Why AI Spend Isn’t Paying Off

David 0:01

Companies are spending heavily on artificial intelligence. They're buying licenses, launching pilots, and training employees, yet many still aren't seeing meaningful business results. So, what if the busiest obstacle isn't the technology, but the fear, distrust, and professional anxiety it creates in the people who are expected to use it? And what must leaders do differently if they want to turn resistance into results? We'll find out on this episode of Shift Shapers.

What Shift Shapers Is About

Announcer 0:32

Change either energizes or paralyzes. The choice is yours. This is the Shift Shapers Podcast, bringing the employee benefits industry interviews with individuals and companies who are shaping the industry's shift. And now, here's your host, David Saltzman.

Meet Dr Gleb Tsabersky

David 0:52

And today we welcome back to the podcast an old friend, Dr. Gleb Tzabersky, a behavioral scientist, consultant, and best-selling author who helps organizations navigate change and make better decisions, which we could all use. His latest book is The Psychology of AI Adoption at Work, From Resistance to Results. What a great title. Dr. Gleb, welcome back to the Shift Shapers Podcast.

SPEAKER_01 1:15

Thank you, David. Appreciate it. And uh I'm glad my publisher at Georgetown University Press was happy with it. So yeah, good.

David 1:22

Good, good, good. And it's available by the time you listen to this podcast, it will be available every place you get your books. The official launch date, I think it's September 1, but um this will air after that. So go get it because it's a terrific book. And we'll give you a little bit of insight from it in the next 30 minutes or so. So let's start here.

Why AI Change Management Breaks

David 1:40

Everyone has a shift shapers moment, that moment that something happened that changed how they see the world. What was yours when it came to understanding AI adoption?

SPEAKER_01 1:52

So I think it was when I saw companies rolling out AI in the way that they roll out previous technology. And that's the problem with companies doing it. They're using their traditional methods and approaches to change management to approach AI. And it just doesn't work very well. That's is a big issue. So thinking about previous technology, let's say you're rolling out wellness benefit software or HR management software or a CRM or something like that, sales enablement software. The challenge with that software is that it's difficult to learn a new software tool that you're attached to your existing habits and ways of doing things. Even if it's you're you switching from various Excel spreadsheets to looking at CRM or whatever else you might be doing, or various Excel spreadsheets to looking at your HR management, new now HR management software, wellness benefits software, all of that stuff. It's difficult to learn and people have are stressed out about learning new things. So leaders thinking that are addressing those problems. But that's not the real problem with AI. There are different problems with AI tools. AI tools are surprisingly easy to learn once you're motivated to learn them. Once you have the motivation, you can talk to them. You can ask them, how do I do things? How do I change my HR policy? How do I look at wellness benefits? How do I do whatever you want to do? How do I draft an email? Everything about your workflows. It's a very flexible tool. But there are three big emotional blockers to AI tool usage. And so that's when I realized when I saw those blockers, and when I talked to and I saw how leaders were trying to roll it out, just giving everyone a subscription to Claude, encouraging them to use it to say, hey, take the education, time that you need. That's not the approach that you need for AI tools. And I can talk about the approach that you need for AI tools, but just answering your specific question, that was the moment.

David 3:58

Was there a particular client experience or a failed pilot that kind of crystallized the problem for you and said, ah, AI resistance requires its own book?

SPEAKER_01 4:08

Hmm. Interesting that you asked that. So it wasn't a specific pilot per se, but it was a research and experience with just a number of clients and just findings from peer-reviewed sources. So, for example, there was a study from MIT showing that 95% of AI pilots are not showing the returns on investment that compared to the resources that were put into those. So I'm not simply looking at my experience. It's a Georgetown University Press, it's Georgetown University, it's a peer-reviewed book, it's looking at research. And I'm a very much a research-oriented person. So I definitely look at my own experience, and the reason they published it is I did over 100 consulting projects with hundreds of focus groups, many hundreds of responses, hundreds of surveys, many thousands of responses to surveys. So there's a lot of original research in the book. But I'll take, try to take a look at the research and respond to the research. So what the research finds is that people are really resistant to AI for not typical reasons.

