SPEAKER_01

The 74% said they were afraid they're going to lose their job in two years if they couldn't show results.

SPEAKER_00

Where is the friction for other companies to be getting the kind of results that we're getting internally? Is it a training? Is it a budget? Like what's the constraint?

SPEAKER_01

It is going to take some upfront money, but it's also going to take organizational change.

SPEAKER_00

Person driving generative AI in the company, how should they approach it so that the employees aren't thinking, oh, I'm just basically training myself out of a job?

SPEAKER_01

The answer is you have to guarantee their job. The domain knowledge that they already possess about your specific business. How valuable is that? Super valuable, right? Because what they've got to do is help you build and improve the agent. I have no idea what it will look like in five years, but what I do know is every single one of these forecasts has been wrong.

SPEAKER_00

Peter Capelli is a world-renowned expert on workforce strategy and a professor at Wharton. He's helping business leaders face the reality of AI, not with hype, but with hard truths about what's actually working inside companies today, while championing the need to preserve human connection, trust, and meaningful work. Welcome to Using AI at work. I'm your host, Chris Dave. Each week we'll be learning how today's business owners, entrepreneurs, and ambitious professionals are getting more done with smart use of tomorrow's tech. Let's get started. Right now, every business leader is asking the same question. What are we going to do about AI? If this is you, ChiefAIOfficer.com has the answer. We give you a simple path forward where we provide executive and team training so your people know exactly how to safely use generative AI in their day-to-day. We also manage the deployment and implementation to make sure tools actually get adopted and deliver results. And we'll also guide company-wide transformation so AI becomes part of your operating system, not just another shiny object. The companies that act now will increase productivity, cut costs, and grow faster than their competitors. Those that wait will get left behind. So if you want to make AI work in your business, visit chiefaiofficer.com and see how we're helping companies of all sizes finally get results from AI. Everybody, welcome to a new episode of Using AI at Work. And our guest today is Peter Capelli. Peter and I, this is round two. We tried to do it before and I had some internet issues, but we did have a chance to start to dig in at least, and I'm very excited about continuing the conversation. Peter has a long history, really uh on a macro level of evaluating um the relationship of employees and today's workforce. And uh we're gonna be talking today about that, particularly its impact of uh the AI and how is that um kind of leading the conversation in offices these days. So, Peter, uh before we get started, um you've got a long uh and robust uh career and biography, but considering the audience, why don't you take a second and kind of um share with them a little bit about this um this esteemed history in this concept and this topic that we're gonna be talking about?

SPEAKER_01

So uh my personal history? Yeah. Uh yeah, so it's going back longer and longer, um surprisingly, as long as we keep keep going. Uh so I uh began my career in economics and kind of hands-on e-economics in England, studying inflation. And as soon as I finished my thesis, inflation more or less ended. So that was that. Uh I went from there to MIT to study um collective bargaining union issues. Uh, and by the time I finished that work, unions had more or less collapsed. So you get the pattern uh going here. Um uh and I was at the Wharton School where I've been forty 40 years now, and uh had to teach things about management. I got pulled into some policy work in the U.S. I worked for the Secretary of Labor on a commission she had that was Ann McLaughlin, and then with a colleague, I ran a research center for the Department of Education on workforce quality issues. And basically what those were about was recognizing the employer's role, uh, which seems pretty obvious, but was not quite so obvious at the time. You know, who you hire changes the labor market, changes what students think and what fields they go into. Whether you train people matters, whether you give them jobs that are increasing in responsibility matters, um, whether you're developing talent within, that sort of stuff. So I did that for about 10 years. I wouldn't say that was wildly useful, um, kind of interesting, but not wildly useful. I did some work outside the US on that. And then starting about 2000, um I started focusing more on issues around human resources as it was being practiced. So how hiring practices work and performance management systems work or don't work, and um, you know, the labor market as it intersects the reality of what employers are doing. So focusing on what employers are doing. So we started interested getting interested in uh AI topics with the rise of machine learning. Um, so we published some things at the end of 2010s about what machine learning was doing in organizations and the lack of fit with human resources and the legal framework which underpins human resources. And most of those issues are the same with large language models. Um, and so I've written a few things about, one particularly about large language models with colleagues, and one thing looking at what a particular company did, RICO, in its effort to try to automate some simple frontline white-collar paper pushing tasks. Um and right now I'm doing a series of podcasts uh sponsored by Wharton and Accenture, where we're looking at what companies actually do with AI, which turns out to be incredibly harder than you would think. But it's a reminder that there's so much talk about AI and very little actual efforts to imply it, right? It's mainly talk. And it's mainly by talk by people who are not actually doing it. There are people who are uh advocates for it, and some of them are people who build the AI itself and are telling you what it could do. I think it's always useful before we jump into that to just remind ourselves of the story about driverless trucks, which were predicted to take over the field by 2019. And if you ask yourself why that prediction was so completely off, you get some insights as to what's happening with AI now. And that is the focus was all on what it could in theory do. But when you started to think about what was practical or what was cost effective, you started to see use not being anywhere near as great. So that's kind of the path that takes us to the conversation today.

SPEAKER_00

Yeah, awesome. You know, as somebody who spends a lot of time inside of companies, uh actually helping companies do the thing when it comes to AI, this is particularly fascinating because uh I don't know that I necessarily um I guess I live in a bubble, right? Not only do I know what's possible, but we actually execute internally, we help other companies execute and that sort of thing. So um I've seen some statistics recently. Now, the sources uh the, you know, I don't know how much what the data set was or the sample size or anything like that, but the uh amount of companies that are making an effort to introduce AI and being successful is pretty low. Are you finding that in your research as well?

