Effective Data Governance For Compliance Risk in Financial Services - with Supreet Kaur of Morgan Stanley AI in Financial Services ยท 13 Mar 2023 Source: https://aiandbanking.libsyn.com/effective-data-governance-for-compliance-risk-in-financial-services-with-supreet-kaur-of-morgan-stanley Transcript: https://podcasts.opennash.com/shows/ai-in-financial-services/effective-data-governance-for-compliance-risk-in-financial-services-with-supreet-5c2e93/ (via OpenNash Podcasts, machine-generated) Welcome, everyone, to the AI and Financial Services podcast. I'm Matthew DeMello, Senior Editor here at Emerj. Today's guest is Supreet Kaur, Assistant Vice President at Morgan Stanley. Supreet has spent the last five years working in data spaces across financial services and life sciences, along with essential advocacy work in mentoring a diverse next generation of data science leaders. In conversation with Emerj CEO Daniel Faggella, Supreet explains to the audience what works and what doesn't in building effective data governance infrastructure and strategies for financial services organizations. Throughout, she emphasizes the need for a top-down approach in ensuring employees and analysts aren't overwhelmed with data they don't need and probably shouldn't be seeing. Without further ado, here's our conversation. So, Supreet, welcome to the program. Hello. Thank you for having me. Yes, glad to have you here and glad we're able to catch up on the weekend to be able to get this recording in and make something fun happen. We've got a lot of good topics to explore, and we're talking about the value of data and AI and the regulatory function in financial services. I wanted to talk with you first about data. You have a very strong technical background. When you think about the increased importance of data in the regulatory function, kind of how do you see that? When you look up close and personal as to how data is serving more and more of a role and why its importance is becoming more prominent, how would you frame that and explain it for kind of a business crowd? Yeah, of course. So obviously, this is the era of the data. It's a data-driven world, and most of the organizations are heading to promote data-driven cultures. So it plays an imperative role in ensuring the success of any use case in hand and compliance is no different. So obviously, we will cover multiple use cases in this as part of this conversation. But just on the data front, I feel that more than what data you have or importance of data, the tougher question is how you're using the important data to capture the required information and who has access to the data. In such a regulatory environment, it is sometimes disturbing to see a few statistics. And I have a few of them handy from Harvard's review, which says that more than 70 % of employees have access to data, they should not. And 80 % of the analysts spend time in discovering and preparing data. So those are alarming statistics that we definitely need to kind of look back and analyze what is our data strategy. Yeah. So clearly, and these big financial institutions founded 200 years ago, were not exactly founded with these considerations in their DNA, right? These were not things that necessarily were a concern then, but now we really do have to adapt and evolve. You brought up two points. I just want to make sure I'm highlighting the ones that are most important and we can talk about what you are seeing in terms of companies addressing these. One of them is the fact that governance and access isn't structured and clear enough to really make for a safe environment here. We've got a lot of data open to folks that maybe shouldn't really have it. And maybe even from a compliance standpoint, it is a risk for them to have it. That's one. Number two, there's all this preparation and access time going across silos, manually exporting things in really sloppy, time-consuming ways. and we're unable to really be able to act on things. So are those the two you wanted to kind of put a pin in here for our conversation? Yeah, definitely. And I think then we can kind of talk more about the strategy and how data can be leveraged to address these use cases, I would say. Yeah, yeah. So we will get into use cases, but I like that you're talking about the issues just with data. So it's becoming more prevalent. We're able to train algorithms to inform decisions, to automate or streamline processes, but we've got governance problems, we've got access problems. It's a little bit of a messy environment in the big stodgy financial services world. And they're all evolving and trying to catch up here. What are you seeing work well for companies to get that strategy in order? Because it would seem clear that any AI use case would be better, easier to act on when we know our data is safe and isn't too open to people. And we also know our data is accessible, simple to use, simple to pull down, simple to leverage, but we're not there yet. How are we getting there? What are you seeing actually work or not work to solve those data problems? Yeah. So it's speaking to a lot of executives in the conferences and day-to-day. I think it's a mix of a lot of things, right? There's no one solution fits all here. But there are a couple of points that I would like to highlight, right? First, as we mentioned about the access, right? data access is a major issue. Now, obviously, with the increasing cybersecurity, companies are focusing a lot on developing proper frameworks, proper documentation as to what data exactly one executive needs, right? One of the analysts doesn't need all the data sources that you have in your data repository. And definitely financial institutions have rich data. They have a lot of consumer data, a lot of PII, which needs to be handled with care, right? We all talk about building responsible AI models and everything, and data is kind of, you know, the start of it. So yeah, so building those frameworks, having a clear strategy in mind as to who needs what data, and that's an evolving thing, right? You cannot build a strategy for one use case and forget about it. You need to go back and forth, think about what you're trying to solve, and then what data access does a particular role require, and you revisit it every three or six months. Yeah. Who needs to be in the room to think about this? Clearly, there is a little bit of both top down and bottom up action. You know, here, there's, we've got to get enough