AI Margins and MSP Growth: Dr. Gleb Tsipursky on Passing Savings vs. Competing Away Profit

[00:00:00] My guest today is Dr. Gleb Tsipursky. He runs a future of work consultancy called Disaster Avoidance Experts. He has a new book out from Georgetown University Press, The Psychology of AI Adoption at Work. The argument is that AI projects stall for human reasons rather than technical ones. I reviewed the whole thing and notes included. I want to spend our time on whether it holds the size of company most of my audience usually runs. Gleb, welcome to the show.

[00:00:30] Dr. Gleb, Thank you, Dave. I appreciate you being on the show. I appreciate you inviting me. Well, let's dive right in. Your smallest case study is a firm with under just 100 staff or so. And to be fair, the words MSP and managed services don't appear anywhere in your book. So I've got listeners with 11 employees, like no HR, no CIO, no learning and development budget, and a technician shortage. Like, so when you think about it in that perspective, what in your book is focused and helps and survives for the future?

[00:01:00] Yeah, I just did a training for that group. Yeah, I just did a training for an eight-people company. So very much in the MSP space and they outsource their IT. So they have an MSP as one of the ways that they manage things. They don't have enough people for an IT budget. So the same principles very much apply to folks who are an eight-people company and people who are an 100-people company and people who are an 1,000-people company.

[00:01:28] When you get beyond 5,000, that's more like enterprise level. And that's kind of when there's more difficulty and when at least some of the things from my book would need to be tweaked. And I do have case studies of the larger companies. But for everything that I talk about, you can apply it to quite small companies that get MSP services.

[00:01:52] Okay. Now, in the book, the company that I thought looked the most like my audience is ImageQuest, right? So they're 50-some people, IT and security. So in that case, he doesn't get a case study. He gets some quotes. And what the owner there describes is a 45-minute optional monthly lunch where everyone shows what they're doing. It's not necessarily profiles. They're the five stages. Is that enough at a 50-person company? What do you recommend for those smaller sizes?

[00:02:20] Sure. So he described what he was doing afterward. So after the initial adoption. But this for MSP service providers, whether you are a 50-people company or a five-people company, what you really need to do is start with a training, a thorough training on AI adoption.

[00:02:41] Because, as you and I said before the beginning of the episode, MSP service providers are not necessarily expert in AI. And here's the reason. They are experts in technical tools and technology. So there are experts in, let's say, Microsoft products.

[00:02:59] They're experts in various forms of technical support, technological support. But AI tools are not necessarily the grassroots ground space of IT folks because they use natural language programming. So natural language programming is very different than the kind of technical work that MSP service providers are used to.

[00:03:22] So you build tools within AI using natural language programming. And so people need to be trained on that. People who are in MSP need to be trained on how to use natural language programming. The mindset is very different than working with typical technology tools. And that's some of the reasons why AI adoption projects fail.

[00:03:46] There's 95% of AI pilots fail to scale. They fail to show return on investment, according to a recent MIT study. And that's a big concern. And of course, a number of those failures have happened with MSP support as well as with internal IT support.

[00:04:03] And the failures happen because of human psychology and because MSPs are not used to providing the kind of support for natural language services that AI tools require. And they're not used to the kind of resistance with which AI tools are met. Okay, I'm following and tracking. Now, one of the things I really wanted to dive into is I loved your chapter 13 about measuring and optimizing.

[00:04:29] That's an area that oftentimes we talk about in MSPs because we're quite good at measuring and look for a lot of metrics. Now, the chapter gave guidance to track cost reductions and correlate to operating costs and margins. So, I'm following along. But one of the things I was curious about is it never once names a cost that the organization pays. Things like licenses or per seat or token spend or consultant fees, the non-billable training hour.

[00:04:57] And it doesn't necessarily latch to a formula to put them in. And if I'm thinking about myself as, say, a 20-person MSP doing things like $30 a seat a month plus fees plus 40 hours of non-billable training, like walk me through the arithmetic on both sides that I should be measuring and optimizing. Sure. So, there was a recent useful study put out by Upwork finding, so again, that's the freelancer platform,

[00:05:25] Upwork and research study showing that compared to how much time a freelancer would spend, so like a dollar a freelancer spent, it's covered by about three cents of tokens. So, on average. So, if you're thinking about exchanging the actual work for the spend that's after the implementation, you are spending a whole lot less after the spend.

