How to Use AI in Business Without Falling for the Hype

Every business leader right now is feeling some version of the same pressure. We need to do something with AI, which sounds strategic until you realize something is not a plan. It is a panic response wearing a blazer. But here's the expensive problem. Using AI tools is not the same as creating business value with AI. And that's what we're talking about today on Marc Haine Live. Welcome to this episode of Marc Haine Live. Of course, this is where small business owners and entrepreneurs pick up some core skill sets to help them create the show-stopping, jaw-dropping experience that our customers and our employees deserve. I am your host, Customer and Employee Experience Strategist Marc Haine. I'm so glad that you're here with me today because my guest is Dr. Fern Halper. And we are going to cut through the AI hype to explore what businesses really need to use AI responsibly, strategically, and effectively. My one request is to please make sure that you're subscribed to this podcast. By doing so, you'll get notifications each week when I bring you fresh new content. And of course, it helps my ratings. on the YouTube God machine. And I could really use your help. And after all, it's free. So, why wouldn't you want to do it? There is a moment happening in businesses right now that feels a little like the early days of social media. Do you remember when every company suddenly needed to have a Facebook page? Not a strategy, not a content plan, not a customer engagement model. Just a page. And for many organizations, that became— that page really became the digital equivalent of a dusty brochure rack in the corner of a hotel lobby. You remember those? They used to have those. Now with AI, we're seeing the same kinds of patterns. Business leaders know that AI matters or could matter. They know that their competitors are exploring it. They know their employees are already using tools behind the scenes. And they know Doing nothing is probably not a winning move. And that brings me to our Question of the Day. So, where in your business are you experimenting with AI without a clear strategy? And what could that cost you if you keep moving forward without the right foundations. I'd love to hear your experiences. Why don't you go ahead and share a link to this episode on your favorite social media platform? And in the comments, make sure you put your comments down about your experiences maybe with AI and hashtag it #ExperienceLeadership and be part of this conversation. I look forward to seeing what those comments look like. And of course, if you like, you can comment here right on YouTube. You can put your comments down here Again, #ExperienceLeadership. Would love to hear what your thoughts are. As I mentioned, today's guest is exactly the kind of person who we need in the room when the conversation is all about AI and it starts getting louder than it is useful. Dr. Fern Halper is the founder of AI Foundations Group and the VP and Senior Research Director for TDWI, where she has spent more than 2 decades researching data management, analytics, governance, and artificial intelligence. In other words, Fern has been studying the engine room of AI long before most of us were asking a chatbot to write that birthday poem for our accountant. You remember the one? Fern has written several well-known technology books including Cloud Computing for Dummies, Big Data for Dummies, and her latest book is Data Makes the World Go Round: The Data, Tech, and Trust Behind AI Success. Fern, welcome to the show! It's so great to have you.

I'm happy to be here. Thank you, Marc.

So, maybe you could tell us a little bit about who your clients are and how you serve them right now.

Well, I have a couple different sets of clients, but a lot of them are executives who want to understand more about AI, about data and AI and the data foundations, what makes AI tick, how to be successful with it, and so on. So, a lot of discussions around that, where the market is, what they need to do, how they're comparing to their competitors.

Yeah, and so it seems to me that a lot of business leaders now are feeling that pressure to do something with AI. They don't know what it is necessarily. What problem does that pressure create when leaders move kind of too fast into kind of like, oh, we have to do something because everybody else is doing something? What's the downside if they don't have a clear path?

Oh God! Well, the downside—

Oh, by the way, I forgot to remind you I only ask little questions.

Yeah, the downside is that they're going to lose a lot of money and they won't be competitive. They risk reputational risk, they risk business risk. One of the biggest misconceptions, I think, in the market today is that AI adoption equals AI readiness. So everyone's saying we have to do something, and that means that people are trying to implement ChatGPT, Copilot, Gemini, Claude. I was at a conference where people were debating the merits of the that the other day, but my research at TDWI shows that close to 90% of organizations are either using or experimenting with, or they're planning to use generative or agentic AI. But I think that using AI is not the same as becoming an AI-driven organization. And so in my book, I describe this thing called the value ceiling. And if I give employees ChatGPT, They'll likely become more productive. You know, they'll write faster, they'll summarize faster, they'll generate content faster. And that can be valuable, you know, but I think eventually the value plateaus. And so the real value, I think, comes when AI is part of products and operations and customer experiences and decision-making.

Right.

And that is going to require not just using a tool, but building trusted data foundations, governance, leadership skills, and organizational readiness. I mean, interestingly, one of the surprising findings in some of my recent research at TDWI is that even though a large majority of organizations are using off-the-shelf AI tools, like I was just talking about, and many are building custom assistants, only about 40% of them are measuring consistent business value.

