Wong: So I would say also look for a mentor. The mentor gives you a sounding board with how to navigate some of these challenges. How do you broaden your scope beyond what you're doing right now? Have some of the key projects that you're able to highlight value to the senior management to be able to show that you are able to handle much bigger projects as well, even cross-functional projects.

0:34:12.6 Geraldine Wong: I think the last thing is network. I think networking is so important in this organization, in this area. I've been lucky that, like I said, 10 years ago, when we started out, there were not many people in this area. The group of us that graduated all still network with each other. So there's a lot of exchanges with what's the new trends, not in the banking industry, but just across in Singapore, whether healthcare, you've got transport, you've got banking. What are the trends that's happening? How are they using some of this GenAI? It's really good to hear bounce off ideas. It broadens perspective as well.

0:34:45.7 Satyen Sangani: And what are the biggest challenges you faced as a CDO? What keeps you up at night? What do you think about that are the most critical things that either you didn't expect or you do deal with?

0:34:55.7 Geraldine Wong: I'll have to go back to the unique challenges that we face. I think it's the partnerships, the data sharing partnerships. While it brings its own sets of advantages, we're also talking about different DNAs from both partners, Grab and Singtel. And being able to convince them, being able to influence them. Like I said how do we... The stake that they both have is very different as well.

0:35:15.8 Geraldine Wong: So how do they share your data with us? How do you keep it in a very secure manner? I think it's one of the things that keeps me up a lot. I think the other one, which probably you can relate to is the talent and skills gap. As we go into GenAI, how do we grow the people internally versus how do we augment the resources that we have as well? What are those resources look like? Today, I don't think there is a structured framework for training someone to be a GenAI scientist. There's many tools.

0:35:43.2 Satyen Sangani: Five years of experience doesn't exist. Yeah.

0:35:45.2 Geraldine Wong: Yes. I mean, there's so many Coursera courses out there. There's deep learning AI courses, but which sequence do you take it? If this person has done 10 courses versus five courses, is that enough? So I think those are some of the things that I would say are top of mind.

How is data governance like a seat belt?

0:36:03.9 Satyen Sangani: Yeah. Yeah. So you've said data governance is like a seat belt. Can you unpack that quote a little bit?

0:36:09.9 Geraldine Wong: So I find that like a seat belt, when you don't get into a car accident, it's completely unnecessary. But when you do get into a car accident, then you say, oh, why didn't I wear a seat belt and keep myself safe. I mean, if I bring the parallels, if I draw parallels to that in the banking industry, it's almost saying that, hey, why don't I have good classification of data and good definition of data? Why are there two columns that have similar fields, but completely different entries in there. And this is because it stems from not having proper definition of proper data governance.

0:36:41.5 Geraldine Wong: It's very basic stuff, but people get it wrong. And when you do credit risk profiling or fraud monitoring, these are things that has intrinsic impact, big impact on your customers. And you don't wanna get it wrong from the basic start. So to me, that's how I see data governance should be. It should be embodied within the organization. People should be very aware of it. So like in GXS, we actually have a yearly training on data governance modules. So you're meant to take it out. It's a mandatory training.

0:37:09.0 Satyen Sangani: Oh wow.

0:37:10.1 Geraldine Wong: Yes. It's a mandatory.

0:37:10.9 Satyen Sangani: For every employee?

0:37:13.3 Geraldine Wong: Every employee, even the contract workers as well. They are meant to take it trust me. They come to me all the time and say, oh, I...

0:37:20.4 Satyen Sangani: This is my favorite course. I love it.

0:37:22.3 Geraldine Wong: So we have a data privacy, a data governance mandatory training that you have to do it every year.

0:37:26.9 Satyen Sangani: Yeah. And you've built these internally.

0:37:28.7 Geraldine Wong: Yes. The content is built internally. It also shows them how... We use Alation. How do you find data from Alation, for example? So these are some of the refresher courses that we ensure that people are up to date with.

How should a CDO manage their stakeholders?