[00:05:53] So, just kind of like what the research shows on average. Now, in terms of the investments, of course, you have to think about all of these issues. How much are you investing into it? What's the billable spend? Like, how much are you investing in training time, consulting time? How much are you investing into the seed? Because you're really not, unless you're doing development work, which your clients are not going to be doing, you might be doing it, you're not going to be investing in tokens.

[00:06:23] Realistically, for the audience, this is one of the differences in kind of large firms that I work with and that I talk to them about, I don't focus on the IT for them because they have their own IT departments that figure out this stuff. I focus just on the psychology for them. And that's my area of expertise for large firms. And so, they need to worry about token spend. For MSPs, you don't really need to worry about token spend because you're just going to get the license seat per hour. Let's just be honest.

[00:06:53] The kind of clients that you're dealing with, they're overwhelmingly not developers. And so, they don't need to worry about tokens. They need to just have a $20 per seat license for Claude or ChatGPT or the $30 for Microsoft Copilot if they're on the conservative side. So, that's what I typically see with MSPs need to think about. So, that seat is the one topic.

[00:07:17] Then, of course, you can either spend your sweat equity into learning the tool and you don't need an outside consultant. So, when I do a training, I tell them very straightforwardly that I can cut your time learning by 5x or you can put this 5x into learning the tool. And so, you can do that. And that's what I typically see on average, how much longer it takes without the initial training.

[00:07:45] And so, you can just figure it out by yourself. The nice thing about natural language programming is that you can ask the tool how to use it. And it will tell you how to use it. It's great. You don't need to, like, necessarily get paid for external courses. And so, you can think about those as the spend.

[00:08:04] And then, in terms of the output, I told you the kind of average metric in terms of what you can expect to save on that amount of time for somebody's time. For, like, three cents compared to a dollar. And that's a really objective metric because when you're looking at how much time a freelancer would take to do this task, that's quite a reasonable metric to use on average.

[00:08:30] And it doesn't necessarily mean it applies to your company, but that's a good baseline to go with. And so, you want to see how many tasks can we replace by using AI tools and how much time will that save? Well, that's an average going to be 97% less cost for that task. We'll be right back after this message. This episode is supported by Halo. Automation is becoming a defining characteristic of modern managed services.

[00:08:58] But automation only works if the core platform supports it. Halo PSA gives service providers the flexibility to build powerful workflows, integrate automation tools, and design service processes around how their business actually runs. For MSPs building a more automation-driven operation, Halo PSA is one of the platforms increasingly showing up in those conversations. Learn more at usehalo.com.

[00:09:29] And we're back. Now, you wrote in the book that time saved alone doesn't assure a strategic win, right? And you cite the finding that 36% of managers squandered more than half of their freed hours. Now, if I apply this to MSPs, saved technician hours are worth nothing to me unless I can resell them, redeploy them to backlog, or take them off the payroll, right? And then the other thing you recommended, though, is you recommended a headcount neutral pledge.

[00:09:57] So kind of reconcile that for me. Like, which of the three is supposed to happen then? Happy to. So what I typically do for a company... So let's talk about the blockers to AI adoption. One of the biggest blockers is not the typical things that MSPs and other folks associate with change management or technology. For change management or technology, when you're adopting a new technology, financial system, new CRM, new sales enablement like Salesforce,

[00:10:27] it's a lot of hassle to learn and you need to change your previous habits. Those are the biggest things. Learning and habit change. People are lazy. They don't want to learn things. They don't want to change their habits. And so those are the big challenges with change management and resistance. But with AI tools, there are three big blockers that I want to share about and get to the question that you asked. One of the biggest blockers is fear and anxiety. Again, not associated with the...

[00:10:54] Nobody's afraid of using a CRM versus using various Excel spreadsheets to look up client information. But they're afraid of training a tool that will replace them. They're afraid of losing their jobs. And so that definitely happens. And we'll talk about that. The second blocker is people who feel a threat to their identity.

