Episode Two: Data Makes the World Go Round with Dr. Fern Halper

Welcome to Redefining AI. I am your host, Lauren Walker-Zaffer. Today we're thrilled to have Dr. Fern Halper join us.

Dr. Halper is a nationally recognized industry analyst, educator, and premier thought leader in enterprise AI strategy and governance. She brings a rare blend of hands-on practitioner experience, academic teaching, and over 30 years of industry thought leadership, having published hundreds of times on data mining, analytics and artificial intelligence. She holds a PhD from Texas A&M University, has served as a lead analyst at Bell Laboratories and is currently the VP of Research at TDWI and a founder of the AI Foundations Group.

She joins us today to discuss her highly anticipated new book, which is Data Makes the World Go Round. The data, tech and trust behind AI success, which serves as a comprehensive strategy guide for business leaders looking to drive real world results with AI. It's wonderful to have you here with us today, Dr. Halper.

Welcome. Thank you, a girl to be here. Thank you so much. and you said in the introduction that you're currently in New York is that correct yes in New York it's a nice weather today I was gonna say it's gonna be very hot today so we're getting a burst of summer a burst of summer summer is always an interesting month it's one that a lot of people look forward to what do you enjoy about the field about AI yes how did you end up here what do you enjoy that?

I think I've always been really interested in what's next. That's why I became an industry analyst. I started off as a data scientist, really, at Bell Labs, analyzing data for AT&T and building models before AI became popular. But I always wanted to know what's next.

I'm thinking about what's next. So for the past 20 years, I've been looking at what's next and what software vendors are putting into the market, what organizations are doing, how they're succeeding and what it takes to succeed. Really, I've been doing a lot of research around that for a long time. Do you think that job in particular will be disrupted by the development of AI?

I can hear in my head a lot of people ask him when they listen to this episode. It's a very fascinating job becoming an analyst. And I think that there's a certain curiosity and curiosity always leads us to discovery. Do you think that this job in particular will be disrupted by the tools that are now accessible on the market?

Yeah, absolutely. I think it's happening already. We saw Gartner stock going down significantly a couple of months ago. I think whether it makes sense that should be the case or not is another matter entirely.

But I think in people's, in leaders' minds, that's something that they think won't be as necessary, that AI can actually do a lot of that for you. So that's, I do think it will be disrupted. You have been working in data and analytics for over 30 years now, and you began with pioneering continuous auditing and predictive models at AT&T and Bell Labs. How has the transition, in your own opinion, from maybe early rule-based systems and data mining to today's generative AI shifted the way businesses view and value their data?

Oh, my God. Where to begin on that? It's been a sea change in some ways. When I was starting off doing AI, I was using early machine learning algorithms, building statistical models, that sort of thing.

It was very highly specialized. And I thought of it as machine learning and predictive analytics. And now with the advent of generative AI, it's become much more consumerized. So everyone thinks that they can do it.

Everyone thinks that they understand AI. That was one of the reasons that I wrote the book, because there came a wave of people who were self-professed AI experts. And I know what I don't know. And I have a lot of expertise in it, but I know that I'm not the person who's developing the algorithms, but yet we had people out there who were saying, I'm the AI expert and writing books and saying, listen to me.

And I said, wow, there is a sea change. And certainly AI has become consumerized and a lot of people are using it for a lot of different reasons. They're trying to be more productive with it. That's great.

Whether they are or not, or just creating work slop is another discussion entirely. but I said, you've been looking at this for the past 20 years. You need to write a book about what it really takes to succeed with AI and what the data foundations and the trust foundations and the technology foundations and what all of that looks like so that organizations just don't think of AI as a set of tools that they can use, but that they really need to put together an enterprise, company-wide capabilities, the capabilities that are actually needed for AI to succeed, which isn't just about tools.

It's about a lot more than that. So that's why I wrote the book. Where do you think this sort of shift has taken place? Because you speak about the consumerization and it is very accessible.

It has become a sort of invitation for a B2C market to take over a b2b market i think that if we're looking at the necessity to maybe take those two apart and go back to the fundament that you're highlighting that is foundational to looking at a proper business view with data foundation a trust foundation do you think that we've gone too far to go back to the elements that you're highlighting as a real necessity. And again, that is data foundation, trust foundation. I don't think so, because I think that what's happening with the organizations is that they'll say, they'll start off, they'll say, oh, we need to do something with AI, even if they don't really understand what the business need is.

