Why boring AI use cases will win in 2026

David Sweenor 0:04 Hello everyone, and welcome to the data faces Podcast. I’m David Sweenor, founder of TinyTechGuides and your host for today’s show. Today, I am thrilled to be joined by Tom Davenport, Distinguished Professor at Babson College and one of the most widely respected and influential voices in analytics and AI. Thomas spent decades helping leaders understand how to turn data and technology into real business value. His writing, research and teaching have influenced, actually, how companies think of AI automation and the impact now of generative AI. So Tom, thanks for joining me on the databases podcast.

Tom Davenport 0:38 My pleasure. Nice to be here in this Hawaiian environment,

David Sweenor 0:44 it’s great. Well, you know, for those who may have been living under a rock for the last couple, you know, decade or so, can you tell us a little bit about yourself and your background and how you got to be such a luminary in this space?

Tom Davenport 0:58 Well, I don’t know about the last part, but I was actually academically trained as a sociologist, but I did a lot of work in statistical computing and teaching statistics and so on. So over time, I think I got more and more interested in the computing and less and less interested in sociology, and I eventually found my way into the consulting industry and worked for first boutique consulting firm called index, which styles itself as the inventor of business process re engineering, although I actually wrote the first article in The first book after I left the company, but then went to Harvard business school for a while, then went back into consulting at McKinsey, and then ran a research center for E and why, and then Accenture taught at the University of Texas. So I’ve kind of bounced back and forth between universities and business schools for a long time, been at Babson now for almost 25 years, but still work with consulting firms fairly regularly.

David Sweenor 2:09 Well, that’s great. And you know, it’s no better time to be in the AI and analytics space. I do have a question, as we’re jumping into 2026 you know what’s real and what’s hype? I see a lot of people bloviating about the potential impact of AI, but I really have trouble understanding, you know, what’s, you know, what’s, what’s durable, and what’s not. You have any perspective on

Tom Davenport 2:30 that, yeah, well, I think the most bloviation is about agentic, right? I contribute to the bloviation by co authoring a 500 to something page book about it. And I do think that it has a lot of potential in the longer run, but I don’t think it’s really ready for prime time now, and so most of the companies that I work with are, you know, piloting. You know, why should agentic be different from any other AI technology, right? Right? The subject of pilots, but trying to do things that are not terribly risky or important to see how it goes, and so that, I think is going, it’s going it’s going to be a while, I don’t know. And Andre carpathy recently said that it’s going to be a decade. You know, I would guess it would be closer to five years before we have real, you know, transactional applications of it. But I think I have to write some pieces about what’s going to happen in this new year. And I think the quest for value is the biggest thing. I mean, obviously it affects not just how companies use AI, but it affects our entire economy, since it’s all built around AI, these and so, you know, we’re starting to see some cracks in the valuations of these, some of these companies, and suggestions that maybe we don’t need a data center on every corner by a nuclear power plant next door.

David Sweenor 4:19 I was gonna ask you about that. So, I mean, do you think we’re we’re in a bubble? Is it going to pop?

Tom Davenport 4:24 I do think we’re in a bubble. And I think there’s a lot of real value to AI, of course, but I think the valuations of some of the companies are somewhat crazy. I think generative AI is particularly overvalued, and the companies that do it are overvalued. I believe in all forms of AI, and I think that’s a quite interesting and important one. But I think because it’s so accessible to the public and to the chattering classes, if you will, I mean. Times of people who write stuff and do podcasts that it’s gotten way more attention than it deserves in the overall kind of pantheon of AI. But I think all forms of AI will continue to be around and will continue to provide value, but that when it just has to come down to earth a little bit.

David Sweenor 5:22 Yeah, I think I agree with you. And, you know, I’m curious here, you mentioned business value right off of the top, like for agentic, is there like a killer use case you’ve seen? Because, you know, what I’ve seen is people trying to schedule calendars, which, which is a bit ho hum to me. It’s not all that exciting. And I can’t find a real decent use case that I’ve seen. I’m curious what, from your vantage point, you know, what are like? What would be like the killer use the first sets of use cases you think would be coming out for, for agentic?

