Data Governance 2.0: Transforming Control into Collaboration with AI

Speaker A: Foreign.

Speaker B: Hi everybody. Welcome back to the Light Zone Data Show. My name is George Firikan and I'm your host. And today we're going to explore something that uh, we're, we're going to call Data Governance 2.0. It's kind of the shift from governance as a control function to governance as a, uh, collaborative AI powered enabler. And to get us through this evolution, I'm joined by somebody who lives and breathes these topics every single day. Najeem Ashufta is the founder and CEO of Driva and a longtime leader in data governance, metadata management and data culture overall. Najeem, I'm thrilled to have you here.

Speaker A: Hi Josh. Yeah, it's a pleasure and honor to be on Lights on Data the show. So, uh, yeah, I'm very excited about our conversation and the topic.

Speaker B: Yeah, likewise. So, uh, you know, I love to start with the idea of uh, data governance 2.0 and what that means in practice. And from there we can explore maybe how AI, metadata and culture overall kind of work together to support this new model. So when you hear the phrase Data Governance 2.0, what does that mean to you? How is it different from the traditional way that organizations approach data, uh, governance?

Speaker A: It's still something that we need to kind of um, define with more certainty what it stands for. But I would state already that I think Data Governance 2.0 is already trying to um, mitigate the ambiguity of what we know so far as data governance. Because data governance at the moment, there are tons of definitions and understanding what data governance stands for and how it should be, uh, uh, understood and lived within organizations. What is part of data governance and what is not part of data governance, then at least, uh, based on my understanding, was always something causing conflicts amongst the data professionals. The data governance manager that we're debating is, is that not part of data governance or not? I'll give you an example, uh, data literacy programs. So is data literacy part of a data governance manager or not? So I think data governance 2.0 is already making sure that we all have, have a, have the same understanding. What does data governance really bring to the table and what does it deal with? And I think that's already a big benefit.

Speaker B: You know, it's, it's funny because data, uh, governance is supposed to provide clarity overall, but even within the definition there's so, so many different, uh, ones. When you do a quick Google search on what it is, you get so many different examples. So even that it's kind of conflicting with, with its hope to start with yeah, exactly Emily.

Speaker A: And I think what Data Governance 2.0 is now highlighting more is the entire topic of stewardship. Yeah, um, and what stewardship should all cover in a sense of, so that it becomes something more beneficial to the governance, um, or to the organization overall. So I would say when we look a little bit into the history of data governance, it was very much focusing on the creation, the maintenance of master data management, transactional data reference data, um, and it was very much about having the rules for the data lifecycle. So now the difference is now that I think data creators, if we want to call it that way, and data consumers in data governance consumption 2.0 are getting much, much closer to each other and how they define data, how the data should be available to them, and how data in different scenarios will be applied and has to in that sense also provide a certain level of data quality. So I think data governance now is much more about um, connecting rather than clearly stating this is what you do and this is what you not do. It's more about making sure that end to end operating models that are not only about data but about tech and people and change and the usage of AI solution and so on, that this is holistically addressed. So it's about bringing the right people to the right conversations in allowing them to define then the necessities of how the data should be, as I said from a metadata perspective should be defined, how it should be available throughout the life cycle, uh, even to the extent of that we say the life cycle of analytical purposes or AI purposes, how it is then throughout these stages to be then defined and the quality is ensured and how stewardship. And I don't talk only here about the role of a data steward. I talk about the entire concept of stewardship, how it should be then addressed and then how it should be fluently, um, handed over, the responsibilities.

Speaker B: Mhm, mhm. So it's looking at that whole um, data lifecycle from data creation or acquisition, maintenance, dissemination, usage, and the usage can be anything anywhere from um, showcasing that data in a report to make it available for uh, into a data product or for AI data analytics, what have you, and then tying it all together within metadata and data quality and observability. Anything that is needed to make sure that data gets converted into that asset that we keep talking about.

Speaker A: Yes, I think so. And I think that um, the data governance nowadays, as I said, it's something that is more kind of present but intangibly kind of little bit present. So it's a little bit like When I have to imagine it like in the past or the way when, when I started it was like um, it's like the auditor came to the room and then it was like okay, we have to check if everything is according to policies, if the SOPs, SOPs are addressed and um, you know that everything is compliant and so on. And nowadays the way I see it is first of all is that we have data domains and uh, these data domain representatives, whether it's a domain owner or a data steward from there that they together with other domains and maybe a central team try to answer business related questions such as uh, uh, when we think about how can we improve our customers service or how can we reduce maybe um, certain procedures in the supply chain that they together with the other subject matter experts when it comes to process, when it comes to the applications and so on, sit together and try to see how they can first of all make sure that everyone is using the same semantics when they talk about data objects and then that there's a clear understanding of how the data needs to be so that the process or the functional um, the functional capability is performing better. Yeah, yeah. So and I think that is where the data governance now comes in. By not only providing data repositories that give us a better understanding of. This is how we define for example ah, headcount and this is how we define numbers of employees. It's also about who is the right person to talk to when you're dealing with these topics. So it's the connector uh, of data people and responsibilities. So that everything. And you said it is becoming more enabled to operate more efficiently.