So they start off with the tools, but then they soon realize that they're going to reach what I call a value ceiling. And yeah, they'll get maybe some productivity gains, maybe some other good benefits but eventually they going to realize in order to really make use of AI effectively they going to need their company data And hey guess what If you don have your company data foundation in place, then your AI is just going to expose all the weaknesses that your organization has.

I say that AI amplifies the weaknesses and the strengths. So if you want to build a bot that's going to answer customer questions. Yeah, you maybe can do that without having a solid data foundation, but a much better bot because you can take your company's manuals or whatever and do that. But if you're actually trying to give a good customer experience, you need to know who the customer is that you're conversing with.

That means you need to get into or data to say, here's this customer. Oh, they're a loyalty customer. I need to be especially deferential to them or whatever. So soon enough, they're going to realize that they don't have their data foundations in place.

And then when, if they start to do a risk audit, which they may or may not do an inventory, what's happening and what they're trying to do and what type of risk that's going to incur on their company, they're going to realize, oh, maybe we need governance and we hadn't really thought about that. So it all starts to build on itself. That's not saying that companies, there's a lot of entrepreneurs out there that are making a lot of money building AI applications. I'm not talking about that necessarily.

I'm talking about companies that are trying to make use of AI. So I think that, and it was actually going to be my second question, one of the foundational questions that I wanted to ask you. So a central theme in your book is this ceiling, as you've mentioned, of generative AI. Now, maybe we can go back a step and look at what this ceiling, why do organizations really skip foundational data steps?

Because inevitably, it's hard to hit it, as you've mentioned. And I think that very much, again, commercialization of AI for an organization take the easy way out. There's an easy way where organizations make use of plug and play without the complexity of any form of data-driven orchestration. So what will really happen to organizations when they skip this foundational step?

And maybe do they really care if it's that important? Yeah, I think that they should care. I think that they're going to end up, if they don't put a foundation in place, they're going to end up doing things like incurring penalties because of certain laws that are out there. They could incur reputational risk.

We've seen that already with a number of organizations where they were using some sort generative AI tool and it didn't work. So the web telescope example with Google, when their stock went down a lot because their tools didn't work. What happened to Air Canada, for example, when I'm sure you're familiar with these stories. So I think that they have to put those foundations in place.

That said, there are a lot of vendors out there who are trying to make it easier for organizations to do that. Say a lot of organizations' data might be their CRM or ERP data, right? So their CRM and ERP vendors are now viewing their stack as a foundational stack, and then they're putting the AI tools on top of that. So if you trust your CRM vendor, or if you trust your third-party data provider that's providing demographic data, who's now integrated with your CRM data, that can certainly help with get your data foundation for some use cases in place more quickly and help ease your way into this.

But that's still your data foundation. So even if they're helping making it easier, you still have to take a phased approach and say, okay, my CRM data check, but what about the rest of my data foundation? And certainly organizations, what I see in my research at TDWI is that they're all trying to get a unified foundation in place that doesn't have to be a physical unification. It could be a logical unification, but they don't want to run into issues where they have 15 different definitions of a customer, especially as they're going to move into more complex AI, like agentic AI.

I do think that organizations, to the value ceiling point of your question, they realize once they, I've seen it multiple times, where they can't, they say, oh, I don't really have access to this data, or I don't understand this data, and I don't have metadata about this data. How am I going to move forward with this? They won't move forward. And certainly on the governance front, I've talked to a number of organizations, a lot of them, that with the rise of agentic AI and having a bunch of little agents running around trying to do a bunch of tasks, they're saying, whoa, let's put those guardrails in place because this has the potential to really blow up in our faces.

and then they understand that. It's one thing to put a generative AI app in place that's analyzing your call center notes. It's another thing to put a multi-agent system in place that's going to actually automate your supply chain or part of your supply chain, that there's a lot more risk involved in that. 100%.

So let's look at, I think that there's two aspects of that. There's definitely the risk that's involved in the implementation of multi-agentic systems, especially self-learning systems that can lead to even more dramatical impact that is unforeseen and obviously undesired. Now, there's one thing that you're talking about that I think has become quite a common observation that a lot of companies, and this is not unknown, they are really drowning in a lot of data silos, especially extremely large organizations.

