Achieving data maturity with Beverley Paratchek

With the rise of AI, leaders are starting to realize the importance of becoming “data-driven.” But for many companies, that's easier said than done. Many are wondering if there is a secret to becoming a company with strong data governance practices at its core?

Beverley Paratchek, an experienced information leader, has led organizations through the data maturity journey, and she takes Anthony and Kris through the key approaches she has used in her career.

They explore the intricacies of data governance, data quality, and the emerging role of AI as well as the need for continuous improvement through data maturity assessments, and the business case for investing in data quality.  

They also discuss:

• The important distinction between data literacy and fluency

• The importance of gamification in growing organizations’ maturity

• The synergy between AI governance and data governance

• And the future of data governance, and the need for organizations to enable their data practices strategically.

Key takeaways

• Data governance is essential for managing data as an organizational asset.

• Data maturity assessments are crucial for continuous improvement in data practices.

• Data literacy should be viewed as fluency, not just knowledge.

• AI governance must be integrated with data governance practices.

• The future of data governance lies in enabling organizations to make informed decisions about their data.

Resources

• 🎧The risks of unstructured data with Joe Pearce  

• 🎧Focus on reducing risk with Dr Miles Ashcroft
• 📏 Benchmark: How much PII does the average organization store?
• 📑 Blog post: Mitigating AI risk in your organization

Transcript

Anthony: [00:00:00] Welcome to FILED a monthly conversation with those at the convergence of data privacy, data security, data regulations, records and governance. I'm Anthony Woodward. the CEO of RecordPoint and with me today is my cohost, Kris Brown, our EVP of Partners Evangelism Solution Engineering. Man, that never gets any shorter.

Kris: No, never ever. I wanna add more things to it, just so I make you go through it every single time.  

Anthony: Today, we have an amazing guest. Beverley Paratchek is an experienced information leader and has most work recently worked at an Australian gaming organization. I hope I haven't messed your name up.

I'm notorious for doing that, but welcome Beverley.

Beverley: Thanks for having me.  

Anthony: It's great to have you here.  

Kris: Look. Thanks Beverley. Look, welcome to the show. I think it'd be wonderful if you could actually start by giving us a rundown on your career. [00:01:00] You know, as I said, most recently you were at a gaming organization, but I know you've got a heavy background also in financial services as well.

And then I also want you to put in there, because of that gaming organization, you did mention in our little chat beforehand, you've started to introduce, a bit of gamification and data governance, which I think that'd be kind of cool too. But talk to us a little bit about the career.  

Beverley: Yeah. And thanks Kris and thanks Anthony for having me on FILED. But yes, most of my career was in financial services across Canada and Australia, and then five years of consulting across everything from utilities, government, transport, education all the way through to. And then, you know, so I had a bit of confidence moving completely into a new field in, not same field, different industry with gaming and really, as you say, embrace that gamification to, roll out and establish best practices around data management.

But throughout all my career data and people leadership has been the common thread, and I've been involved in every part of the data analytics value chain from acquisition through [00:02:00] to structured storage and that value creation. And I just love doing organizations do things in safe, simple and strategic approaches.

So I also, I've done this at a conference once where I've asked people to put up their hands, 'cause data people will often very much go to one part of a spectrum or another one very much around the tech side of things and very comfortable with the platform and the pipelines and the engineering and the other side very much around, you know, the data science and mathematics and so on.

So, I don't know which end you guys are on, but I would describe myself as in the middle. So I've been at different parts, at different points in my career, but I am very much that data person. 'cause I think the closer you get to the data, the closer you get to really what's happening. 'cause that's the representation of reality within the organization and that's where it all flows together.

Kris: Yeah, I think Beverley, I probably would say I'm where you are, but I would've started more technology focused, very, very heavy technology focused. Then through the middle of my career, very much got much closer to the [00:03:00] usage of the data and the understanding of the data. Not so much the data science piece as much, but certainly that usage of.