Speaker B: Right. And ah, yeah and I like what you said and how it evolved from um, providing that compliance into more having an enabler function. Because compliance happens anyways if you shoot for enablement. Right. I see it as a subset and you shouldn't just stop at compliance. And what I've talked to companies in the past, it's very similar to what you're. Where you're mentioning when they've even shifted that mentality that it can be more. That's when really magic happens. This data uh, governance can provide its, its benefits so much more than just being compliant, just being controlling.

Speaker A: Yeah, definitely it is, it should be much more. And I think um, for me it's like when we look into, into schools. Yeah. Let's, let's take American football. Yeah. We have like these, these uh, these playbooks. Yeah. With all these uh, ah, positions and movements and so on. So whether it's a defense, whether it's the offense where regardless if you are at the front or you, you are scoring wise behind it, it gives you a playbook, a scenario how to respond to these different uh, circumstances and for different roles and responsibilities. And that's how I see data governance. Data governance is the. Okay, I don't really know how to continue. So I go to my data governance experts or um, uh, I look into the playbooks and then it tells me, oh, you have issues with data profiling or you don't really know how to uh, identify the data quality defects and so on. So how do you do it to who do you have to speak to? And so on. So it's an answer, uh, provider, a solution provider that solves the issues. Um, and naturally it becomes a different, it becomes as I said, the enabler rather than the oh, uh, we now have to involve data governance because we cannot go into the next phase of the project unless we don't have the approval, you know.

Speaker B: Yeah, and I see this as a natural evolution because data has definitely moved from just reporting to really uh, operating the business in real time and with AI, with automation, you know, product analytics, um, data is something that drives decisions instantly. So you need to have something a bit more nimble and not have data governance like you said, seen as ah, as that roadblock that you need to go through to ask permission for things and move the project forward.

Speaker A: Yes. And I think when we address now a little bit also the topic of uh, what has so much changed is the fact that now when we talk about the interaction with data, we now have not only the data domain representatives or the lifecycle representatives, we have also AI agents that are dealing with the data. So that means we have here also collaboration and interaction with uh, artificial intelligence that is creating, maintaining, using data. And it also has to kind of obviously uh, fulfill here certain expectations. And I think that is something that uh, is of very importance because I think at the moment we already have the situation where it is quite difficult to anchor uh, strongly the work responsibilities of data, whether we call it data ownership or accountability and so on. But when we talk now about the agents that are also taking over maybe accountability and ownership, that's a different ballgame. Yeah, yeah, and, and, but it all starts with a common understanding of what data governance should really bring to the table. And for me it's much more something that is principle based, allowing procedures to be more um, efficient across people and system. And I am a fan of ah, as much data governance, whatever we know, which aspect we address that, that Is necessary. Yeah, so I wouldn't go always full blown, uh, because the level of maturity also in governance is very different from function to function, uh, to entity to entity, from division to division. So we naturally have not average or common maturity level of data governance. So that means we always have then to make compromises in the expectations and in the interaction when it comes to different data governance aspects.

Speaker B: Mhm and ajim, what would this then look into practice? How um, would that data governance program look like from let's say the more traditional version that we're familiar with?

Speaker A: I think, I mean I think um, for me data governance in general is something that will be federated. It has to be federated, yes. Having that said, I think there's always a level of degree of governance that needs to be centralized and it always comes down to how healthy your data governance is performing. So level of understanding and level of experience determine how much you can decentralize of your data governance, how much a domain can take over, uh, your data platform can take over, or how much it can be um, governance principles, rules and so on can be uh, part of the code. It has a computational governance, so it always kind of depends. But the target is to have as much as possible decentralized and then with a certain amount of uh, maybe KPIs and reporting, uh, structure allowing then the central team to coordinate and make sure that there is improvement and the governance principles are in a way um, enabling as I said, functions in business to continue. So that is maybe one key aspect.