0:37:42.4 Satyen Sangani: And as you think about sort of growing your role, growing your remit, success, how do you manage your business stakeholders? I mean, one of the about the CDO role is it's fundamentally, there is obviously an opportunity to it where you can discover new products, discover new capabilities, but there's also an enablement aspect where you're constantly sort of, you're having to manage your stakeholders. How do you do that? And what do you find, what are your best practices and techniques in being successful in navigating your stakeholders?

0:38:10.1 Geraldine Wong: There's a couple of ways that we partner our stakeholders. Firstly, I think we wanted to democratize data and put it in the hands of our stakeholders, meaning they do a lot of self-serve. We are not a very big organization. So enabling them with some of this technology to self-serve, visualize on dashboards, for example, is the first thing.

0:38:26.3 Geraldine Wong: But having that trust in the data on the dashboard is also another thing that we need to ensure. And we need to ensure that the data is being classified properly for them to use is one of the key steps. Once we get past that hurdle of them self-serve, then it's about how do work with them to solve the big rocks, the big problem statements.

0:38:46.1 Geraldine Wong: So it could be how do I increase engagement? How do I gain traction in this product? Who do I acquire? And this is a collaboration that we have with them. We often do champion challenger modeling, so they can create their own rule-based model. We can create our own AI model that then challenges them and we see who, well, if you think about it, it's like an A-B test kind of... Yeah. So we see who wins on the success matrix, and see who acquire more customers with the lowest amount of costs.

0:39:12.6 Satyen Sangani: Does the AI always win?

0:39:14.9 Geraldine Wong: Not necessarily. I'll give you an example in our first run of the FlexiLoan, there was also a rule-based model, but we also created one that was an AI model. And what we did was, first of all, it used to be 95%-5%, for example, 95% of the base was gonna be run by the rule-based versus 5% on the AI. But as we gained more confidence on the AI model, we started to shift progressively and more base to the AI model. So that's how we've been adopting some of these partnership models with our stakeholders share.

What’s the future of the CDO role?

0:39:43.3 Satyen Sangani: I guess maybe to end, you have such an interesting and progressive background. You're doing it in a controlled environment, but with a lot of opportunity. As you think about sort of building the CDO role, growing the CDO role, what advice do you have for CDOs? How do you think about the role going forward? And what do you foresee as this role evolves with, first it was big data, then it was sort of data in the cloud, now it's GenAI. I mean, how do you see this role changing and evolving?

0:40:10.6 Geraldine Wong: I think it's clear that GenAI would be part of the remit. Not every organization would probably agree to that, depending on mandate and all that. But I think progressively, there's gonna be also the digital space. I mean, we've seen a lot of chief digital officers. I'm of the view that there might be a convergence of such roles, like chief digital officers and chief data officers, because there's a lot of AI that's being used in workflows, in digital transformations as well today that we are seeing. There might be a convergence of both these roles into one or the other.

0:40:43.3 Satyen Sangani: Excellent. Well, Geraldine thank you for making it on Data Radicals, we appreciate your time and this was a phenomenal conversation.

0:40:47.5 Geraldine Wong: Thank you.

0:40:51.0 Producer 1: That was such a valuable conversation. Our key takeaway? Data governance is a seatbelt. It's something you overlook until you desperately need it. But when it's there, it can save you from a lot of pain. For GXS, data governance isn't just about compliance. It's about building trust. It's about powering reliable and safe AI for enterprises. It's about unlocking AI's potential in banking, from streamlining back office processes to improving customer interactions with AI. The opportunities are huge. If you're thinking about how to future proof your business with AI, start with data governance. Thanks for listening, Data Radicals. Keep learning and sharing. Until next time.

0:41:29.0 Producer 2: This Podcast is brought to you by Alation. Your boss may be AI ready, but is your data? Learn how to prepare your data for a range of AI use cases. This white paper will show you how to build an AI success strategy and avoid common pitfalls. Visit alation.com/ai-ready. That's alation.com/ai-ready.