Tom Davenport 5:51 Well, I’ve talked to a couple of different companies that are doing either with their software for as a vendor or professional services firms for their clients that are doing hold on to your hat Accounts Payable work that’s really exciting. But you know, it turns out that generative AI is quite good at sucking the important data out of invoices and sending a message to some other agent. In a lot of cases, people don’t trust typical agents, so they send it to stripe or something, right like that that they do trust to, you know, actually pay it when the time comes. But, yeah, that’s not exciting, but it is transactional. And I think a lot of people have not terribly exciting jobs, looking at invoices coming in and extracting the key components. So, yeah, you know it’s, it’s worthwhile, just not very, very interesting,

David Sweenor 7:01 okay, you know, one thing that was, you know, went back to your, you know, one of your first books on the competing on analytics, and I think you had a map in there, you know, the descriptive, diagnostic, predictive, prescriptive, does generative AI, is that another pillar is it? Is it an overlay? You know, how do you how should companies think about that? Or is it, you know, prescriptive on steroids, because it’s actually telling you what to do, and it might do

Tom Davenport 7:26 it for you, yeah, you know, I sometimes refer to it as predictive analytics on LSD. You know, it’s clearly predictive in that it’s predicting the next word or piece of an image or whatever. But I don’t I refer to it as generative as everybody else does, and then all the other stuff I generally refer to as analytical AI. And I’m on sort of a one person crusade to get people to use that term, because even though generative AI is also predictive, it’s not its primary purpose is not prediction. Its primary purpose is to create content and analytical AI, usually traditional machine learning. I don’t really like that term at all, but it The purpose is to make a prediction for the future based on data and from the past and models created from it. And that applies to neural networks as well, and deep learning as well, and so on. So I think we need a term, calling it classical, AI or no AI or whatever. I think is a bad route to go. So I picked analytical.

David Sweenor 8:55 I love that. Got a book up there. It’s all analytics. So that’s how I think about it. So, speaking of predictions, you know, you had to make one bet for 2026 you know, what are most organizations, you know, underestimating today, you know, and what that maybe will come to the surface in 2026

Tom Davenport 9:15 I think it’s that we’ll see a shift from individual level use of generative AI to more enterprise projects that, as someone else described it, I co authored a Harvard Business Review article about this with a couple of researchers from Stanford. But then few weeks later, some people came out with an article called, I think too many pilots. And Harvard review, right? Stop doing so many pilots, and they had a nice distinction between the broad and shallow implementations that we’ve done thus far with generative AI to. Deep and narrow ones. And we in our article, we call it shipping from individual level to enterprise level. And I think that will make generative AI more like analytical AI in terms of the the kinds of work necessary to pull it off and the kinds of value that you get assuming you succeed with it. I do think that, you know, we’ll still tell people, hey, you know, try this out on your blog post. But I don’t think we will rely on that for economic value, because it’s too hard to measure either the productivity or the quality of output, and hardly anybody does measure it. So in some ways, I was talking to a company that said we knew it more as an employee satisfaction and retention tool, which is, you know,

David Sweenor 10:59 okay, that’s super interesting to me. So, you know, speaking of value like, so what do you think the big enterprise use is going to be, you mentioned, you know, the financial, you know, accounts payable, things. Or is it going to be across, sort of evenly distributed, across all the functions? Or do you think it’s like marketing, for instance, is more ripe than than, you know, maybe another function, like legal, I’m making this up?

Tom Davenport 11:21 Well, I do think all of the custom for generative AI, all of the customer facing functions. So and I, I’m doing a little project now with some researchers from Cambridge University, and they have some data suggesting that, in many cases, calls that come into a call center or contact center are not just about service. In many cases, they are often opportunities to sell those customers more. But the people who do that work are generally there because they, you know, can answer nicely customer questions about service and or that’s the hope anyway, and they’re not very good at selling. So I think we’re going to see sort of a blend of these customer facing functions, marketing, sales, customer service and customer success. And the companies that have that function to incorporate how generative AI can really understand customer intent, regardless of right what they have in mind. And maybe we’ll see more of a breakdown between incoming dealing with incoming relationships with customers and outgoing, outbound and inbound, or somebody described it as proactive and reactive, but I think it makes sense to start combining some of these functions and having a channel that makes sense of what the customers want, and either sending it to another bot that can do that thing, or sending it to a human that can do that thing.

David Sweenor 13:03 So how you know to unlock this value Tom, like, how do organizations need to think differently with agentic and generative compared to traditional, whatever term we have you, are there new things that analytical, analytical, political? That’s right, it’s all analytics, but yeah, I’m just wondering is there are new things you need to consider that maybe you didn’t previously?