[00:11:15] And so people don't feel threatened in their identity, professional identity, by a tool that looks up information, like I said, in a CRM versus a spreadsheet for client information. But they're afraid of a tool. They feel threatened to their identity by a tool that can write great outreach emails, that can write great marketing copy, that can create great images. It threatens who they are as a professional.

[00:11:40] And of course, for MSP providers, people are threatened by a tool that can resolve technical issues very effectively. And a tool that can essentially replace a lot of IT functionality. And a tool that can write code very effectively to the extent that development is needed. So people feel threatened to who they are as a professional. And so many of us get a sense of meaning from our work. I know I do. That it threatens their sense of meaning. So that's identity threat. That's the second group.

[00:12:09] And the third group is going to be people who do adopt AI. So the first two groups are really resistant to adopting AI. The third group is people who are going to adopt AI, but they are reluctant to talk about it because of a sense of shame and social stigma. So this is the shame group. Call them the reluctant adopters. And these are all described in psychographic profiles in chapter six of my book.

[00:12:32] And so these are the people who are going to not want to talk about their AI usage and who are not going to use their freed up time effectively. Therefore, they're not telling their supervisors about all the ways that they're using AI, not telling their colleagues because they can be shamed for using AI to write their emails, to do their code, to do their support for IT, for all of the stuff that they are using AI for. And so getting back to the question, so these are the reluctant adopters.

[00:13:01] So these are people like managers who are not using their time effectively. They're reluctant adopters who are using AI, not using their time effectively. That's freed up. Same thing for technical support, like you said, with MSPs who are not using their time effectively. So what can you do with the time that's actually freed up? I recommend having a time neutral pledge, meaning not firing anyone because you're using AI.

[00:13:30] Instead, focusing on growth. So the company can make a commitment to focus on growth instead. And we have recent research out from Stanford. It's so recent I didn't have an opportunity to cite in my book that finds that companies that are adopting AI effectively compared to other companies in the same industry. So same industry. Those companies are growing their headcount 6% faster.

[00:13:56] It's kind of a surprising number that companies are adopting AI, growing their headcount 6% faster, right? But they're growing their revenue 9% faster. They're using less people to grow quicker. But they are growing because they're seizing market share away from companies that are not using AI.

[00:14:15] And so what happens where jobs will be lost in the medium term, at least, short medium term, is companies that aren't adopting AI effectively because they're losing market share to companies that are adopting AI effectively. And believe me, this applies to MSPs as well as any other companies and to the clients of MSPs as well as any other companies. So to the extent that you are advising your clients on adopting AI, I think this is a number you should be aware of and you should be thinking about.

[00:14:43] And you can tell your clients that if you're not adopting AI effectively, you'll be part of the percent where the market share is seized from versus and you will not be growing. You'll be firing people and so on. And the companies that are growing are going to hire away your employees, also they won't need quite as many employees to grow. I follow. It makes sense. I love the metrics here. And in fact, before I get to a challenge question, I actually want to point out that I thought your sharpest metric was Sterling Parker's Ivanti, right?

[00:15:12] And he talked about deflection percentage married to CSAT, right? And the idea, so you're not trapping people in a bot. And I thought this applied in this space really well because for an MSP putting AI in front of end users on their service desk, I wanted to ask you, like, is that the whole scorecard? Is it deflection percentage married to CSAT? For them, yes, absolutely. It's a great scorecard.

[00:15:36] The more simple scorecards you have, the better it is because then you'll have a great headline number for the executives. So let me clarify. That is the fundamental scorecard that they have for the functionality. And of course, within it, there are a number of scorecards you have.

[00:15:54] But you want a clear, simple metrics that executives can appreciate, understand, that rank and file can appreciate and understand, that correlates and that results in good business outcomes, right? Like, that's what you want. So that's thinking about the ideal scorecard, like the fundamental scorecard. That's a great scorecard. We'll be right back after this message. This episode is supported by YouSecure.

[00:16:19] Across your client base, you have identities with exposed credentials, excessive access, and higher target value. The hard part is seeing where those conditions converge on the same identity. YouHealth from YouSecure connects these risk signals around each identity and surfaces high-risk combinations as toxic combinations.