And I think that possibly the idea of becoming AI ready can sometimes be a little bit overwhelming. So how do modern architectural patterns maybe like data lake houses or the data fabric how do they help organizations unify their data to feed a model effectively And how can an executive as well with the support of an executive team or executive minus one how can they also make sure that they do that with the speed that the market is pushing them into maybe into this whole agentic landscape it?

Yeah, first in terms of the data lake house or the data mesh, the data lake house is very helpful because it can help unify all different types of data, right? Organizations have historically just built their AI systems on structured data, the type of data that you think about, my billing data, some of my CRM data. But now with generative and agentic AI, organizations are really looking at new data types. I was mentioning the call center notes.

So they're looking at their unstructured data as well. And in fact, this unstructured, especially these text documents have become this big source of data for these types of AI. And of course, organizations, some organizations have been trying to analyze that for a while, but now the technology really seems to be there for that. So these data lake houses can actually help to unify both the structured and the unstructured data.

I think of it as a physical model. So some organizations go that way. The reality is you're not going to be, they think when they get started that they're going to be able to put everything in their lake house. The reality is you're not going to be able to consolidate everything in one lake house.

If you're big enough, that's not going to happen. So also they're looking at this notion of the data mesh, which is more of a logical unification that requires a semantic layer or some sort of semantics over all of your silos that make it look like there's one unified data platform underneath. So they're moving towards that model also. Both of those can be very helpful.

These things don't happen overnight. So executives need to realize, have some patience and educate yourselves about what these platforms are about and how they're going to be important for you because a lot of executives don't understand what's really involved and they just feel the pressure, do AI, get it done. And the people who are trying to deal with the data foundations, they're explaining. Of course, you could use a phased approach and that's fine.

What's the use case you're trying to solve for first and build what inventory or data? where's the data how am I going to put the right foundation in place what's the architecture look like that's going to help me with those first use cases that's also going to help me build towards the future that's how successful organizations look at it so they're not trying to boil the ocean obviously and it right out of the gate exactly yeah do you think that this has to be motioned from the top though because again if you go back to the complexity of large organizations each individual division has a certain set of data in a certain crm that might be separated from another erp system or a sharepoint whatever there is that they're utilizing and again that doesn't adhere as you've mentioned at the start with any sort of classification tag in there's maybe not any form of contextual graph or knowledge graph or working processes that are documented across the organization because again these are large evolving beasts so if there is this necessity for the foundational element of data to ensure proper practices who needs to who in your opinion needs emotionless.

I completely agree with you that I've seen time and time again that organizations that succeed, they have executive support from the top. I've seen that statistically in my research data. Also, if the CDO or whoever you're calling that person who's responsible for the data and AI, CDIO, CIO, whoever it is, they need to be part of the C-suite. That is what is going to definitely help organizations succeed.

As I said, I've seen that in the analysis I've done of all the assessments and surveys that we do at TDWI, that if that C-level person who's responsible for that is in the C-suite, then they have the ear of the CEO. I've seen at organizations some very successful organizations where the CEO is actually interested in the data foundation and the people who are doing the data and analytics. Once a quarter, they'll read out to the CEO and tell them what they're doing and what's happening.

And I think that's great if the CEO has the time and is interested at that level for that to happen because everyone says, yeah, we need executive support and like a top-down, bottom-up. That's great, but top-down is needed. Yeah. So there's one thing, Dr.

Halper, that I want to touch upon as well. Your book, it makes it clear that data governance and AI governance are two completely different disciplines. Why is traditional data governance insufficient for managing modern AI risks? and what specific controls are then required for AI governance.

So maybe just highlight, why are these two different things? I think that I tend to think of it as data governance for AI. There was data governance, now there's data governance for AI, and then there's AI governance, and then there's responsible AI in terms of the really important ethical and fairness and societal types of issues. So in terms of the data governance, they complement each other and they overlap, But data governance, you're looking at the typical accuracy, completeness, timeliness, consistency, all of those really important metrics that are really important for AI.