Speaker B: Do you see then uh, the central team being more uh, similar to a center of excellence where they're providing templates, best practices to then uh, disseminate to uh, these individual teams within their own units, within their own areas of the business to do their own mini data governance programs as it pertains to their ecosystem.

Speaker A: I think yes, to a degree, uh, this will be one of the main aspects of their responsibilities to provide as you said, the templates, the good practices within the organization that have proven that for this organization this way of governance or uh, procedure works. The other aspect is that they, as I said at the beginning, they are also from my perspective also an entity that, I mean we talk nowadays about data as a shared asset, as a shared good. This is why the topic of ownership is also so difficult because how do you want to have ownership for something that is, yeah, a shared asset, a shared good. So it causes a lot of um, debates and conversations. So um, to what extent is the accountability, which I prefer, maybe in one domain and Then the accountability then will be transitioned or handed over to another. I think that's also a responsibility of the central team to make these clear judgments. If there's nothing coming out of the domains, then based on their responsibility, they should have the decision. Right. To make some calls in these kind of scenarios.

Speaker B: Thank you. Yeah, um, and you mentioned A.I. a couple of times in A.I. agents. Um, I haven't seen it in practice yet to be honest. Within the data governance domain though, I've seen um, service providers providing tools for agents on data quality, on data classification, on uh, semantic tagging lineage. Have you seen a successful uh, agent within the data governance domain being deployed? And maybe some learnings from there?

Speaker A: I think. I mean I have the same observation as you that there's a lot of heat around it, let's say that way, that uh, this is the future and this is how it will be. I have seen occasionally some examples where it really worked out. And um, the governance aspect here that I would like to highlight was that in the scenarios that I was dealing with it was always very much compliance, um, or regulatory, strong industries and sectors. So here it was a lot about governance in a sense of um, making sure that the agents are working in an ethical manner. So when we have to deal with uh, personal informations of the uh, insurance, uh, policy or the patient data and so on. So again here it was very much about governance having an eye on is it ethical? Is it morally acceptable to work that path? How do we make sure that the agents are not using really the data that is, uh, that is m. In a way anonymized or sodomized? Yeah, ah, synthetic data. So to what degree can we allow judgments, uh, about how data products or data insights are valid when we use uh, synthetic data? So here the governance aspect was very much about these ethical questions. And in that sense it was not only one domain, but many domains, um, trying to solve these questions together.

Speaker B: No makes sense. And especially with AI implementations nowadays, uh, you kind of need to design for explainability, not just accuracy, uh, you need that transparency, try and understand the best as possible what's within that black box that we often see, uh, uh, on our end. And for that we need that clear ownership, um, traceable lineage, uh, you know, the human in the loop review for sensitive use cases and so much more. And of course the metadata as well and the semantics that go with it. And data governance is a key for all of this.

Speaker A: Exactly. The question is who has a better understanding of the data output? Is it a human? Is it, Is it A human or is it a machine? Depending how the data has been created and been made available and so on, it can be either or, and we have to make sure that definitions, metrics, logics and how it is interpreted, a human and a machine, where it is necessary, have the same understanding.

Speaker B: Yeah.

Speaker A: And maybe it's. Yeah. And also the question will arise do we really want in certain business functions and in business processes that a machine takes over these kind of decisions or uh, execution steps. So while we explore certain use cases, we have to touch it all these kind of uh, uh, questions and, and uh, and, and, and ask us what do we really want. Yeah. Uh, so there are some maybe no brainer situations where we think like yes, uh, the machine can do that. But then there are ones which is very dryish to answer that question.

Speaker B: Right. It's um. Yeah, that, that semantics of it. It's kind of funny oftentimes a company goes into an AI deployment uh without uh, the company really investing in metadata first and understanding their metadata and then they're expecting for the AI to magically figure things out. And I mean I think it can certain parts of it but then its understanding might be different from your understanding when you're looking at it and might not be the correct one. What would you say are some, maybe some practical steps that organizations can take to make that metadata visible and meaningful? Uh, and maybe in daily work I

Speaker A: think there are different ways of doing it. It always kind of depends again on the circumstances. So there's also the question of what stakeholder do you have in front of you and prefer to do it very top down. Which would be you come through a semantic or you come through maybe even knowledge management. And then the semantic layer, the conceptual layer, uh, and then you go all the way down to a physical um, metadata model. That would be maybe one way especially if you look to regulatory ah, requirements. Um, um, I mean we have in Europe we have now the csrd, uh, which is uh, about corporate social responsibility, uh, directives. This is the full name of it. And here we talk about a lot about uh, KPIs, um, about gender pay gap, emission rates by different uh, uh, energy sources and so on and so on. So here you come through the semantic layer, uh based on the regulatory requirements.