And its value. But as you know, we're moving back towards AI and Anthony mentioned in the intro, I'm very much coming back to the center again and having to be a little bit more focused on the how, as much as the what and the why.  

Beverley: Yeah, and that's where more recently my roles have been around more that I, I call it data enablement, which is strategy, governance, maturity, quality, et cetera.

But I wanted to touch on the gamification piece because usually the first step, if people are getting involved with, you know, increasing data, data management practices, it's often data governance. And you can just see the eyes roll and you know, so many comms people say, oh, here we go again.

But I really find leaning into that is a bit of fun because data governance is pretty simple at the end of the day. And so when people tell me it's boring, like I create a competition where embrace your inner geek and tell me the most boring thing you know about data. You [00:04:00] know, if you're doing data maturity, think of it as you would in a game where you level up. You always have that progress meter of where you are at that time. You could even. A bingo card for self-based training or a trivia, or, you know, recently did a treasure hunt where people had to go in and actually figure out what is the source of that data, what is the definition of that piece?

And they actually had to use the tooling that we had built out. I think you guys are familiar, but I don't know if everyone in your audience will be about Tim Tams, but that is a really good draw. If you can do like a spin to win little game and have a bit of chocolate if your data's absolutely perfect.

But of course, if you've got a data breach, sorry, you're gonna get something that's more healthy so that you know how to take care of your data better. So those are some of the things I've used in the past. The other bit is just, you know, typical strategy 101 is having the iconography really reflect the organization.

So, you know, if [00:05:00] you're in a health scenario, you probably will talk to data health and so on. If you're in a water utilities, you might talk to, you know, safe. Turning on of, of. But always on tap, for instance. But if you're in a gaming organization in that boring circle, that's usually the data governance framework, you know, with it has the different pieces, structure and so on.

You can make that into a draw machine and have the little balls rolling around with the different pieces of the framework. So I think having fun with it and really making it feel like it's part of the organization rather than something put on top is really important.  

Kris: That'll resonate a lot with the audience. Beverley, one of the favorite ones that I've seen in customers I've had to deal with is that they've actually gamified it back to their salary. If you achieved certain number of achievements, off the back of the data governance strategy that actually led to, a portion of their, their pay rise or their KPI pay rise By putting those sorts of things in and having them measured and feel like they're a part of what's going on, as well as then the roll up.

To the greater organization. You know, again, those levels [00:06:00] really helped. I thought that, those challenges, helped to bring at least a little bit of value to, as you said earlier, what is a very valuable part of the organization. And I know, you're a big driver in those and some of the conversations you've had online and, at conferences and other things, which, we'll dive into in a little bit.

Anthony: Yeah. Look, I think you touched on a couple of really interesting things, talking about that gamification around, maturity of data and upleveling as you described it. Which I love. You know, your data, risk position and those things it'd really good to drive in. One of the things I've always struggled with is,

probably to admit my biases, legal background and then background in technology as well. But having strong frameworks to actually make those assessments is really difficult. How did you do that to create, to gamify? You actually gotta have levels and you gotta be able to connect those levels to something.

How did you go about that?  

Beverley: Yeah. And that's where a lot of organizations will, do a data maturity assessment, and I've seen it at different levels of maturity within the assessments [00:07:00] themselves. I think it's a really good tool when data management's a strategic priority, although it's still linked back to risk and compliance.

As you mentioned, it's not the primary driver. You're actually looking at data and using it as a strategic asset. So probably it depends on the organization. I think if an organization's just starting out and they don't have, a metadata repository, they don't have the data governance processes defined for how it should work in projects, then it's kind of mean to come in and say you're level zero or one because they already know that.

You know, that's not really gonna build people up or anything. So I think maturity assessments need to come in at a time when they're ready for continuous improvement. They know the basics are there. They might just might be using them inconsistently, or it might just be project focus is, which is usually your level one.