Tom Davenport 13:26 Yeah, absolutely. I mean, with analytical AI, the primary thing was making better decisions about key business processes, your inventory levels you should maintain, and your pricing, that you should charge, and who should we be targeting with our with our ads and offers for generative AI, of course, you know, we’re creating content, and there are, I think, a very different set of disciplines that are necessary. I’m not sure most organizations have these disciplines, but one is, as I was saying before disciplined experimentation, where you say, Okay, let’s try this with on a group of people, with generative AI and without generative AI, maybe we have different variations of its usage. And see, you know, do they do their work faster or better? I was talking last week to the head of data science and AI for Procter and Gamble, and they did an experiment using 776

David Sweenor 14:40 people. Very precise, very precise

Tom Davenport 14:44 in new product development to try to figure out how to generative AI help people come up with ideas. And it turns out that it did. It made actually individuals with generative AI were more productive. Than teams without generative AI and it also they came up with better balance of sort of commercial and innovation oriented ideas. But the key thing is that they tested it and ended up writing a paper about it with a bunch of academics. So that just doesn’t happen very often. I would say some vendors do it. You know, anthropic has tested a few things and so on. But in general, companies don’t do that. Disciplined experimentation. Another is just a discipline for behavior change. So, you know, we have this new term that’s emerged over the last month or two called Work slop.

David Sweenor 15:41 Work slide, I can see a lot of that my day to day all the time,

Tom Davenport 15:45 yeah, exactly is generated, of course, by generative AI. And so people have to learn how to get rid of that work slop and edit the output and add some value to the to the output of generative AI. There’s some spec, I think it was a McKinsey study that said the 80% of generative AI output never gets looked at. So my gosh, we have to change people’s behavior in that regard. You know, it’s really, I find it really challenging to do with my students, because they, you know, people seek the easiest way to do something, and we’re not necessarily all trained to be effective editors of content, rather than producers of good first drafts or whatever. So I think getting people to use their critical judgment and review content and try multiple prompts and really do something that that makes it interesting. Most of the content from jazz with AI is not interesting. And so that’s not an easy, thing to do, but I think it’s important, and it does, by the way, lower productivity. Some of my students told me, gee, you know, it was easier to just paraphrase a Wikipedia article than to what you’re talking about. Yeah.

David Sweenor 17:15 So that brings up a great question. So both in terms of, you know, education, but like, the future of work. You know, we’re reading, Hey, I can’t figure it out. Is AI gonna Yeah, net, destroyer of jobs or creator? Are the jobs gonna shift? I can’t make heads or tails of it. What’s your perspective on that?

Tom Davenport 17:33 Well, you’re not the only one who can’t make heads or tails of it. I’ve just finished an article. I think it’s coming out in this week or so about the inability to predict what’s going to happen to the workforce. And so we looked at a bunch of different predictions of how many jobs are going to be gained or lost, or what percentage of work could be automated and so on. They’ve been a lot of them, mostly from consulting firms, but sometimes from NGOs, like the World Economic Forum. They have only one thing in common, David, they’re all incredibly wrong, okay? And so I think we should stop that kind of prediction on a large scale anyway, and think about, you know, looking at particular jobs, or even particular jobs in particular companies, gets tricky, because there are a lot of them in large organizations, and we don’t necessarily have the resources to look at each job. But you know, some are going to be automated or supported by AI more than others in the short run, so I think it makes sense to start looking at them individually and say, What? What can we do differently? How can we augment that work? Or maybe we don’t need all those people. Maybe we can get by with with fewer of them. What we’re not seeing is that kind of analysis. And so we’re seeing anticipatory impacts of AI where companies say, Well, you know, we think AI is going to change a lot of jobs and eliminated jobs, a lot of jobs. So even though it can’t do that now we’re going to fire 15,000 people, and I think it’s really stupid. A, we don’t know how many people we should be firing. B, it demoralizes everybody in the organization well before they need to become demoralized. So I really think it’s a bad idea.

David Sweenor 19:39 Yeah, that’s event on both sides of that, but I agree with you. I think companies are using it, maybe as as an excuse, but they they don’t know what’s going to happen. So maybe a related question Tom is, how do you feel about like this and education? You mentioned your your students. You’re an educator, and. Especially early career people like, you know, it’s like, hey, we hired these people to do each other. Are they ever going to learn it now? Or maybe they’ll learn it in a different way? Because, you know, with AI, might handle some of that mundane stuff that you don’t want to deal with.