But with AI itself, you're talking about software, you're talking about models. So it yeah you need to have data governance so that you feeding trusted data into the models But the AI models themselves need to be governed And so they need to be versioned They need to be documented They need to be understood and they need to be tracked because AI models are going to degrade and drift over time. It's not like you can just build it once and let it run and hope for the best, right?

You have to always be tracking it. Those are different skill sets and different issues that you have to think about, but you want your AI to be explainable. That's already in GDPR, right? In terms of if you're making a decision that's going to impact someone like a credit card, yay or nay, or something about if you're going to hire them, it says there that the model needs to be explainable.

So someone's got to be able to explain And if a consumer asks, why was my credit card declined? They have to be able to explain it to you. All of that is different than what goes into data governance. They're both really important.

They complement and overlap each other. But as I said, I think that you have engineers, ops people who are dealing with the AI governance. And you have data engineers, data stewards. If your organization uses a data product model, then you're going to have business people should be involved, obviously, the whole way through because they're stakeholders in this and they need to be involved.

But they're different, at least to me, they're different but complementary. no I think that you've made a really good separation between those two distinguished elements in the different disciplines that are highlighted by the ownership of different people and different skill sets as well let me ask you something quite controversial I'm not saying that there is a yes or no answer or there would be an answer but if someone were to say to you what's more important what's more is data governance more important or is AI governance more important or maybe one can't be more important without the other, but what would you say?

I think that you can't have, you have to have data governance first. And because it's the old garbage in, garbage out. Yeah. I hate to go back to that, but these models, as we move into more complex AI, they're going to depend on data.

If your data governance isn't in place and it doesn't matter if your AI governance is great, you're just going to get garbage coming out of these models. you won't be able to trust it. And trust is more than data quality, but it's really foundational to AI. How do you encourage an organization that doesn't have the sort of necessary discipline when it comes to a consistent approach to governance?

How would you encourage them to breed that culture? I would encourage them by thinking about governance as an enabler. In my circle, some people talk about we shouldn't even call it governance because that makes people say, whoa, or be apathetic or whatever. But if governance is viewed as an enabler to help you really build the trust to your data and you put the processes in place and the tools in place for governance, then you can really start to trust your data and then you can trust your AI.

And a lot of times people come to governance because there's been an issue. And then they're more likely to say, yeah, we need the governance because we need to fix this data. And that's okay. But do you really want to get to a point where on a scale of one to 10, your data quality as one aspect of governance is a one, or you've made a big financial mistake because the quality was bad?

No, you want to do this in parallel with what you're building for AI to make sure that you can really trust and enable the AI. So that's how I would try to come to it. But the reality of the situation is a lot of times organizations make mistakes and then they have to go back and fix it. And that's how they come to governance as being really important.

Thank you. So one last question before we end the conversation, I think it's a really poignant one to end the conversation with. You started our conversation, you said that you've always been interested in what's next. And that's one thing that you enjoy about the field.

Are you optimistic about what's next in the AI field? The curious part of me is optimistic. Part of me has become more scared. That's the right word.

I think that the advent of agents has me concerned on a number of levels, one of which is what happens to people when I hear organizations talking about their agents as team members. That just doesn't. It sits well with me from an intellectual standpoint, but not from a personal sort of standpoint, if that makes sense. In my circles, we joke about, oh, there's going to be the billion-dollar company that's run by one person because there's just a whole bunch of agents that are doing all of the tasks that are needed.

And maybe, yeah, there's some people, a human in the loop, like overlooking these agents, but it's a concerning type of future when you hear about how agents are interacting with each other and trying to gain, one trying to gain control over the other, et cetera. There's still a lot of work to be done. I talk to people who are really in the weeds with a lot of this. And thankfully, they are trying to understand how agents are going to interact and putting the brakes on certain things.

But some of it certainly has me nervous. If organizations don't actually think about it the right way and put the right controls in place, it could completely blow up. So we'll see if that happens. thank you no it's not to make it if it's i think it's important you've been in the field for a long time and it's good to take a retrospective yeah a retrospective look at the development and what could possibly be next and also stands towards that there's no right or wrong thank you very much for the conversation dr halper i think that it's going to be beneficial and i hope that everyone that's been with us in this conversation has enjoyed it.

If you'd like to find out more, then click on the link and enjoy the conversation. Share it with your friends. Thank you. Thank you so much for having me.

I enjoyed the conversation. Thank you.