That's interesting.

Yeah.

So, we'll get deeper into that. I'd like to roll back just very quickly for those people who are tuning in brand new to AI, not really sure of some of the lingo and stuff. Can we please define generative AI and agentic AI for the audience?

Yeah, sure. So, when I think about AI, well, the easy way to think about it is like 3 different types of AI. There was the traditional AI. And me as an old person, I was involved in traditional AI way back like 20, over 20 years ago at Bell Labs in the '90s, using tools like you may have heard of machine learning, neural networks, those kinds of tools that can actually find patterns in data. Well, you know, a couple of years ago, there was this big AI boom when large language models came out. And those were generative, that was generative AI, which was all about, you know, asking a prompt and then generating output From that prompt. And the technology behind it isn't that worth going into, except I think everyone's pretty familiar with generative AI if they use ChatGPT or Claude or any of those tools. You put in a prompt, you get out an answer, you interact with it. It's a lot of fun. Then there's agentic AI, which is where all of this is heading now. And agentic AI is about agents. Think of little agents that are actually going about doing tasks, and they're autonomous and they can reason.

Right.

And they can perceive and they can act. And that's where things can get pretty scary.

Yeah, because AI, you know, agentic AI is all about it doing tasks on your behalf as well, right?

Yes.

So, getting in and having it do kind of some of the outreach or customer service or whatever. I did see a very clever AI adoption from O2 in England where they use agentic AI to do spam busting. They get a spam call. They'll use the agentic AI to actually take up the spammer's time so that they're not out spamming other people and keep people on phone, which I think is hilarious.

Okay, that's a good idea. That's a good idea.

But if you're ever interested, you just Google the spam bot on YouTube. It's hilarious. created this little old lady who just drives spammers absolutely crazy. And I think it's brilliant. But when you look at organizations adopting AI today, where are they most likely confusing activity with strategic progress?

I think that there's a lot of places that they're confusing activity with strategic progress. To me, if you have a strategy, And you're going to use AI, it's because AI is going to fit a business need or solve a business problem. But what a lot of organizations are doing is that they're implementing AI sort of everywhere. It's called shadow AI. So everyone in the organization has sort of their own AI. They're just sort of off doing their own thing and there's no sort of strategy behind it. And that means there's all sorts of— Duplicate efforts. There's no control over it. And that's where an organization can get into trouble, I think. I've had some interesting conversations with small organizations just about what they should and should not be doing with AI and private data, for example.

So, could you delve into some of what those problems look like? Like, how does somebody or an organization that is kind of doing this kind of siloed approach to AI adoption, tripping over and falling inside themselves and creating bigger issues than what AI is actually solving for them.

Now, here's an example. So, the CEO said, let's do something with AI, right? The refrain, let's do something with AI. Give me your ideas. Let's get started. Just do something. I have a number of examples. So, So, people went out, they started to do things. In some cases, they might have been using some company data and they didn't actually— the company data wasn't in good shape, right? So, maybe they built a chatbot and that's fine. You could have a chatbot that answers questions based on your, I don't know, all of your FAQs or whatever. But you really want the chatbot to answer questions to know who they're talking to. And if it's a customer, you want them to be able to go back and look at that data about that customer. Are they a loyalty customer? Have they not, you know, have they spent a lot, a little, you know, what was the last thing that they bought, that sort of thing. But so organizations would go out and build and not get the real value that they needed to get, or worse, that they would be answering the same question differently, you know, for different customers, or they didn't even know who the customer was that they were talking to because they had multiple views, you know, of their customer. In another instance, an organization said, yeah, look, this agentic stuff, this sounds really cool. Let's just go out and everyone create their own agents, right? So, I guess Copilot has an agent builder. You can use ChatGPT now, right, to create an agent. for yourself. And all of a sudden, there were 20,000 agents running around this company. It was a larger company. But then the company realized, we have no control over these agents, right? There's no guardrails on these things. Who knows what data they're actually accessing? If it's private data, do they have the ability to even access it? Should they have been? But we didn't put those access controls in place in the first place. They're not thinking about their agents at a systems level. I don't want to get too technical. But at a systems level, and so they would have duplicate agents running around sort of doing the same thing. That's going to cost money. It's going to make a mess. So, I mean, those are 2 examples of what happens under the scenario that you just described.

Yeah, and I can't imagine having agents kind of doing some of, let's say, doing the outreach or the marketing or whatever and having so many different people's kind of individual agentic agents doing the work. But there's no cohesiveness around brand voice or kind of what the true north is of the organization. And who knows what these agents are doing and saying to our customers.