Speaker B: Right.

Speaker A: So you do um, so what I would recommend how we did it with our uh, partners and projects was we took the regulatory requirements and then we talked with the legal team, the manufacturing team, the HR teams and so on and so on about how much are they already using this kind of information, do they have already reports in place? And then we would uh, liken the shadow a little bit. So this is again maybe a good example where a governance manager would kind of lead the conversation about how are uh, your procedures, how do you fulfill currently these kind of requirements and so on. Then someone who is more responsible for data architecture, data modeling is observing this conversation and is writing down all the data terms and objects and is trying maybe to kind of draw uh, a model on a semantic layer. So this is maybe a way of how to start it and then bringing the logic into uh, a view where you can see maybe a little bit of lineage and relationship between um, these data objects and the semantics. These. The other way is obviously if you're dealing with a very technical team, then I think there are nowadays solutions which can give you uh, quite good overviews on your physical data model and allowing you to start from that side that you say, okay, this is how our physical data model looks like. And now let's start to bridge it towards a uh, logical and uh, conceptual layer. So it depends obviously organization that the experience of the people that you have there and you can engage your conversations.

Speaker B: Yeah, oh definitely. And uh, you know, mentioning of the different people that you have within the organization. Culture. Culture is often maybe the hardest part of data governance. How would you say, or what would be some of the characteristics of that culture for successful data governance 2.0? Um, implementation.

Speaker A: I think what is important is here to when we talk about data governance 2.0 and how we should address the cultural topic is we should stop to kind of isolate the organizational culture in general or when we talk about data literacy or data fluency. Um, I'm actually not a fan to always give it a very, you know, like specific term. I would say data literacy should be part of the company literacy.

Speaker B: Mhm.

Speaker A: So facet of it the same as a data strategy is a part of the company strategy. So by creating these barriers in front of our eyes, then this is business literacy, this is data literacy, this is process literacy and so on. We, we hinder people to see that this is actually everything belongs together. So I think the questions that here data governance um, should ask, especially on the top of management, is what do we really want to govern? And ah, is it data is something that is reactively looked in or is it um, a business model, a product, a service that we want to bring out into the market and we want to generate revenue? And then we maybe look then as data governance managers strongly to the data aspect of it and make Others more aware of it. Yeah. But I would say the data governance 2.0 starts not with the question of um, where is your data? And uh, how have you documented it? And so on. The question is always, which purpose does it support? What's the benefit of having this data, uh, set and data, uh, reports and data products and so on and so on. What's the business model behind it? And I think this is where data governance manager also have to have more uh, stronger understanding of the business model, the operation model overall and the strategy of the organization. Because then they can better also place and advocate for the necessity of ah, enabling data governance. Because it all starts with what do we really want do with the data.

Speaker B: Yeah, Evie, for companies, um, organizations that are listening to our conversation now and they're thinking, okay, how can I improve my data strategy? How can I improve my data governance program or even start one? Um, how can I engage with uh, Driva? What is the first steps that you usually go through?

Speaker A: I mean, how you can engage with us? Very easy. You can go on our website, you can contact us on LinkedIn or have a session with me or my colleagues. And the first thing that we normally do is we try to understand what is really the issue in the room. Because as I said, a data quality problem is only a problem if it's not supporting the business. So in that sense we would like to understand what is really the problem where they do need support and then it's very much about meeting them there where they have the gaps. It can be skill wise, it can be capability wise, it can be educational wise. So for us, execution and literacy goes hand in hand. Maybe it's something where they need more support, the literacy to be better in execution. Maybe they have the concepts and the literacy, but they don't know how to bring it into the operating world. So based on these conversations that we're going to have, we then engage with them. And um, I always say, like, uh, it's a little bit of a tapas setup that we do, so you don't have to go for the whole picture. It can be very use case specific, capability wise and yeah, step by step, taking it from there.

Speaker B: Driver. Yeah. Thanks so much. Thank you, thank you for sharing. Uh, these, thank you, John. These best practices and yeah, uh, data governance 2.0. I'm hoping more and more we're going to hear about us. And not only that, but we're going to see companies really adopting all of these and start implementing it into the right direction holistically. Thanks so much. Najeem.

Speaker A: Thank you, George. Thank you for having me and, uh, uh, having the chat.

Speaker B: Oh, it's been a pleasure. Thanks, everybody. And until, uh, next time, let's keep putting the lights on Data.