SPEAKER_01

Yeah, well, we hear from our colleagues in the consulting world that it's about 5% of companies are giving it a serious push. The year before that, I did a quick poll at an MIT conference, uh, which was AI related. Yeah. And we did a series of questions asking about their use of large language models. And when you got down to have you done anything that fundamentally changed a job in such a way that you could replace somebody was less than 1%.

SPEAKER_00

Yeah. So let's talk about that then, because I uh, you know, here's what I see in the marketplace high demand, yeah, right, like a lot of interest. There's enough um anecdotal stories being shared in the media to where people say, ooh, I want to get those results. Yeah. However, when I talk to a lot of business leaders today, one of the things we do is it's informal, but you know, on a scale of one to five, where do you kind of like where's your AI knowledge? One being brand new, five being um you're an expert. And most of them are six months ago it wasn't the case, but recently we're hearing a lot of people self-identify as threes and fours. And when I ask them, I'm like, that's pretty impressive. When I ask them what like what are you using it for? It boils down to really just some basic things. And it's usually writing emails, summarizing documents, and you know, I'm I'm trying to figure out very helpful, of course, and and um certainly saves time and and that sort of thing. But what I'm not figuring out is are they just not aware of what's really possible, or are they overestimating their own skill level when it comes to this stuff? Or what's your thoughts on that?

SPEAKER_01

Yeah, that's a really profound question, um, for particularly for practice. I think um you may have seen the survey that was a minor story, but it should have been a bigger one. Uh a Harris poll, I've forgotten the company that sponsored it. They said of C-suite executives that 74% said they were afraid they're going to lose their job in two years if they couldn't show results. And when result they didn't define results clearly from AI, but I'm pretty sure what it meant was um cutting headcount. Uh and when asked um what tools they thought were available, and they said uh 60 plus percent said they thought off-the-shelf tools were gonna be able to do it, which is absolutely not true. Yeah. And the pressure is from their board um to do this. So the the story I think begins, at least when I've asked other people, but I should ask you this question too, why you think the boards are so out of touch. I mean, they just don't know. Uh they're just following the headlines in the paper. Yes. And they think uh this is possible, so they're pushing, they don't know, but they're pushing the CEOs, um, who are flummoxed as to what to do. Um, so I think it's expectations for sure are way out of whack. Uh I think the people who know what you could do understand it's a lot of work. And the problem is you're not going to be able to meet those expectations from the board. You can't get quick results with off-the-shelf stuff that are going to cut headcount. So I think they're a little bit paralyzed. You know, they're they're keep looking for some magic application that will show these results and keep the board members happy. I think the only real way is to persuade the board members of the reality of the situation and you know what is possible. And the things that are possible are really important. But you know, they're just not the magic bullet, right?

SPEAKER_00

So let's kind of talk about that. We've got executives that that are, you know, uh under this mandate that are listening to this podcast. How would you advise them to kind of manage the expectations of the kind of out-of-touch board members that have seen the stories? And and look, at some point, yeah, we'll be having a different conversation. But today is what we're talking about. Yeah. Um, how would you advise them to uh kind of counter that uh enthusiasm from the board to have the magic bullet?

SPEAKER_01

Yeah. Well, I think one uh issue just ahead of that a little bit is why are they so fixated on headcount? And this is something that I've tried to been talking about for a while, and that is the quirkiness of financial accounting, which is the rule book or basically the um the measures that are used to determine who's winning in the corporate world, right? And there's so many things about headcount that are quirky in financial accounting. You know, employees can't have value in financial accounting. There's no value attached to people because you don't own them. Training can't be an investment because you can only invest in assets. Can't be an asset unless you're interested. So many things are measured on a per employee basis, revenue and um profitability. So if you could cut your head count, those things should weigh up and you look way better. So there's all kinds of reasons why they're fixated on that one issue. Uh so I think one of the things they have to do is get them off that one issue and say, you know, this these tools can help us do new things, they can help us do things better, uh, and they can, in the longer term, improve our productivity. Um, but in the short run, we're not going to see tons of headcount drop from this. Uh, I think you got to walk them through some examples. Um, and uh let me walk you through one story on this, which uh I heard from a manufacturing company, and it was featured on one of our podcasts a little while ago. They spent a fair amount of time in a machine learning model. I think one of the things we're discovering is machine learning might be way more useful than we thought in a lot of these tools. The machine learning model they were trying to develop was one that could assess the quality of the paint jobs that were coming off their assembly line. And to do this is something that they had employees who were really good at it. Yeah. Uh, and I'll come back to that in a second. But they wanted to see whether they could build a model, an algorithm that would be as good at judging the quality of the paint job as employees would do. It's a pretty tractable problem, as you could imagine. You know, you got you can train it on a bunch of data showing, okay, here's the what the color is supposed to be, how close is it to that particular color. You could see how you could do that. It does take a lot of work, and one of the problems, as you know, is every time you change the color, you got to change the algorithm. Um, so roughly, right? Uh got to be trained differently on different colors. But they did it, and the reason they did it was uh to prove to the company and the executives, and presumably to the board, that you could do this and it was going to be useful. Now, it turned out that they still wanted to keep the employees. It wasn't wildly more valuable than the employees, particularly if you did a return on the investment exercise here. Yeah. And the employees also knew something that AI models, the machine learning models, didn't, and that was they had a sense about why, when the colors didn't match, why it was, what was going on. It might be humidity, it might be temperature, it might be something else in the assembly operation. So it wasn't like this was going to be a game changer, but it did persuade the company that uh you could do something like this. Here's what it's gonna look like, here's what it costs to do it, but now look what we got. And that provided the lever through which they could secure other investments to do fancier things and things elsewhere with AI. So I think one of the things that reminded me of is this is in part an organizational change process.