Tom Davenport 20:15 Yeah, I think those people are the most at risk. We’re, you know, we’re starting to see a little bit of evidence, not much yet, that AI is the cause. There’s some jobs like in my friend Eric Brynjolfsson and his colleagues did some work in this regard. Some jobs like customer service and programming that show substantial declines at the entry level. And those clearly are jobs that are affected by AI. There are some other jobs, they don’t mention this in their article, but there’s some other jobs, like digital creators, that are also affected by AI, but are not really showing any declines. They’re still increasing at the entry level, so but I do think they are the most at risk, and we need to figure out new ways to educate them, to introduce them to the workforce earlier, so they have more experience. I was listening to another podcast the other day, and somebody suggested, you know, you hire somebody at the entry level, it’s a risk that they’ll be able to do work quickly after you hire them. Because, you know, you don’t you don’t know how good they are, they don’t have experience, you don’t know how quickly they can learn it. So I think we’re going to have to look for evidence that people have learned something about a particular set of tasks and a job, so that they can show they can add value pretty quickly. And universities as a rule, we’re kind of slow to move in that regard. There are some who are starting to do it, but we need to move much faster.

David Sweenor 22:03 You have to go back to like, like, handwritten tests in the classroom. Or do you? I’m just curious, do you let people use Gen AI? Do you talk about it when you’re instructing the next

Tom Davenport 22:11 Oh, yeah. Well, you know, I teach about AI. So it’d be a little silly for me

David Sweenor 22:17 to say class, right?

Tom Davenport 22:20 What I do is I make them show their work. I say, Okay, show me if you’re going to write an essay, great, you should be using generative AI, but show me you know how to use it the right way. Show me all the prompts you tried. Show me the edits that you made to the output. Check the sources. One One of the things I’ve found is that, you know, because I teach this course called AI for business, and I decided this year, it was kind of like teaching course called electricity for business, because there’s so many different things you can do with AI these days. But so I tell them, pick something that you’re interested in. I think this might be good for job purposes too. You know, say you’re interested in healthcare, great. You focus a lot of your your research and running on AI and healthcare. If somebody is writing up an essay on AI in, I don’t know, automated reimbursement or pre approvals of healthcare, and they have a citation that says the use of AI in automated pre approvals for reimbursement, it’s almost 100% likely it’ll be made up because, you know, these systems want to please and if you say, I’m writing about this, they’ll say, right, right. I’ll give you a source about that. So I just make them do the work, but it’s really quite difficult. They don’t they’re not used to it. They don’t like it. Some I was just reading my course evaluations yesterday, my courses finished for this semester, and some of them said they appreciated it, but most of them, I think we’re not, we’re not that happy

David Sweenor 24:06 about it. You know, I was arguing with my my son, who’s, who’s, you know, going to be entering college, and I’m like, You got to learn the fundamentals. That’s how I learned, right? My background was in physics. I went out, like, to the fundamentals. And, like, maybe I don’t know if that’s the case anymore. Like I have this calculator. I know how to do long division, but I don’t do long division, so I’m always curious at like, what level do you need to understand the fundamentals? Moving forward, if you

Tom Davenport 24:35 know how to do long division,

David Sweenor 24:38 you’ll have to check the work give it to an AI,

Tom Davenport 24:43 yeah, I think this whole de skilling thing is another really important issue. And you

David Sweenor 24:49 know, are you wait? I have not heard of that D skilling. Yeah.

Tom Davenport 24:53 Dumber, I wrote a little substack piece about it. I’m not advertising my substack. We’ll put it. In the notes, I don’t make money out of substack. They’re all free. But it was on de Skilling and healthcare. And clearly, you know, de skilling, of your marketing skills, or whatever, you know, the world is not going to come to an end desk skilling healthcare, clearly there could be some problems. So I talked about several different episodes where I’d been confronted with healthcare deskilling, one of which was my doctor sending a message. I asked in an email, do I need to do something about a cold I had had been dragging on for months, it seemed. And so he sent back this message quite quickly, of course, saying Tom, I noticed my name was in a different font than the rest of the message, right? The message started, thank you for sharing the information about your patients upper respiratory disease. So I sent back a message, so what model are you using? And he said, Oh, you caught me. Wasn’t hard, yeah, right. And then I had another, you know, there are all these systems now that will record your visits to doctors. And my wife and I both went to doctor for annual checkup, and my wife’s, in particular, the clinical notes coming out of it said that my wife’s son went to Northeastern University, and she was having a lot of fatigue and sleeplessness, and she’d said neither of those things. Turns out the nurse had said something about that, and it got trapped into the clinical notes so and a few other examples. So I think there’s a real risk of us losing some critical skills in the headlong rush to get stuff out more quickly with with AI, yeah, I think people

David Sweenor 26:57 are going to get burned out, because companies always want to do more with less, and you don’t have time to check these things at scale. It was relatively easy, in my mind, with, you know, analytical AI, you’re getting predictions out, and you can test for drift and all that with content at scale, written, audio, visual, I don’t know anybody’s going to check that stuff.