buy in at the top to understand the business goals, what are these departments responsible for based on those responsibilities that we need access to certain things. But then we also need the subject matter expert perspective. What's frustrating for them? What is very hard or impossible for them to access to actually do their job? We need their perspective too. How have you seen collaborative conversations form to learn, hey guys, what could be our governance and access strategy for data in this department? Like what has been productive in terms of the kind of people in the room and the way those conversations are fostered to come down to a strategy that we can actually use? Yeah. Yeah. And as I said, right, if you're thinking that you'll be able to build a static strategy, that's not true. I mean, that's not true for anything. So just to at least have those conversations and get those conversations started, you need SMEs, but you need data SMEs as well as use case SMEs. You need the business stakeholders and you should know what they're trying to achieve. And if you already have a data-driven culture in your team, some of these things will flow naturally. Like everyone's so comfortable with the data. Everyone knows what's there. They can talk in data and they know what they want to achieve. And then comes like a discussion as to, okay, what data sets do we need? And do we even need sensitive information? Sometimes you don't need that as one of the important features of your maybe AI model, right? So then you can mask it for that particular use case. And you need your senior leadership and the compliance team, I would say. We should definitely not miss on those folks because they can help you be on track to what needs to be achieved and what is, I would say, what's a gray area, what's not accepted, and what's definitely accepted. So all of those legal conversations also need to happen. Yeah. And so multiple voices in the room for our listeners, this will not be an uncommon thing. The idea that not one person understands all the moving parts of AI in the enterprise. AI is a team sport, as some of our guests have mentioned. When this is done well, is there somebody in leadership who kind of mandates, okay, look, you different spindly parts of the business, compliance people, subject matter experts, IT folks, whatever. I know you're all busy doing your own thing because they all are, right? Everybody's busy doing their own thing. But we are going to have this many of this conversation and we're going to have this outcome. Like, does it need to, on some level, come from the top to say, we need a better governance policy by this time horizon? Is that often the impetus that gets this done? Or have you seen that bubble up in different ways? Yeah, I think, again, it kind of depends on where you are in terms of your governance strategy. Are you just starting out? If you're just starting out, you probably might need the help of senior leadership and some more handholding. to get to a point where you can actually have a team that can run on its own and define governance for you. And it can make sure that it's having the right conversations with the right people to make things happen. So it will depend on how and where you're starting from, but definitely reach there. That's interesting though. I actually, I don't think we've ever had this distinction brought up on the show. And so I think it's cool that you're touching on this, that maybe in the earlier, earlier days, it's maybe more and more important that that impetus kind of come from the top, because if we have no momentum and there's no current collaboration process, cooperation around governance in general, the spark has to come in from, so it sounds like it's more important in that early stage to have the top-down direction than later when the balls are already moving and we're trying to figure things out. Yes, exactly. I mean, if I would say if the ship has sailed already and, you know, you're already in trouble. There's nothing that your senior leadership can do to save you. They will try their best, obviously, but why to even reach that situation? So it's better to have those, I would say those governance gates, those processes, those frameworks already in place so that you don't have to reach that stage. Got it, got it. So, okay. So it's good that we're starting here around compliance use cases. Of course, there's so many use cases we can dive into, and we will in a bit. But let us not forget, already we might have access problems that are a compliance risk, financial risk, a legal risk right now. And also, we have other access problems that are just keeping people from doing what they need to do. It's information they should be able to get, but they can't. It's kind of funny. It's almost like there's stuff that people shouldn't have access to that they do, and then there's stuff that they darn well should have easy access to that either they don't have access or it's not easy. So both of those, it sounds like what you're advocating here is let's think about those two parts of the ecosystem and let's think about collaborating to solve for a bit of that. Otherwise doing real AI use cases is potentially very counterproductive. So I like that place of starting here. I guess that takes us a little bit into some of the areas where data is starting to wake up in regulatory and compliance and financial services. In this series, we've spoken about everything from insider trading to other kinds of financial prime and, you know, AML and you name it, there's just so much. When you think about use cases where you're starting to see AI inform decisions or streamline processes, what are some of those that are good examples for business people to understand? Yeah. As I also mentioned that financial institutions have rich customer data. And And the first thing is just how to harness the potential of that data source, right? Like other data sources kind of get married to your consumer data. And consumer data can tell you a lot about the customer journey, as it says, know your customer, right? If you want to have a deep dive into the financials, you know, how the consumer has been engaging with different firms, how has it, I would say, the credit card history, all of those parameters can be tied back to the user and consumer data. So just even pondering upon how can we develop efficient data pipelines and data quality frameworks to handle this data