Yeah, and saying to your customers, that's a good one because I was listening. I was talking to someone the other day who said that they got internal email from their company saying, this is your so-and-so agent, like somehow it was their buddy. And they were shocked because they had no idea that part of their process in the company now was going to be actually interacting with this agent. So, the company did not set that up properly.

Yeah. Are there any symptoms that show up inside the organization when AI adoption is being driven by all the hype instead of by a real strategic I mean, there can be multiple, you know, you can have sort of multiple proof of concepts and whatnot.

But I mean, it just becomes, IT gets upset, the business gets upset, there's no control. And so then what happens is, you know, that organizations realize we need to put control, some controls in place, and some of them do that better than others. I mean, at a minimum, Organizations should have some sort of entry point so that they know what tools their employees are using, even if they didn't decide that this was— these were the tools that they, that they should have. At least they should have an inventory of them so they're not paying duplicate license fees, you know, for example.

Right.

They can get a better deal. So, things just become very messy and you start to you know, see a lot of problems in your systems. And, you know, a lot of organizations— well, there was an article today, right, that I don't know if Uber— did you see that article that Uber ran through its budget for AI in 4 months, its yearly budget?

Yeah.

So, then they had to, you know, sort of roll back things. So, it's untamed. I don't know about I don't want to say anything bad about Uber because I don't really know what was going on there. But that was what the article was talking about. So, I think that's happening in a lot of companies. A lot of people I talk to say, we're just running through the budget so quickly. I'm using up all my tokens and I can't do anything. And well, sort of there's no plan.

Yes. So, where do leaders, in your experience based on your research, where do leaders kind of underestimate or even overestimate the things like the risk around privacy, security, governance, around trust building? Like, what are you seeing?

I see that governance, you know, there's all different types of governance. And people don't like to talk about governance because some people view it as a headache. And when the leaders came out and said, let's use AI, and then they hit this value ceiling, right? Now, they needed they realized that they needed— that AI isn't just a tool. I guess that's one of my messages. Like, AI isn't just a tool, you know, it's a set of organizational capabilities. And it doesn't matter if you're a small organization or a large organization, because you can put this, you know, together with varying degrees of sophistication or whatever, you know.

Right.

But at a minimum, you know, you have to be organizationally ready. You have to have your data ready. You have to have your governance ready. You have to be able to operationalize, you know, what you're trying to put into production. And, you know, you also need the skills and the literacy. So, those 5 things, I think leaders underestimate. And when it comes to governance, I think that they're thinking about security and privacy. Yes, they think about that, but they're not thinking that governance can actually be an enabler. You know, if I— put the right controls in place, not after the fact, you know, but when I'm sort of putting together my AI and what data it's going to access and everything, and I'm thinking about that as part of the system that I'm building, then I can actually trust the data. And then I'm not going to, you know, fall into, you know, issues like what we've seen in— I don't know if you ever heard of the Incidents Database, you know, where they talk about— various AI incidents or, you know, some of the very popular problems that organizations ran into because they weren't thinking about governance in the first place. And of course, you know, organizations, I've heard that, you know, some of the regulators are going to start coming down harder on organizations. And if they thought that they could get away with things, they couldn't. I mean, even with traditional AI, you think about, a hiring model, right?

Right.

And, you know, there was that famous story about, you know, one of the big hyperscalers who, you know, had a certain model in place for hiring engineers. But historically, engineers have been all male, right? And now females are, you know, are coming into the picture. Well, they have been for a while. And those females weren't given the same due diligence as the males because the model had trained on that. Well, someone needed to think about that.

Yeah, it's interesting because I don't think that people understand that AI is also feeding on our own biases and our own processes. And even something like policy manuals, the way that some of them are phrased when we were human beings doing it, when it gets interpreted by AI, could have different definitions and different biases bled into that then taint future communications.

Yes. Yeah. And bias is like a big deal. I talk a lot about that in my book. And, you know, the reality also is that the regulations GDPR has a piece of it. And you may not think about it if you read it. But it says, you know, that the model has to be if you say, don't give someone a credit card. You have to be able to explain why you didn't give someone a credit card. And, you know, if— and that includes undetected bias. You know, all of a sudden you're going to realize that I didn't give someone a credit card. Well, it was because of the way that my model, you know, was being trained. Right? That happened with the Apple Card where women were— yeah, you know the story.

Mm-hmm.

It's time and time again, bias rears its ugly head.