And then there's just the mechanics of the model itself, just being really clear on what each level looks and feels like. And then also there's, depending how you do it, you can actually evaluate like the maturity of the central data team and those core [00:08:00] processes versus the more decentralized hub and spoke model and how it's actually happening on the ground.

And as I say, I personally like to lean into the decision rights around data. Like the data council will be the one who will endorse the model. They're the ones who are going to set the target and the timeline for where it needs to be. The data owners need to accept their baseline. That that's probably been, assessed by the custodians or the stewards.

And then, your risk and compliance people need to come in and they need to do the linkages back to the effectiveness of the data controls and maybe have an action plan linked back to the customized roadmap. And then lastly, things like your board risk committee will have a role to play in monitoring to see whether this is within acceptable tolerances of what's needed for the future.

But if I talk to the model itself, it's usually attribute. Everything from data quality, metadata ownership strategy and project governance, retention lifecycle, like you can go on, right? But my [00:09:00] recommendation is to customize it for the organization, and especially getting into things like understanding what is the linkages to business outcomes or, you know, that return on investment around data is, is I think where there's a lot.

Kris: So what does that mean though, for the, the business? So, so you're asking them to invest in these maturity assessments and then obviously the, the upskilling and progression, but how are you, how are you selling that?

Because I guess the data now to talk about the data, but the data that I've seen around, you know, organizations who have those very strong mature data practices are performing, you know, more than twice as well as those who aren't. And when I'm talking about performing better there, we're talking about like revenue or profit.

Efficiency, like real, tangible business outcomes. Is that it? Because, and maybe you can help me understand here, but what's, what's the investment to progress? Because you don't get double the revenue for free. That that doesn't happen.  

Beverley: And not go backwards. As well, but I have seen organizations go backwards [00:10:00] as soon as, the attention isn't there anymore and it takes as much to embed those new practices as it is to introduce them.

So yeah, the main selling points I would say is transparency. You know it, I've been reflecting even, recently around data quality. It's really hard to make the case for data quality because there's a lot of what I call unsung data heroes out there. Just making sure that those insights and those board papers and those financial reports come together good enough, or that they can stand behind.

But how much are they hiding in what's going wrong or being that hero working through the night, or. Doing a manual thing to make it work. So these assessments really add transparency so that senior leaders know what's happening and they can make those trade offs of where to invest. The other selling point is actually giving agency to people within those roles to make a difference within their substantive role, not their in top of my day job type of role.

'Cause a lot of people just have been dealing with data issues for years and just don't really [00:11:00] think it's ever gonna change. And, you know, when finally someone listens to them and can really frame it in a way that resonates with senior leadership, they will pick up on that and drive it.

And the last is just that coordinated effort. 'cause data is the most pervasive asset in an organization, and it's not something that just one person. And there's a lot of good work happening already. It's that consistency and coordinated effort across an organization to grow together and then to really build that trust between data producers and data consumers.

Kris: So projects and what kind of efforts, you know. How do they get to that place? Because you were saying you can go backwards by losing that concentration on it. But what are the projects and efforts that you've seen in organizations or that you've been a part of that, really help to derive those outcomes?

Beverley: Yeah. And if I take one, which might be topical for your audience around data retention or what I like to call data deletion, for instance, is really getting that discipline around getting rid of dead data. So data that's no longer needed, not for whether it might be [00:12:00] a legislative requirement or a regulatory requirement, or some other business operational need, like that's super, really brave and goes again.

Everything people have been comfortable with for years. So that takes often, it is a dedicated investment just to get the ball rolling, to get rid of that technical debt I would call of, of what's been happening over the years. But then the discipline of doing that going forward really can't be, as I say, an add on our attack on.

It needs to be considered part of people's roles. And that's where the investment is, right? So having those roles report to a data owner or someone who has the operational risk ownership and process ownership around, that's really important because they might say that particular dead data or database just.

Really can be deleted. But this other bit, you know, it's really low risk. It's sitting somewhere with no secu, like there's no access, and you know, I'm fine. Just leave that where it is. And [00:13:00] those are the trade-offs, and that's the decision rights that you need to define across an organization. And then support them to make it happen and share, share, share the stories.