SPEAKER_02

Yeah.

SPEAKER_01

You got to persuade the organization we need to spend some money on this, um, and we will get something from it. I don't blame the CFOs and the board and everybody from saying, yeah, yeah, yeah, everybody promises, right? So they had to show them something, right? So um I think that is one example. Um I think here's an example of what was really hard to do and turned out to be way harder than people thought. And this is one that we wrote about in the Harvard Business Review, and this is from this company, Rico. Rico is the copy company. We think of them that way. They think of themselves as basically a data uh imaging company, trying to take data, put it into a digital format, right? And they do, among other things, back office work for lots of companies processing paper. And this particular one was for insurance companies, health insurance companies. So uh it's quirky how quirky this process is in healthcare. Uh, so we'll explain why in a minute. But basically, envelopes come in, and the first thing they got to do is open the envelope and figure out who it's from. And then what is it? Is this a claim? Is this an appeal to a policy? Is this responses to questions we have asked? Is it something else? They got to stack them, basically. Just sort them out, right? And this is harder to do than one would think because you you might say, well, just tell the people filing claims to use an online system. The problem is you're dealing with thousands of hospitals, thousands of providers. Each hospital may have a hundred different insurance companies they're dealing with, so they just use their own form. And if you tell them, no, you got to use our form, they're gonna say, the heck with you, uh, too hard, right? So then they've got to move the data from the paper form, once they figure out what it is, into each client's database. Okay? So that sounds pretty straightforward as well. Um but you start thinking about building the model that will do that. Um, you have to build it separately for each form. Every time the forms change, you almost need a different model to do it. They discovered that they could get a large language model, a fancy one, that could actually do that pretty well, just looking at the existing forms. So they didn't have to do build a lot of separate models and things. They had a f AI was good enough there to do that simple task. What they concluded, though, was it was far more expensive to get the large language model to do it than to get the employees to do it. Interesting. So a big aha for a lot of people, and particularly for boards, is this stuff is not free. It is not the case that they're giving it away, right? They give away ChatGBT, but that's not gonna do it for you. Yeah. And so then, you know, you might say there's this view among people who are not really in the world of data science and a and IT generally, that it's just obviously going to get a lot cheaper. And people who are close to this than and I'm not, tell me not so clear, right? One reason is because supply and demand and demand is shooting through the worth roof. Yes, electricity costs are zipping up as we all know. Yep. Computer time and power is not, you know, Moore's law doesn't work anymore, stopped working a while ago. It's not clear how easy it's going to be to get additional power, computing power, and to pay for it. So it's not cheap.

SPEAKER_00

Yeah.

SPEAKER_01

They figured out an alternative way to do it, which basically is very smart engineering, just stacking different levels of AI, right? So how much could you do with the cheap free stuff? And it turned out they could sort about 50% with the cheap stuff. And then with the next 50%, the harder stuff, they could use first an intermediate level cost one, and then they could save the expensive AI for the very hard ones at the end where they couldn't read handwriting or something like that, right? It turned out, I mean, you could make this work. Uh it was a big onfront investment. It did take a lot of time. It took about a year. Uh six people working full-time on this. Um, and those were outside people too. So pretty expensive thing to do. It still didn't cut headcount by that much. And the reason is they still needed people to track down the problems after they identify them and provide the answers. So, what was the problem here? I d I don't know. Well, it turns out they couldn't read the handwriting. Okay, or I got to go back to the original form, or I maybe got to call somebody, you know. So the work changed from simple, really boring uh hand coding to problem solving. That was a good thing because turnover of those employees before was terrible. Yeah. And their own quality with work like that gets boring, was terrible too.

SPEAKER_00

Yeah.

SPEAKER_01

So they improved the quality a lot. Um ultimately they will be able to do a lot more per employee. Upfront costs were big, uh, and headcount didn't fall. And I think their view, which I I read' uh right to me, simplest stuff is gonna be a lot harder to do than we think. And I think the reason for this is it's gotta be a hundred percent right. Right? Yeah. When you're opening that envelope and you're saying, is this a claim or is this an appeal? Gotta get that right. Yeah. When we come to what Chat GBT can do, it doesn't have to be perfect, right? It just has to be better than what we were what we know now. And often we don't know anything, so it doesn't have to be that good. But these things have to be perfect. So you gotta build those models. Expensive.

SPEAKER_00

Yeah. It's a great lesson for anybody who's uh thinking about these kind of uh complex processes that involve humans and thinking, oh, we'll just have AI do it for cheap because compute is going to be getting expensive. Demand for compute is causing some of these models to actually throttle user accounts to some degree. One of the things that I want to kind of go back to was this this head count. I've got a few questions here. Yeah. Do you think that it's that that is a metric that boards historically were able to say, oh, this is a good thing. We're reducing costs by so do you think that they're kind of like, even though we have this new technology, they're still leaning on uh old way of looking at things?

SPEAKER_01

I think so. They're stuck on headcount. Okay. Yeah, and that's why this was so appealing to them because the promise was we'll cut headcount. Headcount really benefits us with respect to investors because of the measures they use. They also hate fixed costs and they think for some reason labor is still a fixed cost. It's not. You lay people off all the time, but they think that's so I think that's right. They're really fixated on headcount.

SPEAKER_00

Okay. So as a as an executive who is having to uh kind of translate the reality into the the viewpoint of a board who is still kind of stuck on, they don't understand it well enough to know that the whole paradigm is changing. Uh the best advice for them would be what perhaps uh some effort at educating the board that the paradigm has changed, that that is not necessarily the metric they need to be measuring?