Yeah, I'd love to get into some actionable steps that listeners can take to become more strategic with their AI. And we'll get to that right after this. When the spotlight shines on your business, are customers applauding or yawning? In other words, how is your business performing? Make your business a star with a new book, Lights! Camera! Action! Business operational excellence through the lens of live theater by Marc Haine. Marc uses his business and acting experience to help you see your business like a live show so you can create a performance your customers will never forget. Buy Lights!

Camera! Action!

today at your favorite online retailer or directly at marchaine.com. Welcome back! I am speaking with Dr. Fern Harper. And the conversation around AI, I think, is absolutely thrilling. But Fern, you yourself mentioned your book just before the break. Could you tell us a little bit about your book?

Sure. So, the book, as you mentioned, is called Data Makes the World Go Round: The Data, Tech, and Trust Behind AI Success. And I wrote it because what I saw happening was that with the advent of consumerized AI, the off-the-shelf tools, organizations weren't thinking enough about what actually goes into being successful with AI, those 5 areas that I talked about before. And so this is basically a culmination of many, many years of looking at what organizations are doing right and doing wrong. And it sort of lays it all out about what organizations should be doing, how they should be thinking about things, how they should be measuring, you know, what the outputs of their AI efforts Or I have sprinkled in through there a lot of expert insights and various stories of companies that did it right.

Yeah, it's really interesting when you mentioned about the companies turning around going, oh, we have AI. Let's find out what we could use AI for. That's kind of sort of like handing people a hammer and go, here, you have a hammer. Go and build something.

Right.

And so, how should a business leader begin identifying the right business problems for AI to help them solve? Like, how do they, if they've never really explored it outside of, help me make my email better, how do they define what AI can help them with?

Yeah, that's a really good question. And I guess a lot of leaders come to the customer side of things and look at the problems that they're having. with customers or selling products, you know, because those are— or services, because those are some big areas, obviously. And, you know, they say, well, what can I actually change about this? Or, you know, how can I make the customer experience better? How do I prevent customer churn? You know, those sorts of things. A lot of organizations come to it, you know, some organizations come to it because they've had a problem, you know.

Right.

But if you haven't, had a problem that you know of, then you want to ask, well, what is the biggest business need? Yeah, I guess it could be something like, I want to grow my revenue. Okay, well, what does that mean, right? That's pretty vague anyway. So then you have to drill into what does growing your revenue mean? What is your revenue? What are the revenue sources now? What could be improved? Could it be improved by using AI, you know, thinking about use cases there. Because, I mean, come on, like since the '90s, we've been using AI to do things like customer churn prediction, market basket analysis, you know, next best offer, right? It's not like this is a new concept at all. It's just—

Yeah. Well, I think to your point, I think what made— what's different now from the '90s to now is it's kind of sort of making AI accessible to even even people at lower levels. Like the average mom-and-pop wouldn't have, they don't have the database and the expertise, let's say, like Target would have to be able to put out stuff and be able to cater to the needs or define even the needs of who their customers are and how they can help. AI now kind of brings that and makes it more accessible, doesn't it?

Yes, yeah, for sure, for sure. And so, you could ask the question, what am I doing? What could I be doing better? Or Yeah. There's that sort of way. I have a chapter in the book where there's a quote by a guy named Donald Farmer, who is a good AI strategist. And, you know, he talks about making, you know, how am I looking at the business problem? Am I looking at the business problem sort of in the old world or like in this new world of AI? And how can AI, you know, actually help me improve sort of that way? And you can build applications that maybe you couldn't have built before, right? And that provides a value-added service to your customers. There's a lot of organizations that are doing that sort of thing. So, you know, there's all different ways you can go about it, but it always starts with, you know, what's the need and what's the business problem that I'm trying to solve?

It's interesting because I found, you know, when I do sessions with people, I tend to do a lot of sessions around kind of AI 101. where there's an organization with a group of people who are like, I don't know what I can use. I don't know what I can't use. I don't know if it's cheating.

Right.

We have all these different kinds of inclinations within different people. And so, this idea of like data safety and that sort of thing, I have a very simplistic definition that if you wouldn't post it on LinkedIn, you shouldn't post it in AI. But that's super, super simplistic. When we think about this idea of kind of human in the loop to make sure that we're not handing everything off to AI but we're really using it as a tool to help us and maintain our power in whatever we're doing? What are some of the best practices you would suggest?