Three Emotional Blockers To Adoption

SPEAKER_01 5:12

And let's talk about those three reasons just so that folks understand what I'm talking about. So three big blockers. One, you already talked about the beginning, the fear and anxiety. People aren't afraid of looking up HR policies and wellness benefits in a new database versus using various Excel spreadsheets and folders on their computers, right? That's not something that's a fear. But people are afraid of being replaced by an AI tool and that by using it, they're training it to replace themselves. And so in chapter six of my book, The Psychology of AI Adoption at Work, I that's the group that I call psychographic profile that I call the AI alarmists. So that's the AI alarmist, psychographic profile. So that's one that feels fear and anxiety. That's for them the biggest challenge. There's a second group. And this second group, for them, the biggest challenge is identity threat. People aren't threatened to their professional identity by looking up information, like I said, in a new database versus in various folders in their computers or Excel spreadsheets. But they're threatened by a tool that can write a great HR policy. They're threatened by a tool that can they can chat to and they can find out information about their employees super quickly, much more quickly than they can find themselves. They are threatened by a tool that can write a great sales email, that can write a great article, that can create a great image. That presents a threat to who they are as professionals, their professional identity. And so many of us get meaning from our work. Others get meaning from other things, but so many of us get meaning from our work that for them it's a really serious blow to their sense of meaning and just for who they are as professionals. And so that's a group I call the pragmatic resistors. For them, the job issue is secondary. It might be an issue, but it's secondary. For them, the biggest issue is the identity threat. And so those two groups are the ones that don't adopt AI on an individual level. They're resistant to it. Now there's a third group that does adopt AI, but they don't want to talk about it. I call that group the psychographic profile in chapter six. That's the reluctant adopters. What they feel is shame. So shame is the big emotional blocker for them. Again, we don't feel a sense of shame around adopting previous technology. But people feel a sense of shame about saying, hey, did you know that I used AI to write my new HR policy? Did you know that I'm using AI to respond to people who ask questions about the wellness benefits? Did you know that I'm using AI to draft my emails, to create articles, to do my financial reporting, to do my the to do, I did a training for tax preparation firm to like do the tax prep for a client. They don't want to talk about that. And they don't want to admit it to their colleagues and to their supervisor. And so what happens is that they use AI, they get a lot more done, but that usage is not translated into bottom line results because people spend their additional time just going on Facebook or something like that. And so that's the third group. So that's the third group. I call that group the reluctant adopters. And so those are the three big groupings that I find in my research and that are really blocking effective AI adoption. And that's why you have that statistic of 95% of AI palettes don't show return on investment. They don't, because these groups are not using it effectively, they are not using it. That's the first two groups, and the last group is not using it in a way that shows up in bottom line team and company results.

Middle Managers And Quiet AI Use

David 8:54

That's fascinating. So as you look across those different psychographic profiles, do you find a difference between employees and middle managers? Absolutely.

SPEAKER_01 9:04

So you with middle managers, you have a lot fewer AI alarmists. So you have people who are skeptical about AI tools, who think that, oh, I know better and so on. But you don't have as many people who are worried about being laid off due to AI tools. So you do have some, but they're much smaller profile. You do have some who are worried about threats to their professional identity by using AI tools, but you have many fewer of those who are middle managers. But the reluctant adopters, I definitely say still many of those among middle managers. And I'll give you an example. And this actually goes to the very top. So not simply middle managers. Even executives find themselves among reluctant adopters. I was giving a training for a vistage group. And this is a group of peer executives from middle market companies. So this is eight people in Pittsburgh who are leaders of companies ranging from 50 to a thousand people. And so the executive was telling me how, and this was on the larger side, how just a few days ago he had his VP of sales do their annual presentation on their biggest client of the year. And the biggest client of the past year. And they do this, like I said, every year, and they talk about how they got the client, how they approached him, all the case studies, everything, what the implications are. So a really in-depth case study on this. Now, previous years, so after the presentation, which went great. Like all the this is a presentation to the other executives. The other executives loved it, it was great. After the presentation, the VP of Sales approached the CEO privately and said, Hey, I just want to share something with you. You know, it usually takes me about three days to do this. I work with my executive assistant, it takes us a whole bunch of time. Three days pull together all the data, create a PowerPoint, and so on. This time, I just delegated the whole thing to my executive assistant, and it took him an hour and a half using AI tools. But I don't want you to tell any of the other executives because they would think I'm cheating and I'm lazy. So don't tell them. And so this is an example at the very top of the leadership team of how the they are among the reluctant adopters. And I see middle managers definitely among reluctant adopters, as well as rank and file team members. So that's where you'll find a number of middle managers among the reluctant adopters.

David 11:39

If

When Mandates Create AI Work Slop

David 11:40

we look at that same kind of a question from the other end of the periscope, the word resistance sometimes can make employees sound irrational or obstructive. When is that reluctance towards AI actually kind of a reasonable response from an employee to poor leadership or unclear incentives or genuine risk? So quick question.