Anthony: Do you find there's an issue though, as you start to work through those things at data literacy? So one of the things we hear from our customers and we talk a lot about in terms of maturity, is there's this real variability of literacy across different business units and then how they both. Hold the data of data providers and then also how they provide it to consumers, both upstream and other places in the business.

And how do you grapple with that in the framework? 'cause it's, you know, the framework's a little bit blunt as an instrument and it doesn't understand these different people elements.  

Beverley: Yeah, there's a few things there. One, I find and when you talk about those trade offs and if they are really like a throw over the fence or not my problem type of thing, you'll find an over bacon of data controls across the lifecycle.

'cause people don't trust what's happened upstream. So that can be a risk as well that gets [00:14:00] uncovered with data lifecycle management. But I think you were touching on literacy. I actually prefer to use the word fluency. 'Cause I don't know if you've ever learned a language, but to call someone illiterate probably isn't the best way to start.

There's different levels of fluency and not everyone needs to be fully bilingual in order to fulfill their responsibilities. So I think of it as levels of fluency and you're starting getting into the data culture and the capability build. And that's where a maturity model, if it's right for the organization, really sets baseline understanding of what good looks like.

Right, because people have looked at, well, I'm at a level two and a half. What do I need to do to be level three? And everyone has that same model, that same understanding. No surprises. When the inevitable audit does come, if you have taken those self-evaluations beyond self-evaluations and into evidence testing and even moderation, then you know that's gold is having an audit come in and say, yeah, you've done this [00:15:00] right at world class level and makes everyone proud.

They know where they're at. There's no hidden surprises.  

Kris: I do like that. Ever. So slight change of literacy to fluency there. As I said, it's interesting how, the negativity that can be associated to a piece. I was sort of sitting there going, what's the ill of fluency?

It's definitely not ill fluency. It's just a level. As you say, I'm level two fluent. It gives me ability to roll it. Back to the model, which is interesting. To tie in a little bit of the topic of the year, it feels like for the podcast, but, how does something like AI governance flow into this?

What's your thoughts in terms of that? An organization who has a higher level of data maturity? How does AI governance become a value add to them?  

Beverley: Yeah, and I think I read an article you published recently, Anthony, around data governance being an input for AI governance and they are very symbiotic and you do need one for the other.

I think you were touching on that it just can't be attack on AI governance can't be attack on the data governance. It's something different. And I 100% agree. Like if, if I, when I talk [00:16:00] about data governance and what it is it. You know, it's all about owning and managing data as an organizational asset, and it's often around capability building, whereas AI governance is very much around is it the right thing to do?

Is it legal and is it actually a good model that's performing with the right controls? So I think data governance is very much capability decisions type of rights, whereas AI governance is probably where it is at right now is probably more around gatekeeping and approvals, which are different ask of people and different audiences.

So the way I think they work together, like a data mature organization would do better at AI governance or even AI initiatives because those decision rights from data governance. Help set up an organization around talking around things around who is an owner for AI use. Like that's one of the voluntary standards in in Australia, at least, around defining who is that owner of the use, who is accountable for model of performance and when unexpected impacts do happen.

Does that accountability still lie there as well. The second [00:17:00] one would be around. You know, and data governance, we're very good at defining metadata and that vocabulary of the organization and that context of data within the context of business outcomes is gold for any AI implementation. And then lastly, you know, data governance.

It's very good at defining your business rules and your classifications. You need all that to define the guardrails and really codify those guardrails so that you can monitor and have the controls in place over time.  

Anthony: And there's a really interesting crossover there. Right. I completely agree with all the things you said about quality though.

Because we can't have the AI sets a process. We can't even talk about trust and bias and all the things that, that come out in that AI, as you said. Stay, you know, the stage gate until we talk about quality, and I know that's part of what we're talking before about maturity models, but if, I'd love to drill in to how you think about quality.