Okay, yeah, I want to say something about human in the loop. So, in some ways to me, human in the loop has become like a compliance checkbox, right? Are we sending off these autonomous AI agents or bots or whatever to do something and nobody's overseeing it, check, we have human in the loop. But I think that people haven't really thought past that. I mean, because there's different types of human in the loop. There's sort of the good type of human in the loop. I would call it a good type at least because it keeps people really engaged where the human in the loop, think about, I don't know, like healthcare research, right? And the human is definitely in the loop. They're interacting with the AI, they're coming up with discoveries, and then there's human in the loop where the human is really just a checkbox and saying, yes, I, you know, yes, send me your marketing copy and I'll read it over. And they'll make some changes and say, okay. Or it's a call center agent. They have automated call center agents, but if there's an alert that says someone's really upset or they have a question that the agent, the automated agent can't answer, it goes to the call center agent and then the call center agent gets involved. But, you know, there's this thing called automation bias where people start to actually just, you know, as the AI is pumping out all this stuff, they're not necessarily using their critical thinking abilities. Let's put it that way. Because in some ways, you know, their brain has become inured to what's actually coming out of these AI and these AI systems. And they say, yeah, that's good. Yeah, that's good. And then think about what happens when agents 5 to 10 years from now do actually sort of take over and, you know, there's a lot of autonomous agents dealing with your supply chain and, or whatever, you know, they're, they're doing the volume of what those agents are producing. Can, can people really keep up with it? You know, so I think that human in the loop makes sense. Like, there should be a human in the loop, but I think that it needs to be a human in the loop the right way, doing the right kind of work, if that makes sense.

Well, and to your point, like I set up an agentic AI for my business to kind of be an intermediary when somebody's just curious about something. And so, they type stuff in. And to your point, I have no means of being able to grab that data and grab that person who was curious and be able to figure out how to serve them next. And so, my next iteration is Oh, I need to figure out like I need to receive an email from my agent saying that I talked to Donna and here's Donna's email and here's what she was asking. And being able to then bring me in because ultimately, I have somebody coming to my website and it becomes a lost lead because they're going in. They're typing some information. The agent is answering their questions and then they disappear and I never hear from them again. So, I do know that from my own experience being kind of this Luddite kind of stepping into it and experimenting and getting curious with it, that I'm making mistakes as I go.

Right, right. And there's been a lot of research on automation bias certainly that shows what happens when humans are dealing with machines. And we're going to make mistakes. So, I guess it's a matter of determining where it makes sense to make mistakes. And if those agents You know, you really have to be thinking about what those agents are doing, map them out, you know, look at every point where they could actually be making a mistake, where they're gathering data, you know, outputting data, you know, the input controls, all the output controls, all of, all of that. I think, you know, I think humans, humans are in the loop now. I think the issue is whether they are going to remain meaningfully engaged in their work? And I think that that's sort of the next question. Because recently I just argued that we're asking the wrong questions about human-in-the-loop system. I think that there's a difference between participating in a decision and simply monitoring a decision, because one is going to develop judgment and the other monitors for exceptions. And I think As I said, I worry about cognitive disengagement. Think about a junior analyst, right? Historically, that analyst learned by exploring the data, and they formed a hypothesis, and they discovered insights. But now, what if they're just validating the AI-generated insights? They're not going to develop the same expertise, will they? I don't think so.

And ultimately, you'll create more distance within the humanity of our businesses. We create more distance between our stakeholders and ourselves and what their needs are and what our needs are because we're not privy. We're not helping make those decisions anymore. We're not privy to the problem because if AI is just solving the problem, whether that's actually solving a problem or saying, piss off. We don't want you as a customer anymore. And we don't know anything about it, right?

That's sort of like a bad sci-fi movie, right?

Right, isn't it? Yeah.

Yeah, we've seen it. We've seen this movie.

How can leaders educate employees so AI becomes a useful tool instead of kind of a source of fear, confusion, and uncertainty?