SPEAKER_01 12:02

I think it's incumbent upon leaders to make it clear to employees and set up a culture that where the employees are incentivized to use AI. So if employees, if I come into a company and I observe employees being resistant, I don't blame the employees. It's the leaders who are responsible for ensuring that you have an effective approach to AI adoption. And so leaders, they don't know how to do things. That's why I wrote my book, The Psychology of AI Adoption at Work from Resistance to Results. They might not know it, but it's incumbent on them. Oh, they know it's incumbent on them, but they might not know how to do it. So the problem is, as I talked about at the beginning, they're using their old methodologies, what they're used to, to try to solve the problem of resistance to change for AI tools. And that just doesn't work very well. So for example, if you try to force it down people's throats and say, you must use this tool, what happens often is what's called AI work slot. And so this is something that Harvard Business Review recently published an article about based on research from BetterRup. People, on average, spend about two hours a month, according to the responses from this research, about two hours a month correcting bad work that's generated by AI that's sent by their colleagues to them. And we're not even talking about what's sent outside the company, just internal work. And $186 correcting all these problems per month per employee. So it definitely adds up fast. But why is it their sloppy work? Well, if you're forcing AI down people's throats, what happens is that they will not use it well. And they will just get the initial result from generative AI, not give it the right prompt with the right context, and they will get the results and they will just send it onward. And they'll say, Well, you don't like it? Well, you told me to use AI. I'm doing what you told me to do. And so this is what's called malicious compliance. And so that malicious compliance is what you get when you use typical methods to try to overcome that change management resistance. So that's a big problem, right? And so this is where leaders need to learn and understand. I talk about that in chapters seven, eight, and nine of my book, about how to address each of these demographic profiles and how to effectively address the underlying emotions that are causing all these problems.

David 14:39

Could you go into a little bit more depth? That was kind of the question that was on my mind is if I'm a leader and I need to talk to the alarmist or one of the other profiles, how do I tailor my speech and my direction to achieve the kind of results that we really want in my company?

How To Lead Each Resistance Type

SPEAKER_01 14:55

So, yeah, so let's talk about each one. And you have to have a different approach to each one. So, for the first, for the AI alarmists, and this is something that the large majority of companies can do, you can should make a commitment that if AI adoption improves efficiency of the company, that you'll focus on growth rather than job cutting. So that is a commitment, again, that many companies can do, most companies can do, because you can take market share away from companies that aren't using AI. And so that's definitely something that can be done. And there's research supporting this. So research from Stanford showed that this is again published research, showed that companies that are adopting AI effectively in their industry compared to companies that aren't adopting AI effectively, they're growing their headcount by 6.5%. But they're also growing the revenue by 9%. So they're growing the revenue faster than they're growing their headcount. They're being more productive, more efficient, and they're doing more things with fewer people than they would have needed if they were just growing purely using the same methods that they were doing before. So AI is enabling them to grow faster, be more productive, be more efficient, need less people to accomplish the growth, but they're growing and they're hiring people. So the jobs that are lost are companies that aren't adopting AI. And those companies are the ones that are losing jobs, not the companies that are adopting AI. And so, and this is in the vast majority of cases, of course, you have some tech firms that are we're not talking about here, but this is what the research shows in the vast majority of firms. So that's one. The second is something that all companies can do, and say that look, the people who will have a much more secure role here are the ones who learned how to use AI effectively. Because if you learned how to use AI effectively, you'll be much more productive. And of course, we'll keep you around because you're producing. And the people who are not using AI effectively, they're producing much less and much worse than before. Kind of the much worse than you are. So of course we'll keep you around. So this is both a company-wide approach and an individual approach. So creating that dynamic. You need to flip the script essentially for the AI alarmist and give them the understanding that, hey, it's not that you're training some a tool to replace you. Basically, you will be replaced if you're not going to use AI effectively. So that's for the AI alarmists. So when you're communicating to them, that's a matter of leadership communication and policy shift. So those two policies that I talked about and leadership communication about them. Now, so that's one. That's about the AI alarmist, the fear and anxiety. For the second group, the identity friend. Something that I see time and again and again, this is the typical approach to adapting technology, is that leaders try to focus on the most productive use case. The most productive use case. So let's say in tax preparation firm that I was talking about, the initial desire was to focus on the actual tax preparation process and the for HR policy, drafting the policies, for let's say for article sales outreach, writing the emails. Now, the thing is, that's what really causes people to be triggered by an identity threat. That's what truly threatens their identity. So instead, when I work with firms, what we do is we do a survey and focus groups and ask them, hey, what are the tasks that you hate the most? What would you like to get off your plate? Where do you have to look up information across a whole variety of systems and find the information and create a packet of information that enables you to do what you really want to do, what you really like to do. So those are the things then that we create prompts and agents that are going to be useful for people that they can build and they can customize and they can use for their needs. And so you create those and you train people on using those things and doing those things that help them get the time in their day to do the things that they're passionate about and the things that human judgment is best at. And so you essentially go slower in terms of less productive in the short run to go fast. So go slow to go fast. You adopt it with those human orientations, a human-centric adoption, not taking away the things that people love to do, but taking but removing the things that people hate to do. And that's great. So that's really helps people with the identity threat, the pragmatic resistors, actually start using AI. So that's the second group. Now for the third group, the reluctant adopters, there are three things that need to be done. One is leadership model. So I told that CEO that what he needs to do after the training that I gave them is go back to his team and tell the VP of sales to send out an email to all the executives and brag about how he used to take three days to do this presentation. But now he only did takes one and a half hour, and that's not even him. Like the vast majority of work was done by the executive assistant. And tell his team members, like the people who are in his team, and talk about this openly, broadly to everyone and model the thing that you want everyone to do. So that's one. Second, we talked about the training. So you need to train people on the things that they don't want to do and the things that are annoying. And then at least you get that consensus from everyone. You change the social norms that, hey, there's a bunch of things now that we're going to be using AI to do after a training and after everyone has a mutual shared expectation. And so that changes a social norm, at least around those things, and addresses the social stigma. Great. Now, people, of course, will start using it for other things because it's so efficient and effective over time as they learn how to use it. And the you to facilitate that, you need to set up a channel, ideally synchronous, but you can always set it up asynchronously to communicate about new AI tools that you build, and then praise and reward people who build tools that are especially effective. So an asynchronous channel is easy to set up. It's a Teams channel or a Slack channel or a Trello card where people just talk about hey, here's an AI tool that I developed that's really cool and beneficial, here's a prank, here's an agent. And then the leader needs to reward and praise them, and including for tools that other people can use and that save them time, financial rewards, so some kind of bonus, some kind of incentive. Ideally, you'd have a synchronous meeting as well to talk about this, but I know people don't want necessarily aren't comfortable with more meetings. So integrate into existing meetings, but at least have a Slack channel or a Microsoft Teams channel. Channel about this. And so you get that reward and praise for tools that are effective, and you expand the sphere of what AI is now used for. And so those reluctant adopters increasingly come out into the open and they use AI tools, and that helps address their concerns. So that's the free demographic profiles. Talk about in chapter six, and chapter seven, eight, nine talks about how to address all of those issues.