SPEAKER_01

Yeah, I think uh, you know, I've never been a big fan of lying. And I think the problem is if you try to deceive them, uh you better get out of there before they find out that it's not gonna happen. Right. Yeah so I I just don't think you're going to be able to cut headcount quickly and easily with anything that we can see, you know. Chatbots can replace people, but most every job where you can do that has already been replaced by a chatbot. They're just getting better now with AI, right? So um I think the I I think you can walk them through some examples and say, this is a success, right? Back to the manufacturing company I was telling you before. You know, that was what they were trying to do. They knew it wasn't going to be a huge cost savings, but they said we want to be able to demonstrate a success so they can see what it looks like. And then you can say headcount going forward will be less. Okay. And in the meantime, we reduce some of our turnover problems on employees, right? And quality goes up. So we're improving quality. Um, and they may not care so much about that. Longer term, we're improving headcount. It's going to come down. We're reducing turnover, those things cost a lot. Part of the problem is they don't see turnover costs, even though they are big, they don't care much about them because they don't appear on the financial accounting forms, right? Labor does in a big way. So it's a quirk of financial accounting. But I think you're right, the thing we have to do is educate them as to this is what this tool can actually do. You've been misled in thinking that cheap stuff will work, and you've been in let misled in thinking this is plug and play, and um you can easily cut headcount.

SPEAKER_00

So, would a better redirection of that conversation as an executive who's had having to address the board perhaps be we're going to be able to increase bandwidth without increasing headcount. Yeah. Yeah. Okay. Yeah. Yeah. That's when they'd like, probably. Okay. So that's a good tip. Um, and one of the things I know that there was a study that Boston Consulting Group did with Harvard in September of 2023. They kind of took two groups of BCG consultants and put them on a project. Half of them went and did their regular routine. The other half said, hey, wait, before you go do that, here's Chat GPT. This was September 23. It was still relatively early. And at the baseline metrics, after 12 weeks across 18 points of data, um, they showed that those who were using simply Chat GPT in knowledge work were producing 12% more, about 25% faster, and at a 40% higher quality. So that doesn't necessarily mean that we got rid of people or we reduced people, but the people that we did have through augmentation, through being human in the loop, human plus AI, was actually getting a better result.

SPEAKER_01

So um You know, I'd have to go back and look at that study, but as I recall, it made some things easier and some things harder, um, depending on what the tasks were that you were doing. And is often the case, even with computer programming, as I understand those studies, um, that it uh speeds things up initially, but it slowed things down on the checking part. Now, maybe they've been able to do something about those things, but you know, it throws up lots more ideas. Yes. Some of them not so good, right?

SPEAKER_00

And so human discretion needs to still be involved in what is good and what is not.

SPEAKER_01

Yeah, and that may, you know, might that might get better. Um but I think for sure you can say that there are some tasks that they're where they're going to help. But maybe we could turn to white collar, because I think that is where the action is. Okay. Uh, and not what people were promised, right? They were promised they'll get rid of all low-level jobs. Yeah. Turned out, I think, is not playing out at all. Um, but let's talk about that, because that's where the action is, and that's where the results are surprising. And if you think about uh some of the examples that I've heard that where this works well, here's what you have to do. First, you have to know what AI is capable of doing, what a large language model could do. And that knowledge is kind of specific. Um, not, I don't think, all that wild widely held in a particular context, you know. And then you have to take the job and you have to break it down and say, let's look at all the tasks you do in a job. And that is even more narrowly defined than a job description. So, you know, what do you have to do on this particular task? So, you know, we're gonna be developing a marketing platform. Okay, what are the steps in that platform? And then within each general step, you got to break it down further.

SPEAKER_00

Yes.

SPEAKER_01

By the way, this is the work that human resource people used to do. It's called job analysis. When you were gonna create job descriptions, you're gonna hire, you're gonna do all that stuff. We largely have abandoned that, but it's got to come back now. And the reason is then what you do is you look at each of those steps, and somebody who understands AI working with you, not by themselves. They're not gonna get anywhere by themselves, has got to say, with you, the expert on the job, here's where I think we could plug an agent in, and they could take over this task. And then you go down and say, I think we could plug another one in here. Okay. And once you've done that, then you have to rebuild the job. You know, for most jobs, you might have 20 or 30 tasks you got to perform. And you're not gonna eliminate all those tasks. You know, it's kind of like the story about programming, where it turns out computer programmers spend only about 30% of their time writing code, typical programmers, unless you're in a big code shop or something, right?

SPEAKER_02

Yeah.

SPEAKER_01

The rest of your time is negotiating for budgets, talking to clients, specifying parameters, all that stuff, right? So then you've got to reconvine the jobs and say, okay, what you know, maybe we've moved 10% of your work away. What else could you do? And let's look at what the next job next to you. Maybe that person has lost 30% of their job. Maybe we can put these together. But if nothing else, the workflow has to adjust.

SPEAKER_00

Yeah, that makes a lot of sense.

SPEAKER_01

Then you need uh the person who's gonna do that work to help the agent get better. And that means you know, the agent sent me this answer to this, but that wasn't a good one. Here's why. We gotta fix it, gotta go back to the programmer, the AI expert, and tell them how to fix it. So it's a lot of work redesign, right? With an IT person who knows what's going on. So you know, and it takes a long time to do this. Once you do it, the big discovery is things move way faster.

SPEAKER_00

Okay.

SPEAKER_01

Uh and you can get more done. You could get two projects done in the time maybe it would have taken you one to do. You can get this consulting project done in half the time. You don't necessarily save headcount, it's possible you might, but mainly it's faster and better.

SPEAKER_00

Just opening up more bandwidth capacity.