so that it can be easily ingested for our AI use cases and multiple AI use cases, right? So I think just spending time on that is crucial at this point. Yeah. Okay. And what does that look like for you in practice? In other words, is there a specific strategy conversation that tends to unlock that and put it on the table in a productive way or what has been your experience there? Yeah. So I feel, so it's funny, right? It's again, a chicken and egg problem. I would say we have too much data, right? But everything is not accurate. Now, obviously you cannot kind of go manually and check how the data looks like. But there is a human touch that's required to even ensure that the data is accurate, right? So you can have some automation pipelines ready, some basic automation there, just to check a few things, right? And those few things will obviously, again, depend on the use cases, the type of data you receive, but you can set that up in place. And also checking basic statistics around your data. What does my population look like? What does my sample size look like? Is it highly skewed? Do I have a lot of missing values? And is my data bias? Because if any of those parameters are coming into play, your models will be impacted, and tremendously impacted, I would say. Yeah, yeah. So almost you're advocating on some level for, I'm just taking this by what it seems like through earshot, some kind of almost a technical audit of the state of our data in these different departments, areas, what have you. This kind of seems like the technical handiwork of, okay, we're thinking about leveraging this data for this use. Let's take a look at how complete it is, what kinds of information we have on people, what the compliance considerations are for that. Is there inherent bias in the data as you're articulating it now? I will just limit it to technical. And that's why I said that the cure is where the role of a business SME comes into play, right? I feel again, it's a collaborative process, because technical analysts or data scientists, or whoever is working on the data, they can tell you things that are happening, you know, just by numbers, by statistics, but there has to be a human who kind of comes into play and questions all of those statistics, those numbers and looks at it. And with its business expertise gives you those meaningful insights that will help you to see how accurate your data is. Yeah, yeah, yeah. Clearly, again, it's a multiple team member thing, right? So there is a technical element here, but there's a context element. So, you know, I, as a pure data scientist jumping into some call center data or some transaction data, I may be able to say, okay, these things are statistically anomalous. But as an SME, I may be able to say, well, it's the Christmas season. There's nothing actually weird about that. So like, I don't even care to look at it because it doesn't matter to me. Like I know the business and that doesn't matter. But if we combine our heads, then maybe we'll be able to figure out those two things. So obviously you're bringing this point up. Now I would estimate that this often doesn't happen very well. And it's certainly challenging. It's new for all legacy enterprise. It's this process of looking at an area where we could apply AI and then assessing the possibility landscape, you know, through an SME lens, a technical lens, what have you. What is it? I'm wondering almost like what this looks like when it's done well, because it feels like you'd have to be a very mature enterprise to say, OK, we're thinking about leveraging AI to decrease false positives for this kind of AML activity. For somebody involved in steering that project to say, hold up, let's first pulse the state of the data in terms of access, quality, and bias in those different departments that we might leverage and just understand what we're working with. That's a very mature perspective. I think it's very admirable. It also might be rare today. Are you seeing that conversation happen more frequently? Are people like you who understand the process a little bit more often the ones advocating up to say, hey, let's do this first, guys? You know, how is that conversation happening? Because it's obviously a very new one. Yeah. I mean, in my experience where, you know, I've worked with a startup of 60 people and then 6 ,000 and now 60 ,000, I've worked with folks with all different kinds of mindset. it, right? But I think one thing that has been pretty clear in everyone's mind and one thing that I've been advocating or we've been advocating as a team is to focus on data. And it's not a joke when they say that data scientists spend 80 % of their time. In reality, when I started my career as a data scientist, that's exactly what I was doing. So I would say it fall on the onus of not not only data science practitioners, but also leaders to give them that space and that empowerment to question some of these things and develop that culture. And that is where it, you know, all of your, that is where you form like that collaborative team where you have a data governance, you have data scientists, you have your data experts, right? And you're achieving things. So, yes. Yeah. I want to crystallize this as maybe a take-home lesson because there's a lot of our listeners who are aware of these change management issues. They're aware that AI is different than IT, which is a drum that we beat a lot over here at Emerj. They're aware of that, and they would hope to be part of the catalyzing force to encourage more collaboration, to get more value out of these technologies. They want to see that shift over time, and it's hard. it seems like if there is a take-home lesson here from what you're saying, part of it seems like, okay, well, you need the right leader that understands the value of that unpacking, collaborating, investigating. You need that right leader person. Are there other key lessons? If you're talking to an audience, maybe VPs, directors, what have you, they know they're about to sink their teeth into some really tough problems here with data and AI. Maybe it's even in compliance, like the area you and I are talking about, they're prepared for the challenges of change management. You know, they've listened to the show. They know it's not easy. What are some things that they could do as part of these teams to help to foster that shifted culture where we have our SMEs, our data scientists, our, you know, kind of