Yeah, I think a lot of it is, well, there's a couple different aspects to that. I think one aspect is change management. And actually, You know, there's ways— some people are much better at change management than others. But, you know, there's different techniques for change management that then you have to adopt. You know, there's Kotter's 8 Steps. There's the Procy models. There are lots of different models that have some good ideas that you can take and then use, but always remembering that people are people and that, you know, they carry their baggage with them. And, you know, maybe they're upset for a reason you don't even understand. But anyway, I've seen organizations use change management Well, you know, they also educate their workforce. So, even if you're a small group or a large group, I think that literacy is really, really important. And what I've seen, like, even mid-sized companies do is that, you know, they'll develop a literacy program and they'll have— everyone has to take like 5 hours of some basic literacy about what to do with sensitive data, et cetera. And then other personas in the organization, will take, you know, other, um, other classes and, you know, it goes up and up. Um, and don't— you don't want to test the people because that's, that's not a nice thing to do. But some organizations do test people also, but I'm not a big believer, um, in that. But, you know, people get certificates, they get happy, and, you know, some organizations with change management and this literacy, they actually, in terms of trying to get people on board, like say the operations people aren't really on board. Well, they actually market it. They create a whole marketing program around what AI can do for you. And they have nice videos and people get excited and they have people talking to other people. And a lot of it is about relationships. You know, really that your organization trusts you and trusts that you're not going to get rid of them. It's so interesting, back to the human in the loop thing. I was doing some research on human in the loop at TDWI. And so, these were with, you know, data and AI and, you know, more technical people. But I think a lot of it is still the same. I think that they haven't really Most of them haven't become a human in the loop to some degree. Those who have like being a human in the loop, I think, now because they're either— they're doing it the right way because they're interacting. But for those that are either monitoring, validating, reviewing, or whatever, I think that they're okay with it for now because they know that they're needed. And so, they're not necessarily going anywhere. Some people say it frees up time for more strategic work, you know, et cetera. But as I said, you know, then when you ask what's going to happen 5 to 10 years from now, they're going to say that agents take over and they haven't really thought about what that means for them in terms of human in the loop. So, I'm not trying to raise flags or anything. I just think it's something that CTOs and CIOs need to think about and other, you know, the people who are behind the scenes sort of dealing with this because right now it's just like this box that means nothing. I have a human in the loop, you know, so come up with the categories of the types of, you know, areas that humans need to be in the loop and look at what their responsibilities are and, you know, make it something, not just this amorphous human in the loop.

Yeah, yeah, that makes perfect sense. Hey, could you let everybody know how they can get in touch with you if this is resonating with them and they they'd like to work with you?

Oh, sure. The best thing to do, you can check me out on LinkedIn and contact me through LinkedIn. It's FB Halper, or just email me at the AI Foundations Group. So it's F Halper, spelled like you see it there, at aifoundationsgroup.com.

Terrific. I'll make sure it's all in the show notes. Are there any foundations needed to be in place before we can create real value with AI? Is there like what, when we tell people about doing stuff foundationally before they build? Obviously, before you build a house, you build a foundation. What about with AI? Somebody who's teasing into it now, what does that foundation look like for them?

To me, it sort of goes back to those 5 principles that I was talking about. Before, I think even small organizations need to be thinking a lot about their data foundations because they're only going to get so far. And the reality is that the new types of AI that we've been talking about, a lot of organizations are using like their text data, their documents, their call center notes, their customer interaction histories. You know, that stuff is amazing to get insights and to do things from, but you're not going to be able to get the insights or get the trusted insights until you have a data foundation in place. And it doesn't have to be complicated, but you're going to need some sort of data foundation. That's the first thing. And then, as I was saying before, along with that is enabling the governance, like putting the guardrails in so that people can't do things and make make a mess, you know, basically.

Yeah.

Those are 2 of sort of the technical areas that I think are really important. And also operationalizing, you know, once you build an application, if it's a chatbot or whatever, you're going to have to put it— it's going to keep— it's going to have to keep running. So, it's not just a matter of build it, it's you have to build it and then you have to put— when you put it into production, you're going to have to be monitoring it and making sure that everything's okay with it because, you know, Everything's going to degrade over time. And so someone has to be— that's a good job for a human in the loop. You know, they have to be sort of looking at all of that. So those are some of the foundations. And really important, if you want to grow your AI, is to measure the outcomes that you're getting from your AI. When you think about the business needed and the problem, and you start thinking, what data do I have to solve this problem? How am I going to build it? How am I going to govern it? Then you also have to think about, how am I going to measure it? So, you really want to think about the measurement because if you can show positive value and not just how many people are using it, I mean, obviously, if you have a subscription service, maybe that's a good thing that a lot of people are using it. But come up with metrics that mean something because what I've seen over and over and over again is that this is a virtuous circle. As people start to see that AI is helping and they can prove it because they have measures, then it just keeps going. growing and growing. It's like one of those ragweeds going down a Texas town, just growing. But in a good way.

The tumbleweeds. The tumbleweeds go rolling.

Yeah.

You mentioned data foundations. Now, I've seen a lot of chatter up on YouTube about how to build agents. And they're doing exactly what you just suggested is an incomplete way of doing it, which is take all the questions like if you have a database of kind of all the questions customers are asking or employees are asking, create an FAQ, upload it, get the AI just to tap into those things gives us really a limited data set to play in. Is there anything else we need to be looking at to help us build that foundation, that data foundation you were talking about?

Well, not to get too technical. But yeah, it's not just about the data. It's also about, as I was saying, understanding the data, being able to access it, and so on. So, what A really hot topic now in the data and AI world is semantic layers. So, actually being able to build business definitions of your data so you actually know what you're accessing. So, when people had BI tools, semantic layers have been around forever, right?