'cause quality isn't just about. Understanding, you know, the data in front of you [00:18:00] and is it in the right columns? And if people check that the, the names are spelled properly, and cross Referenc is much deeper than that. So how do you, how do you go about understanding that and unpacking that?  

Beverley: Yeah. Good question because data quality, like if you go back to the textbook version, it's fit for use, right?

A lot of people will complain about their data, but they're still using it.  

Anthony: Yep.  

Beverley: So is it actually a quality issue? So I think there's a lot of different assets around quality, but. If you were gonna go beyond their traditional dimensions, and I often add precision, I think in the era of AI, I would also add variability or representation of data.

So earlier in my career when I was more on the analytics side of things, we were doing credit limit, credit limit assignment optimization. And before we even got there, you know, our chief data science is like, we are gonna have a program of testing and, and cell design and segmentation for, you know, six months to a year so that we actually know what.

Is the optimal answer, rather than basing it on exactly what we've seen in the past, [00:19:00] which is quite limited. So I think that representative layer of quality is really important within AI. Otherwise, you're gonna be leaving money on the table around actually what's available to optimize. And digital is so good at this, like they're always doing AB testing.

Anthony: No, and it's a really good test, to break into the representational areas. But what about things where the quality, and this is probably gonna get a little deep, and I apologize to the audience upfront, but when you think about representational importance, the model's performance themselves.

So one of the issues you run into a lot, around data quality isn't just the quality 'cause it's in use as you say, but once you change that use. The performance changes, which actually degrades the models and degrades the data sets itself. How does that mix into that world?  

Beverley: Yeah, and because that use part of the equation for data quality is huge, especially around privacy.

Like if you have consent to use for particular, outcome or service, you just can't pick up that piece of data and use it for something that wasn't [00:20:00] consented to. So there's also a whole piece around data's never gonna be a hundred percent. So how do you actually scope, the models and the logic to make it so that it can deal with poor quality or what's considered poor quality data as we have in the past.

So I think, there's definitely a piece around that. In that, you know, it just needs to be able to have a safe answer if it comes across something unexpected, like a person would and work through it or escalate it because it's outside of the level of delegation.  

Anthony: Yeah. And okay. I dunno if it's worth touching on and things like ground truths and then trying to do that back to understanding the calibration of the sets.

Is that something you get down to when you think about the maturity, or do you find that's too far for the average person?  

Beverley: Ground truth, I haven't heard that term before. What do you mean?  

Anthony: Well, ground truths are ways to look at the representational nature of the model and make sure that there is something to compare it to all the time.

So challenging the assertions or representations of the [00:21:00] data sets that you're looking at.  

Beverley: Yeah, And then, there's a whole piece there because then there's also the adversarial type of testing.  

Anthony: Red Hat. Yeah, exactly.  

Beverley: Yeah. And this is where the appetite needs and the strategy needs to come back to.

Because as soon as you embark into those types of things, that's where the governance needs to match you on your ambition. Around controls and so on. Even traditional data governance is now gonna be dealing with a whole nother data set around just even the inputs and the prompts and, what is the rationale or being captured for some of the decisions being made?

'cause if you're in a highly regulated industry, you need to be able to prove that.  

Kris: So Beverley, you've sort of given me a quick little segue there and I always like to ask a question along these lines as we sort of get towards the end of the podcast, but put your crystal ball, put the magic hat on, whatever it is that you use to predict the future.

Where do you see,  

Anthony: so, you know, Beverley, Kris uses a magic eight bolt, so it's,  

Kris: everything's a, maybe always, maybe.[00:22:00]  

This how all of my decisions are made, and there're always a maybe wonders why I'm able to get anything done, but where do you see data governance going? What's your prediction? You know, we've got a really interesting world at the minute. I feel that, data is more important than ever. I get the impression from yourself that that's, that's, organizations are only just realizing that and that, you know, the, the, the data they've had for a very long time could be highly valuable.