Finding Champions And Shadow AI

David 22:28

So if if visible peer champions, let's call them, are probably a better way to go than these top-down mandates, how do you find and develop the right champions? In your example, it came out of a meeting where a vice president said to their boss, look, I don't want to talk about this, but how do you go and find those people and bring them to the fore?

SPEAKER_01 22:50

So you want to do a survey. One of the surveys and focus group questions is what are you using AI for already? And you want to do that anonymously. So people, because there's a phenomenon called shadow AI usage, and people feel like, oh, if I reveal how I'm using AI in a public way, then people, then I will be punished in some way. And definitely you have that. I was remembering in one focus group in a law firm where a young lawyer talked about, okay, I'm using AI in this way, in this way, in this way. And one of the older lawyers was like, Well, I'm surprised that you're using AI in all of these ways. We don't even have an AI policy. You know, we should talk quickly about like, you know, leadership team sending an email telling everyone not to use AI. And I'm like, this is typical and highly counterproductive, you know, in terms of what would actually get people to adopt AI effectively and use it successfully. And so what you want to do is figure out in these surveys, like who's using AI. Like in that survey for the law firm, 76% of people were not using AI, but that means that 24% of people were using AI. And so you want to figure out what they're using it for, what what are the case studies. Focus groups are especially helpful because then you can approach people individually afterward and talk to them about what they're doing and whether they'd be willing to do so publicly. But basically, even after if you even if you only do surveys, you need to identify, okay, here are the ways that people are using AI, and then you need to publicize that and establish policies saying, like, yes, great, please use AI for these things. We will reward and praise people who will be using AI for these things. We encourage them to come forward, talk to their managers about how they're using AI, and be role models and leaders for the rest of the organization. We want to set up an AI champions group to do that. So that's kind of like the practical ways that you would actually accomplish these goals.

David 24:51

One of the things that I think executives talk about a lot, and you just alluded

Smart Guardrails And Editing Skills

David 24:55

to it in the law firm example, is having guardrails. Oh my gosh, we don't have an AI policy. We need an AI policy. And someone inevitably says, well, if we put too many guardrails out there, we're going to slow down adoption. But that's not the case, is it?