Mm-hmm.

You think of your BI tools and you needed to have some sort of business definition of every piece of the data that you were slicing and dicing in those BI tools, but, you know, just sort of take that and expand it. I mean, that's what's going to give your AI context, because if your AI doesn't have context, it may see revenue, but it doesn't know what revenue actually means, or it may see customer, but it doesn't know what customer means for your business. So, you need to be able to explain. It needs to have that context so it can actually act properly and not hallucinate. If I think that that makes sense. So, there's a lot to the data foundation, but, you know, and it depends how big your company is. I mean, you may not need a semantic layer if, you know, you only have one source of data for your customers and, you know, you have some metadata about your customers and that's enough. It's not like you have 15 different siloed systems that have the same customers that are called something different, right? It really depends on what your situation is.

But again, it really kind of creates that red flag that as you scale, you have to understand that you've created these elements. And that as you scale, or as you develop, and as you evolve, or your business evolves, we need to be going back to the basics of going back to the AI. What did we create with the AI? How do we keep it up to date with what we're doing so it's not becoming an anchor to our business?

Right, right, exactly. And it's, yeah, It's something that once you commit, you're in.

Yeah, yeah. I'd like to get into some pitfalls to watch out for and some cautionaries as people want to tap into this. Some things we should be aware of. And we'll get to that right after this. With the Pro Speaker Presentation Speech Timer app. No more checking your watch or calling for time. The Pro Speaker Presentation Speech Timer app keeps you on track with easy-to-see timers, even changing color for visual prompts during your speech. And you can set audio cues to practice or set it to vibrate so you don't even have to look. Be the pro you know you are. Download the app at speakerpresentationtimer.com.

Off with data quality. And of course, there's data governance and then there's AI governance, and they're different, but they overlap a lot. The data governance, we're concerned about the completeness, the timeliness, the accuracy, the integrity, the trustworthiness, the bias, you know, all of those things about the data that's going into the AI. And then with AI governance, you know, we're actually concerned about the AI system itself. So we're thinking about the software sort of there and, you know, what are the outputs look like and who's accountable and, you know, those sorts of questions. Of course, there's who's accountable on the data side too. But this guy, you know, what he did was that there was a problem, I guess, at company. And there was a problem with the quality of the data. It was pretty bad. And so, what he did was he brought a couple of people into the room and he just said, if everything looked good, what would that look like? And then, people just started sort of shouting out, here's what this would look like. Here's what— we wouldn't have this problem. We wouldn't have that problem. So, that actually helped them form the basis of their data governance program in this case. And they were able to put together dashboards to track You know, what was happening with their data. On the AI side, what I've seen organizations do is look at free publications, basically. So, I'm thinking small here now, not thinking like you're going to build some huge AI governance program. But, you know, they'll look at NIST, the National Institute of Standards. They have really good AI They have some good AI foundational things to look at. The OECD, there's a lot of different organizations out there that have, when they started thinking about AI governance and responsible AI, they put, we have our 7 different areas that you need to be thinking of safety, privacy, those sorts of things, and they describe them. So, I think that those are really good areas to start with. And then, What I've seen a lot of organizations successfully do is sort of combine different models together that make sense for their organization. And then, you know, they have their own. They'll look at all the pillars that these third parties have and then put together sort of what are our pillars that are really important to us and our business. So, that's a good way to get started.

How does one stay on top of kind of all the changes that are happening? Because I mean, this is still very much kind of infancy, this mass adoption of large language models. And it just seems like everybody's just leapfrogging over each other. The politicians seem to be doing a lot of infighting about what needs to be metered, what needs to be watched, what needs to be policed, and so on. How does one stay abreast of all these things? Because just it's It's an absolute whirlwind.

Yeah, I agree. It's a complete absolute whirlwind. And now, yeah, in the US, it seems like we're sort of maybe taking a step backwards. That's unclear to me at this point. But then the states are trying to take a step forward, which is only going to cause more confusion because, you know, then there's 50 states each with their own, you know, requirements. But if you're doing business in those states, you have to be worried. And of course, you know, the EU has their, their own regulations, etc. I've seen some good— it's hard to keep abreast. I don't think one, unless you have one person who's keeping abreast of it, you know, that's, that's difficult. I've seen, you know, there are, there are legal sites, you would just have to Google, you know, legal compliance AI, and that they actually lay out all of the regulations and what's coming up and what people need to be thinking about. There are some that do that for certain industries, you know, in financial services or whatever. You know, if you're a larger company and you can afford, you know, to buy software tools, there are software tools now also that actually help you monitor governance and tell you whether you're in compliance, right? Because they're tracking, and in their software, they have all of the compliance, all of the regulations, and, you know, they know the questions to ask you about whether you're in compliance. So, You know, but those are bigger ticket items, obviously, you know, as part of your AI governance strategy. But interestingly, in a recent TDWI survey, the biggest investment area was for 2026 on the technology side was AI governance tools. So people are—

So we're onto something.