Tom Davenport 27:17 Yeah, you know, maybe, maybe just as we have systems that check for drift, you know, ml ops type systems are checked for drift and analytical AI will eventually have systems that check for for work slop in generative AI, but we don’t now, that’s for sure.

David Sweenor 27:38 Oh yeah, you can see, if you push these things hard enough, you know it’s, it’s like, you get these short, choppy sentences and like, I’m like, just like the cat in a hat, Dr Seuss type, type stuff here is not good quality of anything.

Tom Davenport 27:48 And yeah, with too many M dashes and too many bullet points and so on, it’s just, yeah, we

David Sweenor 27:54 see it on LinkedIn with the ever increasing emoji laden post for use of money, how to use an emoji of like, come on. But so given that you know what’s, what’s the state of governance? Are companies taking it seriously or or not?

Tom Davenport 28:10 Well, I don’t know if I ever told you this, but I really don’t like the term governance. I like the term enablement.

David Sweenor 28:19 That’s good. You’re a one man, crusader

Tom Davenport 28:23 for that and enablement is making it easier to do the right thing. People don’t like governance, for sure. They don’t want to be told what to do. So I think if you make it easy to do the right thing, you know, automated checks for this or that, or, you know, I’ve done a fair amount of that. I wrote a book a couple of years ago, or co authored a book on what was then called Citizen development switch to by coding, now by vibing. The a lot of companies had ways of checking, if you you know we’re starting to develop an application that was beyond your capabilities or it wasn’t using the right kind of data, or whatever, and I think that should be our approach, rather than calling it governance and issuing rules and policies and so on. You’re starting to see that now with platform governance, platforms that are often called in AI that will look at every use case and say, sorry this one a is unlikely to work. B uses biased data. C is totally opaque to anybody who’s going to be using it, and gives you the opportunity to revise it, sometimes, even before it’s developed. With a, you know, people fill out a questionnaire, and this isn’t, this isn’t going to go well, but it’s least semi automated, and gives people some insight about how to how to do it. Better.

David Sweenor 30:01 That’s super interesting to me. I don’t, you know, we can’t even get my personal productivity machine to work properly without making up crap. So doing this at scale with multiple agents, I just feel like it’s going to be a comedy of errors. And, you know, maybe I don’t. Probably some company is going to get into hot soup this coming year for sure.

Tom Davenport 30:19 Well, yeah, and they already have. I mean, I was reading an article yesterday about all the lawyers that are getting in trouble using it, and now there is a new profession of lawyers to try to find out the AI hallucinations in legal briefs. So it’s good. It’s created some new jobs, I guess.

David Sweenor 30:42 All right, well, we’re coming up near the end of the time. So, you know, what’s the one? What do we have to look forward to? We’ve been talking a little bit about doom and gloom with the risks of AI. But what’s the positive? The flip side of is it going to rewrite business models? And you know, what do people need to think about?

Tom Davenport 30:58 I think that it is, you know, we it’s not going to do it on its own. We have to, I think that this whole area of business, process re engineering is coming back, enabled by AI. People are, I mean, people don’t like the term process quite as much as they used to, so we’re calling it workflows. But whatever. Okay, people realize that this can create new, holy, new ways of doing work, but we have to kind of examine it a bit and say what, what exactly can it do? Is it ready today with agentic? I think you should be examining your workflows to think about how it might work when it’s when agentic is really functioning well, which it isn’t now, but, you know, start getting ready for it, maybe putting pieces in place. And I think that’s true with generative and analytical AI as well. And I’m not sure anybody’s reading it, but I have written a few pieces in Harvard Business Review or whatever, about how a AI can really transform, you know, the process of thinking about how you do your work.

David Sweenor 32:07 Yeah, I use it quite a bit for for brain storing, but you have to push it. I mean, I’m teaching some, some of my clients I teach teaching, uh, AI, and I’m, like, never take the first output, iterate on it and then, and then ask it for alternative interpretations of it, because whatever you get out that’s one one perspective, one way of looking at it. There’s probably a dozen other ways of looking at it. I think that’s what people don’t do. They sort of get one thing and it’s dogma, and they just stop, yeah, right way to use it, yeah.

Tom Davenport 32:36 And they may not even read it to see is it of decent, you know, quality so doing things the right way is never easy. I guess that’s the lesson here.

David Sweenor 32:47 There we go. Well with that, Tom, I appreciate your time on the databases podcast. This has been a super informative discussion. Great, great one to kick off the new year. So I appreciate you being here.

Tom Davenport 32:58 My pleasure. Nice talking to you. Cheers. Bye, you.