[00:16:40] It shows where human risk is concentrated across your tenants, helps your team know what to fix first, and provides evidence of progress over time. YouSecure works with over 2,200 MSPs and 16,000 organizations worldwide. Know where human risk is concentrated. Know where human risk is concentrated, and what to fix first. YouSecure.io slash YouHealth.

[00:17:06] I want to ask you something about the book that I kind of had to read twice because I thought I'd gotten it wrong, and so I want to be really precise about this. You write that guardrails move AI alarmists into other profiles. And you also write that you don't advocate choice architecture for them, that their departure removes, and I think this is your words, a source of toxicity, and that the rest need to get on the bus or leave seats empty for more progressive passengers.

[00:17:37] Absolutely. I looked at that and I said, those are two different books. So what's the good instruction for a manager, say, on a Monday? How do we direct them? When you think about the AI alarmists, they are afraid for their jobs, and they are afraid for the organization of what will happen. So there are two things that you need to do to reach them effectively. Two things. And you need to get rid of those AI alarmists.

[00:18:02] And if these two things aren't reaching them, you need to get rid of any that are remaining, basically. But here are the two things that you can do to flip the script for very many AI alarmists. Again, this is a large group. So there was a recent Pew Research survey showing that 52% of respondents are more anxious and excited about AI usage. 8% are more excited than anxious, and the rest are equally excited and equally anxious. So you can imagine where the executives fall.

[00:18:31] They're obviously much more excited. And then you have a lot of people who are really anxious. So you have a large proportion of your employees or the employees of your clients are going to be in the anxious camp. How do you reach them? How do you reach these people who are fearful for their jobs? One is the job commitment that I said. The kind of the, we're not going to fire people because we are more effective and more efficient. Our focus is on growth. So that's a really important part of the communication.

[00:19:01] And that's something, again, leaders can do compared to how you're going to reach the other groups, the pragmatic resistors who feel the identity thread and the reluctant adopters who feel shame and social stigma. It's harder to reach them in some ways because with this group, you reach these, this group for leadership, communication and policy change. So you make that commitment, you make a focus on growth. So to have people believe you because, you know, at first, if they just communicate once, they won't believe you. They won't hear that. They need to really hear that. So that's one.

[00:19:31] Two, what you need to do is communicate to them that the company will definitely keep around people who are the most productive. People who are efficient, people who are effective, people who do more with less, basically. And the people who are the most productive are going to be the ones who are going to adopt AI, not the ones who will not adopt AI. They are the people who will adopt AI and who will build tools customized to their workflows.

[00:19:57] Again, this is one of the things that MSPs are not necessarily used to with AI tools. With previous tools, you don't have adaptation of tools to workflows. You have SOPs. So you have standard operating procedures for how you use these tools. Everyone writes these SOPs. They follow them. That's not how AI tools work. They don't have SOPs. They're incredibly flexible. Each individual employee needs to adapt it to their workflows.

[00:20:23] And so salespeople to salespeople and even different salespeople need to adapt it to their voice and their style. It's really different than how they do things. So you need to adapt it to your workflows. That makes it especially important for people to understand that they will be productive for the company by using AI tools effectively and building the prompts and agents that will enable them to be the most effective.

[00:20:50] So you need to essentially flip the script for these folks. So you need to flip the script for the AI alarmists for them to realize that it's not that the company will cut jobs. The company will grow. It will even hire more people. And you can cite the information. You can look it up, this information that I said. It's available in the research. So that's one. And second, flipping the script right now, they're thinking that, hey, I'm training my own replacement.

[00:21:16] But what you need them to think about is that I will be replaced if I don't train myself up on AI tools and if I don't use AI tools. And so with this messaging consistently applied and with actual policy of showing that you're doing what you're doing, you'll be able to reach many AI alarmists. And after consistent good faith efforts, if you can't reach them, you need to get rid of them because they will poison.

[00:21:42] They will be toxic for the company going forward and your AI adoption efforts. They will undermine them. One of the things that I see that you need to also do is make sure that these people aren't using AI maliciously. So there's a concept called malicious compliance that results in AI work slump, where people are just using an AI tool badly. Let's say they're just generating the initial thing that the AI tool says.