Yeah. People are looking at the tools that are out there and trying to determine, you know, which ones might be good for them. I mean, there are even tools that will sort of take your data and tell you whether, you know, they're gen— whether the data is gender biased or age biased. You know, they'll have like 26 different metrics for bias.

Yeah.

So if that's important, you know, you can look at those, those types of tools. As an industry analyst, I try to keep abreast, you know, with a lot of the tools that are, that are out there. But, you know, obviously some are less expensive, some are more expensive.

Yeah. Are there any other pitfalls that we should be aware of as people want to start climbing into it? Like I've been endorsing curiosity as a main virtue when it comes down to AI, to try things, to look at problems, to see if AI can help solve problems as a way to be able to experiment and get our toes in. But as people do that, or as people kind of step into it, any other pitfalls they should be aware of?

Well, I do think that AI is going to expose whatever strengths and weaknesses are in your organization. So, if your organization is weak in its data or people don't want to change, it's going to expose that very quickly. If your organization is doing good by governance and the leaders are on board and everyone's on board, you're It'll expose that too. So, it's just really going to expose whatever you're trying to do because it's sort of, it isn't just a tool, you know, it's an enterprise or it's an organizational capability. And people need to think about it as a capability, not as a tool. But I think all too often people think of it as a tool and that's how they get into trouble because they don't put all the surrounding scaffolding that they need to hold the tool up.

Right.

You know, the tool is at the top of the stack. So, you know, there's all these other things you needed to do before you even get there. If you actually want to do more than, you know, just using it for the types of use cases that we've been talking about. If you actually want to go in and use, you know, your company data is your— that's your prized possession. Yeah. And it's what makes your company your company. So, if you don't use it, then you're you're going to lose out on the value that you could be getting from AI. But in order to use it, you have to make sure it's in good enough shape to use.

Well, and some people would hear your comment about it'll expose whatever weaknesses your business has or your data has. And some people go, oh, it's going to expose. To me, that's thrilling to know that we can actually do something And there will be some way to be able to meter where our weaknesses are so that we can get ourselves better. I think it's such an exciting time and exciting opportunity. But I do know that some people would look at that and go, oh well, no, it's going to expose all how bad we really are with our data management.

Yeah. But if you want to be competitive, you're going to have to deal with it. And that's what I say to leaders. You're going to have You know, educate yourselves and understand what's involved with this. So, it's not just a tool, it's a capability.

I love that. Fern, this has been fantastic. What a great conversation. Could you remind everybody one more time how they can get a hold of you?

Sure. Either connect with me on LinkedIn, so that's F as in Fern, B as in Boy, Halper, or send me an email, fhalper@aifoundationsgroup.com. There's a contact form on my AI Foundations Group website. And certainly, take a look at tdwi.org. tdwi.org. There's a lot of webinars. There's a lot of research there. TDWI provides a lot of training. So, if you want to get yourself educated, TDWI is the way to go with that.

I love that. As we wrap up, do you have any last thoughts about what we've been talking about today?

Well, you know, I hope I didn't come across as a naysayer because I'm really excited. Not at all. I'm really excited about AI. I just think that, you know, there's a right way, there's a right path, you know, forward. And, you know, experimentation with AI is not the same, or proof of concept, which is what most organizations are doing, is not the same as, you know, a production system that's actually providing a lot of value. So, I think that organizations have to think about getting past just the experimentation phase. And if they see something that is worth doing, then take it to the next step. Yeah.

Bern, thank you so much. I appreciate you being here, sharing your knowledge, your expertise, and of course, your new book. Congratulations on that. It looks like a brilliant piece of work. Thank you so much for doing this show.

Oh, thank you so much for having me. I really enjoyed it. Thank you.

If you have any questions about today's episode, or you'd like to have a complimentary 30-minute brainstorming session with you and your team, feel free to book yourself on my online calendar. The link is below in the show notes. It's the one that's marked meetwith.marchaine.com. As always, I am at your service. And if you haven't done so yet, please make sure you're subscribed to this podcast. It'll give you notifications whenever I bring you fresh new content and experts just like Fern to help us make our businesses better and leverage knowledge. to make our businesses better. Thank you so much for joining me today. My name is Marc Haine. I hope you stay safe, stay healthy, and I hope you dare to be the exception. See you next week.