SPEAKER_01 25:10

So if you put too many guardrails, yes, you will. And but you need to put the appropriate guardrails that would still allow innovation. So you need to, again, going back to that example, go slow to go fast, it's good to slow things down in order to make appropriate guardrails. There have been new stories about, let's say we're talking about lawyers using AI inappropriately and hallucinating uh cases that were not actually in the record. So you definitely want guard trails, but you want to be smart with your guardrails. You want to see, okay, it's not a problem if you use AI to draft legal work. It's a problem if you don't check it afterward. If you don't check it, that's the problem. So what are the things that we want to do to make sure that AI can be used? So just slamming down guardrail and saying no AI legal drafts because somebody handed in a court a legal brief that had some hallucinations and the lawyer got penalized for it, that's not the right step. The right step is okay, we don't want to be stupid, but we all want to be productive and effective. So how do we have guardrails that aren't stupid, but effective, highly effective guardrails? So for example, the law firm before I uh like the initial desire of that law firm in the guardrails was to check with, make sure that if you are like a basically check with a senior partner on the case before any time that you use AI. That would have been a stupid guardrail, I'm sorry to say, because then it would have completely prevented people from effectively using AI. Or like if you need to take the time to check with a senior person who really doesn't know, I mean it's junior people who are adopting AI. He doesn't know, or she doesn't know about like how you can use AI tools. They're like, well, no, don't use AI tools, or let me think about it and get back to you. And like, who's going to use AI tools if that's the the thing the thing that you hear all the time? And like, you know, maybe the person says in the week that yes, they can use an AI tool. Okay, sure. That's helpful, right? No, of course not. So you need to have guardrails that are smart. Smart, effective guardrails, where say you're responsible for the output. You need to check it very thoroughly. You need to, if you're not sure, talk to someone senior to you. Make sure that it's not going to be AI work slab. Talk about what AI works lab means, and have the output be carefully checked and monitored and guarded. But your skills, the skill that people need to develop in the future, increasingly, is the skill of the future, is not generating the initial draft of the legal brief or the initial draft of the HR policy or whatever it might be. You need to develop the skill of being able to evaluate and edit that. And so that is the skill of the future, being able to evaluate and edit the outputs of AI tools. And so that is where the guardrails need to be to facilitate the skill development and the responsibility for people evaluating those outputs.

Monday Morning Plan For Leaders

David 28:26

So suppose a leader who's listening to us have our conversation today bought the tools, ran the training, but still is only seeing kind of scattered adoption. What should that person do on Monday morning?

SPEAKER_01 28:39

I think they need to communicate about the job anxiety and the identity threat. So that is my fundamental uh issue. If they're seeing scattered adoption, then they probably have not done a good job of communicating and establishing policies that address the AI alarmists. So again, the job anxiety, committing to growth rather than cost cutting, communicating about, hey, we're your job is going to be much more secure and safe if you use AI tools here. And then, hey, we're going to use AI tools, focus using them on the activities that are the most annoying, the most frustrating. So go slow to go fast. Don't try to push people to do things that they love. So those are the things that most likely are happening. I don't think it's going to be the matter of the reluctant adopters once you're on the training. You should see them flip pretty quickly. And I've seen them flip pretty quickly. But it's those two groups that if you don't approach them effectively and address their concerns, that's going to be the problem that you're going to be facing.

David 29:41

You know, it's a fascinating subject, and there's so much more in the book. Um, I urge anybody who's running a company, who's got people who are starting to use AI or dip their toe in. The book is called The Psychology of AI Adoption at Work, From Resistance to Results. And the author is our friend Dr. Gleb Tapersky.

Book Resources And Free Sample

David 30:00

Gleb, thanks so much for coming back and joining us on the Shift Shapers podcast.

SPEAKER_01 30:04

Well, you're welcome, David. And I want to mention that if anyone wants a free sample of the book, they can go to disasteravoidance experts.com forward slash AI book and put their information into that. They'll get a free sample of the book, the introduction and the chapter. So again, that's disasteravoidance experts.com forward slash AI book. If you already bought the book, then you can get a free manual and on the seven critical mistakes that leaders make in AI adoption and an assessment on the best means for AI adoption for your company. Again, the same website, disasteravoidance experts.com forward slash AI book. And we'll be sure to put that in the show notes.

David 30:46

Gleb, thanks again. Great book. Thank you, David.

Announcer 30:50

The Shift Shapers Podcast is a production of Shift Shaper strategies and may not be reproduced or quoted in whole or in part without our express written permission. Copyright 2020, all rights reserved.