SPEAKER_01

Yeah. That you could do other things with, right? Um and you actually get better. The tasks are probably done better if they could be done by an agent. You know, like review the literature on this topic. Yeah. Take me a long time to do that. Uh agent at the very least will give a first cut that's pretty good. And then I got to help the agent make better cuts later on. You also can see from this, though, you need the cooperation of line workers, line employees, where this is not going to go anywhere. If you try to get some consultant to parachute in and the employees don't want to help them, this is going nowhere.

SPEAKER_00

So can we talk about that? Because that resistance um or potential resistance, because the narrative is AI is going to take jobs, right? There's also the counter-narrative that AI will create jobs that don't exist. Like, for instance, you're talking about the we've we've broken down this workflow into steps, we've identified some areas that we can introduce automation, augmentation, agents, whatever. But you mentioned something, and now there needs to be somebody who is reinforcing the agent with what's good and what's bad. Yep. It's not a job that exists necessarily wide scale today. Um, but the workers aren't hearing that. And and even if they are hearing that, that's change, right? And well, I like doing it the way I did it, right? Why do I have to do it a new way? So there's there's going to be some friction in this process. Um, how would one go about? Let's say I go into a company, they want to introduce uh at least generative AI solutions into some of their workflows and processes. I sit with the workflow owner and we start to map out what they're doing step by step, just like you described. Um at some point, or maybe throughout the entire effort, they're going to be thinking, oh, they're just trying to figure out what I do so they can get the robot to do it for me, right? Yep. So how how does somebody who can't bring in the change management consultant that but they want to do this in their department, maybe not at the enterprise level, maybe it's SMB, maybe it's lower middle market, whatever, but they got to figure it out. How does that person driving generative AI in the company, how should they approach it so that the employees aren't thinking, oh, I'm I'm just basically training myself out of a job?

SPEAKER_01

Yeah, it's a very important question, and it's an old question, and we know the answers. Uh so this problem uh has been around at least since the 1960s, which was the decade with the fastest technological change in modern times. Um and the answer is you have to guarantee their job. Now, you might say, well, why do you mean guarantee their job? Well, in a typical company, you know, turnover rate across employers in the U.S. is about 30%. Most of the companies we're talking about, it's probably a lot lower if you take out fast food and all that stuff. Say it's 10%, right? Uh and I I like what PWC said. Other companies have done this before. They told people, as long as you're willing to learn uh and to change jobs, we guarantee your job. You should. I mean, if somebody is a good worker and if they're a bad worker, you should be not be employing them anyway. If they're a good worker and they're willing to help you in this, uh, you should guarantee their job. Say, look, you know, it may not be this job here, but it will be a job, and we're not gonna cut your pay or downgrade you. And there will be opportunities just from regular turnover. Yeah. So, you know, it's not like you're you're bloating the organization by guaranteeing these jobs. You're not gonna get the productivity increases that would eliminate their work for years anyway, or at least a year or two. So you've got to tell them that. Look, if you'll help us with this, we guarantee your job.

SPEAKER_00

Uh what about the you know the domain knowledge that they already possess about your specific business? How valuable is that?

SPEAKER_01

So if you just go and super valuable. Yeah, super valuable, right? Because what they've got to do is help you build and improve the agent, right? And the way you do that is you know whether what the agent is producing is useful to this particular task. The agent at first is gonna produce generic answers. Yeah. And you've got to get it fine-tuned where it produces answers that work for your organization. Nobody can easily do that except for that current employee, right? And if you try to bring in a consulting company to do this, you're gonna bleed money forever to get them to do the same thing that your current employees can do for you right now, right? So, you know, you got to persuade people maybe a little further up the food chain that this is such by far the best way to do this. Yeah. And it's also, by the way, just not that you might care about being humane. It's also the humane thing to do. Yeah. And the thing that will build loyalty with employees rather than undermine it, you know, so why not do this?

SPEAKER_00

You you mentioned that humane element. One of the things that I don't get the question a lot, but when I do, I have no answer. And that is fast forward five years, the technology's improved, uh, the adoption has uh accelerated, and the the AI is capable of doing all the things that we've talked about today where there's still a challenge in the humans. With with the study of you know the workforce that you've done over your career, what do you see happening or what what would you forecast would be happening with all of these employees who maybe they're doing the boring stuff, the not sexy stuff that that is easy for AI to come in. Where do they go?

SPEAKER_01

Well, I guess I would say uh I have no idea what it will look like in five years, but what I do know is every single one of these forecasts has been wrong. So that's comforting, okay. Since the 1990s, you know, when we started getting more computers, everyone has been wrong. They've way overpromised what the change will look like. They've way overpromised headcount reductions, and they've been wrong every single time. So, and they're always wrong in the same direction, right? So I guess I would bet on the path here in the past that there's a pretty good bet that that's not gonna happen. Um and that if it does happen that it takes over more and more of their jobs, you just slow down your hiring rate, right? Which is doesn't mean you lay people off, right? You end up laying people off only if you have some dramatically fast transition, which we're not seeing, right? Yep, agreed. And we have and we have yet we have yet to see it anywhere, right? I it maybe there's some particular job where we've seen it in the past, but you know, if you think particularly about jobs, most jobs which have multiple tasks to perform, white-collar jobs, you know, we we just haven't seen it, right? So I I guess I wouldn't be too worried about that. And it and I guess I would just remind companies, businesses in particular, seriously, how far out do your business plans go? Yeah. And in most companies, it's one year. You know, they got a five-year plan, but if you update it every year, you got a one-year plan.

unknown

Yeah.

SPEAKER_01

So if that's the case, look, you're planning one year out right now. Um why uh how much time do you want to spend worrying about something that might happen in five years? Right?