It also carries a lot of risk. Where do you see. Data governance going, what's the, what's the next 12 to 18 months look like with all of the innovation that's happening around us?  

Beverley: I think I was asked this question six months ago. I don't know if my answer has changed too much, but  

Kris: Okay. We will roll it back and play it back to you in a while too, so don't worry.

We, we, we are keeping all of these.  

Beverley: Well, I prefer the term data enablement 'cause I think data governance is meaning different things in different organizations. Like it's getting into content management, knowledge management, getting into, what is the inner interfaces with things like data security and data privacy.

So it will be different [00:23:00] for every organization, but I think taking that enablement hat of what is needed is important, but also what's needed now, what can wait for later, and that's where I really like the capability build hat of data governance in that, having a data governance checklist isn't really going to solve the world.

It's giving people the tools to make the decisions or to raise their hands when they see something that needs to happen, which could be anything from, you know, this was being emailed and it shouldn't have been. Or we're collecting data and making 10 copies of it as we pass it through the workflow.

Like those are things now, which could be exasperated if, you know, agentic AI isn't well designed.  

Kris: Is the, this is the first prediction we've had that's been a little bit negative. Like, it's almost I don't necessarily disagree, to be honest, Beverley I think there's a real situation here where poor practices of the past replicated in AI in a way that has infinite possibility, creates incredibly poor practices in the future.

Yeah. So I think it's a really interesting position to take.  

Beverley: Yeah, I don't know if [00:24:00] it's a position, but it's just, you know, I'm learning just as much as everyone else's. And I think, as AI becomes that, what do you call, almost like a broker for transactions, for our daily admin, that whole piece around data ethics is going to be more and more important.

The traditional approach to data ethics will need to be scalable to a massive degree.  

Anthony: Yeah, I mean it's probably a whole other podcast, I think we started this conversation talking about models and frameworks. But those models that exist today, CMMI, and all the various iterations of those things really are.

Tick boxes of behaviors. Not, they're not truly, there are elements of the, are empirical, but most of the things are not truly empirical. Right. Because you know, you can't just put a test harness around it from a technology perspective and go, yes, it's one of these as a binary answer to that result.

Do you think we're getting to that point though? Is that, do you think that's where the agent and AI elements will take us, where we have this culture where we can actually go. This is good grounded data versus this is data needs more work and we start to put [00:25:00] some more concrete answers around these things.

Beverley: Yeah. And I think it'll come back to what's needed. You know, none of your data is ever gonna be perfect, but is it, is it at the level that can be acted upon with confidence? And what is your, what is your confidence interval around that? Right. So I think it's 'cause you could spend. Heaps of money and you know, probably have data pipelines going everywhere and you're gonna come right up against the timeliness issue of data quality.

So if you pack it with lots of data controls and manual checks or, you know, so there needs to be a, a way to scale  

Anthony: that. Great answer. I'm actually sitting in San Francisco today at the Databricks Conference. And this has been my whole day having this same conversation. So it's really good to hear, I think, a real practitioner rather than a bunch of, technologists talking about how, how theoretically they'll get it done.

So, so it's very enlightening to, to capture that conversation. Very fascinating. I.  

Beverley: Yeah. Thank you.  

Anthony: Look, I think this is a really fascinating area that [00:26:00] we could drill into for quite some time. There's so many different aspects of qualitative and quantitative and data mining, and loosely coupled and tightly bound data, and all sorts of elements we didn't get to unpack.

But I do appreciate you, on the podcast today, Beverley. It was, a real pleasure to dive into some of these things.  

Beverley: Yeah, and I apologize for my darkness. It's definitely not nighttime here, although I think it's nighttime for you. Anthony,  

Anthony: Not at all. That neither is a very bright light in front of me, but, uh, it's doing its job, which is, which is great for the day.

Kris: Thanks. I'm Kris Brown.  

Anthony: And thanks for listening. I'm Anthony Woodward. We'll see you next time on FILED.

‍