management folks around data science working together in this productive way? Is there any kind of parting advice that you would leave them? Yeah, I feel like this shift is very, you know, is gradual. It's not sudden. And it takes hiring the right folks, having the right leaders, and having a common goal at the end of the day, right? One of the important things that most of the organizations are trying to do is, oh, they want to harness the potential of data, but then you need to set up the right team. And if you look at the collaboration that we spoke about, all the teams' skill set are mutually exclusive. There might be some of the skill set that's definitely overlapping, but every team and every person is bringing something fresh to the table. And that's what we are talking about. Right. So having that diverse perspective is a huge shift, not only to the culture, but the problems that you're trying to solve and the solutions that come up. Right. I was actually reading about, I think, an article a few months back where it spoke about how companies are building these innovation labs right in their team. So there are a group of folks who just think about, OK, what are the what is the next use case we can do, considering all the data that they have? So they know all about the data, but they're just thinking about the use case. They're not the one who are implementing it, but they're just thinking about it. And then they're bringing it to the broader teams to have those discussions started. Right now, again, this is not a model that is possible for every team or every company, because again, you need to be at a certain level to be able to do that. But if innovation is one of your core values and you think that you are at a stage where you have the data, but you don't know what to do with that, that setting up that innovation lab and setting up that setting is important to foster some of those. Yeah, well, I think some of the people who are tuned in, I think, certainly have those labs in place. Some don't, as you said. You know, there's a spectrum here. And I think people have had different kinds of experiences with innovation labs as well. We can end on this point, get a little bit of your perspective, but I want to shake this out because you're touching on something important here. I think some people see a lot of things die off in the innovation world. You know, like in other words, there's a lot of ideating and showing and tell, but innovation isn't tied in enough with the business where we're really understanding the problems this business unit is facing. We're really getting to enough meat and potatoes of positing ways to solve it to actually bring back something serious. It's almost kind of like an isolation idea box that doesn't ever have enough teeth to go back and hammer in a nail and really get something done. But maybe, you know, in some cultures, I'm sure the innovation department really is more actively a part of the circulating ecosystem of adding value, solving. And like you said, you know, being able to focus some of that attention on these data problems that are so important. What might be the difference in closing here on this interview between the innovation departments that end up being that little, we had a consultant on here once who called them kind of the sunshine group of the organization, right? We're just coming up with fun ideas, you know, versus the ones that, as you're touching on, might be able to start to untangle these knots, pose new ways to solve problems, be productive. What would be a core difference there that our listeners should know as they leave this episode? The key difference is the collaborative portion. Again, you have, but you have an idea, you have built something. But if you're not talking to the right stakeholders, and you're not socializing it enough, picking people's brain, their thoughts, accommodating some of their feedbacks. And then, you know, I would actually say, twisting and turning your business problem or the idea that you had so that it can fit into the business, bring business more money, that is a key differentiator. So you might, you know, you might have an X idea, but you might have to add some spices from on top of that by speaking to other people right and then come up with and then have all of those conversations and sometimes it's not just about spending time with data only sometimes you need to spend time with the business stakeholders and just understand what they're saying and come back and marry both the things and then come up with a problem and then an mvp which is very important because some of the use also die because it's a show and tell and then And it's, but it's just to show and tell on PPT or Excel and all of that. But sometimes you need to go beyond that, take a little bit more risk, build a minimum viable product and then end that. Yes. And what you're bringing up, which is a great point to close on, is that those MVPs will only ever see the light of day in a deployment if we have real buy-in, deep understanding, a tossing and turning of the ideas with the relevant stakeholder groups to the point where that thing is not just a show and tell. We're all just going to sit there bored and watch. Okay, innovation people, what do you got for us? But we're all excited. Okay, this has been a big problem that we actually want to solve. And like you said, data and business people both being able to collaborate with that group. Certainly a common theme in this episode around collaboration. I might even make sure it makes its way into the title because you've emphasized the importance so much here, Supreet. I know that's all we had for time in this episode, but we've got a lot of great educational points across and I appreciate you being able to join us. Thank you so much for being here. Thank you. Before wrapping up today's episode, we also want to point you in the direction of Supreet's recent appearance on the AI and Business Podcast. You can find that at the URL podcast.Emerj.com. Again, that's podcast, singular, no S, podcast.Emerj.com. The episode is titled Three Metrics for Measuring Enterprise AI Success. I would especially key into the first answer Supreet gives about assessment and personnel if you are intrigued just by the title of that episode. On behalf of Daniel and the entire team here at Emerj, thanks so much for joining us today. And we'll catch you next time on the AI and Financial Services podcast.