SPEAKER_00

So So let me let me ask you about like I I work with it. I I'm like in our business, it's very integrated. We're seeing a lot of the impact from it. Where is the friction for other companies to be getting the kind of results that we're getting internally? Is it is it a training? Is it a budget? Is it a what like what's the constraint?

SPEAKER_01

Yeah, I think it's those two things for sure, because if you think about what we described, and I guess what you're doing, it does take some upfront money to do this. You do have to bring in some experts who who have done this before. Uh if you've never done it, uh doing it for the first time by yourself is going to take the learning curve.

SPEAKER_02

Yeah.

SPEAKER_01

Uh way up the learning curve, right? So it is going to take some upfront money, but it's also going to take organizational change. And change means, you know, not just reallocating bodies and things, um, but also making some commitments, and they might be some to employees, for example, right? Um and all that stuff is enough to stop. You know, right now, as you probably, I'm sure, are seeing, companies are getting cheaper and cheaper. And I don't know, it's a little hard to explain historically why. You know, the performance is typically great in the economy. You know, returns in companies are great, but they're getting cheaper and cheaper and cheaper, and feeling the needs to squeeze and squeeze and squeeze. So it does require some bandwidth to do this stuff. And if you've cut headcount so much that nobody has the time or the staff to do it, then you're gonna have a problem. Uh, and I think there's a lot of that. Um, ultimately, as we said, this is workforce transformation. Yeah. Where are those people in your organization? My bet is if you looked around, you don't have any because you got rid of them a while ago. Yep. So, you know, they they don't know where to start except to go outside and bring in a consultant to do it. And that's not gonna work very well for long. Yeah. You know, bring in a consultant to work with your people, perfectly fine. Expecting you can outsource this, not gonna work well, right? And the size of the bite you have to take to bring in consultants to take this whole thing on, yeah. The ROI on that is not gonna be great. Right? So I think that's what's holding us up.

SPEAKER_00

Who should be leading this? Because I get it. The organizational change, including the like the the culture conversation internally, who would typically or who should be leading this in these companies?

SPEAKER_01

Yeah, well, I think that is a great point because you know, my sense is that the interest in management, and by that I mean, you know, the inside of the organization, has just not been there for the top leadership in the last decade or so. They're focused primarily externally on financial deals and MA stuff. Yep. That's what gets attention. And this kind of exercise internally requires their attention, and they're not inclined to do it. And some of it is, you know, the people who are in those jobs often now are people who got there from doing outside deals and financial stuff. And you're telling them, look, you got to lead an internal organization change process. That means we need to see you in the factory or in the plants. You got to talk to people and all that stuff and say, really? I you know, I just not good at that. Can't we outsource this? I mean, I think that's some of it is that, you know?

SPEAKER_00

And it doesn't seem like that that's an investment in the future. Yeah. And if they're if they're planning for the next year, yeah, well, they're they're going quarter by quarter. They're not thinking about the future necessarily.

SPEAKER_01

That's true too, yeah.

SPEAKER_00

Yeah. What will happen as the technology accelerates and they have been focused on the next deal or boosting revenue and not necessarily pruning the garden, so to say, to prepare it for the next harvest.

SPEAKER_01

Right, right. Yeah. So, yeah, how can you manage that process? Well, I think um we hope that the in, you know, all this kind of goes back to the investment community.

SPEAKER_00

Yep.

SPEAKER_01

Who more than any other group is driving the agenda for these boards and the C-suite, right? And, you know, boards and company leaders do care an awful lot about whether they're doing what everybody else thinks is the smart thing. And uh, they can't, you know, say maximize short-term profits. Okay, well, how do we do that? I mean, there's nothing in saying we're going to maximize shareholder value that tells you what to do or how to do it. So I think we might as well play with the bias that the boards already have. And the bias is, oh my God, we got to get on this AI bandwagon, and persuade them this is the only way the bandwagon goes. And we don't want to be the only people who are not on that bandwagon. Yeah. Right? You'll look stupid. Stupid. If you say what they ask you, you know, at the conference board or the business round table, and you say, Well, what are you how's your AI going? What are you doing? Uh nothing. Nothing. Yeah. You'll feel dumb.

SPEAKER_02

Yeah.

SPEAKER_01

And people will look down on you, and then the investors won't like you because you'll be seen as this backward-thinking company. You know, so I think that's kind of the way to push it. It does come back to the conversation about persuading the boards about what's possible. Trevor Burrus, Jr.

SPEAKER_00

No, I thought that was a great point that you made earlier. And there have been a number of them that are certainly takeaways and it will help me in the conversation that I'm having with business owners for sure. What other trends historically can we kind of look to s to kind of almost forecast what's going to happen with AI? Is there anything in the workplace yet?

SPEAKER_01

Yeah. Well, I think if you if you went back to like the 90s when we started seeing these big investments in uh IT equipment. Uh and we weren't getting the payoffs from them. You may remember the famous quip from uh Robert Solo, the Nobel Prize winning economist, that you can see the effects of uh investments in IT everywhere except in the data. You know, that uh the the talk was all you couldn't see it. Well, eventually you start to see it, yeah, but it takes a long time. And ultimately you start to see it when not when you're dropping the computer in, but when you start reorganizing the way work flows. That's when you start to see it. So I think that that would be what I would do is go back to the earlier lesson.

SPEAKER_00

But it makes a lot of sense. What was the lag time for investment in infrastructure before it started to uh people were like, oh, it turns out it was a good idea?

SPEAKER_01

You know, I'm I don't know. I'm only guessing on this, but I think it was at least 10 years. Wow. You know?

SPEAKER_00

Are you expecting uh a compressed timeline for AI?

SPEAKER_01

Well, I think what we can expect is that you can see job by job effects reasonably quickly in a year. So back to my model about the or my story about paint uh quality assessing paint quality. You know, they did that within a year, and you could show the board and everybody else, look, bang, here's what we got in a year. This can work, right? Uh and I think that's the way to sell this. You're not gonna sell it on productivity, corporate productivity numbers. You're not gonna sell it on REI, ROI for the overall corporation. I mean, you know, there's no single thing that we've ever been able to show, just do this and it moves the needle on share price, other than financial manipulations and stuff, right? So so forget that, you're not gonna show that. But you can show job by job improvements that will ultimately reduce headcount, meantime improve quality, cut turnover costs, that stuff. So I think that's that's a way to sell it.

SPEAKER_00

So rather than expect this like uh holistic or macro impact, start to identify certain areas or departments where it's obvious AI can have a faster impact here. And perhaps as as a leader of a company, maybe you target those areas first to satisfy the demands of the board, to fill the, you know, to fulfill the the promise of AI, but simultaneously you're also investing in that longer-term organizational change, but you're still showing wins in the meantime.

SPEAKER_01

Yeah, you're showing wins. I think that's a big thing. And you're giving them a sense of what a win is going to look like. Yeah and some of it is is going to um require telling them the stories from other places too. Yeah. Uh right now the problem is it's very hard to find stories. Yep. You know, you you get uh you get a claim, you know, that so-and-so reduce productivity or equ headcount by XYZ. And you say, Can you show me that? And yeah, it turns out, you know, it's yeah. So, but a real story where you could walk them through it. You know, back to the another point, just maybe this is where you were headed earlier, too, is to think about who leads this thing. We are seeing a lot of discussion about putting the IT and the HR people together.

SPEAKER_00

Interesting.

SPEAKER_01

And I I here here's the the right motivation for that. And that is places that are talking about that understand this is ultimately on the white-collar side about work reorganization. And some part of that is about the employee side. Um I don't think you're gonna get very far merging those two groups, though, because the IT people don't know anything about human resources who don't know anything about IT. Right. Putting them together is gonna get chaos, right? Yeah. Um But figuring out how to get those two domains to work together is absolutely crucial. And I think the way to do that is at the person level pair rather than some structural change.

SPEAKER_00

Yeah. You know, so so kind of we focus in at least in our business and what we teach, we focus on generative AI application. We don't we don't really get into the analytical AI that much. We don't see it necessarily, at least at the gener generative AI level, we don't see it as an IT role. We see it as somebody that understands maybe it's an H maybe it's an organizational uh optimized, you know, uh expert who is seeing um, you know, what what people are doing, because they're hiring for them, they're describing the jobs, they're analyzing the jobs, but they also see it through the lens of what's pop what's possible today with generative AI. What are your thoughts about the the AI discussion not necessarily even being an IT discuss at the generative level, not necessarily being an AI, I mean an IT discussion.

SPEAKER_01

Yeah, I think you know, I think it's a case where it would take too long to develop people who understand both well enough to do this by themselves. Yeah. So I think it is about putting them together. Aaron Ross Powell, Jr. And on the HR side, you know, there is there's this quirky element of what they were calling business partners, and that means having an HR person at each location or for each work group or something like that. I think the engineering efficiency folks always thought that that was wasteful, and it might be. Um but if you've got that person, that's half your partnership right there. Yeah, right. That's a person who is in our plant, knows something about the work there, and knows a lot about the individual people and has a feel for the culture and all that kind of stuff stuff. If you partner that person up with an IT person who understands enough about various types of AI you might be using, which might be generative but more likely machine learning stuff, and you partner those two people up, and maybe you bring in a consultant to walk them through uh a couple of jobs, and then you let them go and see what they can do, you know. I I would say that's the that's the seems to be the most cost-effective bet, I'd say.

SPEAKER_00

So let me ask, uh you know, I saw a a news story the other day. I think it was IBM that um had a uh significant cut that they announced, with the bulk of those people being in their HR department. Are you familiar with this that happened at IBM?

SPEAKER_01

Uh, I didn't see it per se, but it wouldn't surprise me that it's often the case. It's sort of like saying to the investment community, we're cutting middle management. It's something they always like because they don't think middle management does anything. Yeah. Uh and cutting HR just sounds like you're cutting infrastructure. So uh I I think though that if you're in the HR function, you really need to be able to show to people that you are to leaders, that you're doing something that supports the business in a tangible way in something which is important for the future and is not just supporting the people's side, which frankly they don't get. I mean, leaders don't get, boards don't get, they don't understand turnover costs. A lot of that's a fault of HR, just not showing them what these numbers look like. Yeah. But nevertheless, they don't get it, right? So it is super in the interest of HR to get onto this train. Um so I I think if they're paying attention, they they'd be supportive of it. You know, whether you can get the IT departments to get on and partner up would be interesting. But I think, you know, if the CEO tells them to love that idea, right?

SPEAKER_00

Yeah. Peter, you know, this is uh like I'm fascinated by this because I'm not I'm not the I'm not a high empath, let's say. I I'm looking at the numbers and optimization and things like that. And I don't think about the things that we've talked about today as much as I realize I should be thinking about that. So I want to just say like for anybody that is is a business owner, if you haven't kind of come to the conclusion that we need the people that understand the people involved in this AI discussion, it is not just the it's not just the number cruncher, it's not just the the guys in the IT department, that it really we we need to have uh continue to introduce or make sure that the humanistic side is still part of this conversation. Um, I think that you will uh wish that you had a later date. So, Peter, I know that you've got um an update. Is it an update to your book or is it a brand new book that's coming out?

SPEAKER_01

Uh it's an update. Well, no, it's a brand new book. Uh it started out, they just wanted me to update the old book. This was about remote work.

SPEAKER_02

Yes.

SPEAKER_01

Um and then there's so much that's new since I wrote the first one, which was mainly looking at the evidence that it existed before the pandemic, which had been studied more than people seem to know. Uh, and looking at what the problems were. That was the first book. You know, what are we gonna have to wrestle with? And now we've got some better sense of how to what we have to wrestle with and what's not going quite so well. So I think the the book uh uh is really about um uh recognizing the importance of human interaction, which has kind of got lost in the conversation, you know, particularly the view that you could just do everything by yourself and what could be wrong with that. And now we see all the things that are wrong with that. You know, new employees are lost, um employees are not cooperating as much as they should with each other. Um, you know, the things that happen in the office, people helping each other, getting quick answers to problems, working out problems before they escalate into something big because it's face-to-face, and you know, occasionally maybe getting good ideas from each other, all that stuff is not happening the way it should.

SPEAKER_00

So when does the story when is it when is the book coming out?

SPEAKER_01

And if you don't mind sharing, at the end of September, it's called In Praise of the Office. My co-author is Ranya Nemi. Uh and we begin with that, you know, a hundred years of research showing that interactions with other people matter. And then the question is does that mean everybody has to come back to the office? It doesn't mean that. Um it does mean though management has to be purposeful. If you want to be hybrid, uh you could do that, but it's gonna require some work. So it's very much like our conversation about implementing AI. It's gonna require some work.

unknown

Yeah.

SPEAKER_01

And the change in this case is not that hard. Things like, for example, why is hard hybrid not working so well? The first and most obvious reason is people are just not showing up. So they have anchor days and people don't come. And the attendance is terrible, you know? I mean, really terrible. Um, why is that? Well, it's partly because top leaders have not made it a priority. They pushed it down to the lowest level, thinking that's the solution to most problems, just let them decide based on what suits them. But then the problem is each supervisor, they don't want to be the bad guy. They look around their employees, say, well, Bob's people don't have to come in every yeah. Okay, fine, come in when you want, right? Yeah. And then when they do come in, they discover nobody else is there, so then they don't bother coming in and nobody punishes them. So, you know, attendance is terrible. And so they're not getting any interaction. So the theory, fine, execution, terrible. And the other simple problem is meetings are terrible, right? We spend way more time on virtual meetings. Um, the meetings are bigger, they go on longer than they should. People are typically off-camera, they're doing other stuff, so they're not paying attention. And increasingly, I'm hearing they don't actually even log in. They just have their agent log in and take notes. So why are we having these meetings in the first place? You know, so those are problems which are just right there in your face. You could solve them with a stroke of a pen, but it's got to be somebody up high in the organization who does it. And they're so far they haven't paid attention. So Peter, I feel like we'll see the bridge of that.

SPEAKER_00

Uh that's fantastic. I I if it's got anything to do with uh some of the stuff we've talked about today, I'm eager to uh dive into it. You know, um if for people who want to, it's obvious that we really just scratch the surface. If they want to get more insights into how you're looking at things and how you're answering these questions that we're all having, um, where are you putting out content outside of your books? Is there a place where they can go and and kind of find your thought?

SPEAKER_01

Uh well, I have this this article we described about what Rico is doing, trying to automate simple tasks. That's in the Harvard Business Review, so it's on the website there. Pretty easy to find that. Uh, we have a paper where we talked about the basic uh issues and and limitations of large language models on things like the fact that you can't replicate studies, so that means that you know, if you get a report that you don't like, you just ask a different large language model, you get a different report. And now if you're a leader, you you got competing reports, and how do you do it's all about you know you gotta set some rules about how they get used and things and that was in the Sloan Management Review about a year or so ago. Um but I we tend not to put a lot of stuff uh just floating until we publish it. Uh but I write a column for HR Executive Magazine in particular where I've written about some of these uh issues, so you can find that stuff there. It's pretty easy to find me. I'm the only Capelli two P's, two L's. Believe me, there's not a ton of them in the world of this kind of stuff, so it's pretty easy to find me.

SPEAKER_00

Well, Peter, I really appreciate your time. This has been um Oh, my pleasure. You know, a lot of times we talk about tactical things and and like specific to using the tools, but this opportunity to really ask some questions that I've had about like how how do we look at what's gonna happen and the the like how do we get it to stick, I think that I've gotten a lot of answers from this. So that's great. We're gonna have a link to the code. We learned a lot from this too.

SPEAKER_01

This is very helpful for me too, Chris. I'm gonna have my agent come and listen to your podcast here. So I'm I learned what we're doing.

SPEAKER_00

Honestly we see anytime that like I'm I'm in the weeds, right? Like I'm in the trenches with these not necessarily at the enterprise level, but um I'm having this conversation all the time. So if I can ever be a resource for or perspective or what do you think about this, I'd I'd I'd great thank you. I'd love to be of service. So good.

SPEAKER_01

Thank you, and you you'll hear from me. Awesome.

SPEAKER_00

Okay, everybody, thank you so much for uh being on the podcast today. And uh, Peter, we'll be um, I'm sure we'll be in touch. Thanks, everybody. Thank you. Thanks for tuning in to using AI at work. Don't forget to subscribe for more conversations about how to use AI at work. And a special thank you to our sponsor, Chief AI Officer, for empowering businesses with AI education and training. Visit their website for free AI readiness assessment and AI strategy guide to help you get started using AI at work. It's www.chiefaiofficer.com. Follow us on Twitter at the handle usingAI at work, and visit www.usingai at work.com for free resources to help you harness AI in your role.