[00:22:10] They're not following the instructions of how to use prompts. They're not analyzing it with the right context. They're just like, okay, here, I drafted an email and I'm sending it to you. And it looks bad. And I'm drafting a PowerPoint, the report. It's on good quality. But they're like, well, I'm doing what you wanted me to do. I'm using an AI tool. And this is not just me saying it.

[00:22:30] This is something that research from Stanford and BetterUp published in Harvard Business Review found that workers report that 15% of the things that they receive from their colleagues are AI work slops, so poorly made AI output. And this poorly made AI output costs them each time they receive it. It takes them about two hours to correct it, to address it, correct it, all the things that need to be done.

[00:22:58] And so on the average, it costs companies $186 per employee per month to address this AI work slump. So either way, they're not using it or they're using it badly despite being instructed a number of times, good faith efforts for how to use AI well. So these are the people who still are in the AI alarmist, psychographic profile, stubbornly after a number of psychologically informed strategies. Yeah, you need to let them go.

[00:23:26] Now, you clearly understand the MSP model. So I've got two questions I want to ask about margin and understanding the model. The first one dives into, because three of your case studies in the book end with the firm handing the AI gain to the client as a discount. 8% off fees, 10% to 30% fewer billed hours. All three are framed as wins, and there really isn't any margin math associated with that. Now, the reason I was thinking about this is for an MSP delivering fixed fee managed services. Yes.

[00:23:55] Isn't the lesson that AI competes your own margin away? Well, so of course it does compete your own margin away, but you can deal with that through growth. So here is where you're seizing the space away from other MSPs that aren't adopting AI effectively. And that's the reality of the situation. So the example that you gave with the law firm that, yeah, you're going to be spending less time and you're going to be billing your clients less for your law firm work.

[00:24:25] You're going to be spending less time and you're going to be billing your clients less for your MSP work. But you're going to be making a higher profit. Just like all of the other companies, like I cited the companies, the research that you're growing your headcount by 6% more. You're growing your revenue by 9% more. Well, where's the difference going? It's going to the pocket of the shareholders, of the owners, whoever owns the company, right?

[00:24:50] So this is where in the period right now where smart MSPs are adopting AI and they're lowering their fees. And by doing so, they're growing their company. They're growing their company faster and using less resources than they would have. And they have a higher profit margin, even though they're lowering their fees. They're not lowering their fees to the same profit margin that they had before.

[00:25:15] Let's say if their current profit margin is 19%, they're lowering their fees, but their profit margin, even with the lowered fees, is going to be 22%. Great. Like this is where you want to go. Now, in the long run, you know, we'll have to see what happens. But in the short and medium term, there's a lot. This is a very addressable market space.

[00:25:39] And there's a lot of opportunities for you to seize market share from MSPs that aren't using AI effectively. Now, the other half of this is I want to reference through the roofing case, right? The roofing case in the book frames the ROI as $32,000 of your consulting time paired against $80,000 to $140,000 a year of software. That's the internal build beating vendor spend.

[00:26:06] Well, in this case, my listeners are the vendor, right? So where's the version where the outside IT provider creates the value instead of being disintermediated? So where they need to create the value, the MSP, is they can learn how to help companies adopt AI effectively. So if people are interested in continuing the conversation, reaching out and learning more, what's the best way to reach out and to get the book?

[00:26:37] Well, best way to reach out is on LinkedIn. So just connect with me. I'm very available there. But tell me that you heard me in the business of tech podcast, because otherwise, I will not accept LinkedIn requests at a random. I get way too much LinkedIn spam. If you want to get the book, just go to Amazon, Barnes & Noble, your local bookstore. It's published by Georgetown University Press. It's a peer-reviewed book. So it's pretty much available everywhere. And if you want to get a free sample of the book, check it out.

[00:27:05] Go to disasteravoidanceexperts.com forward slash AI book. So you'll get the introduction and a free chapter there. I really appreciate you joining me today for this lively conversation and really appreciate your answers. Thanks so much for your time. And I really appreciate you reading thoroughly for the book and citing all those case studies. It really speaks to your skill as a podcast host. You did a great job, Dave. Thank you. Oh, well, thank you for saying so. It's been a pleasure. The hardest part of running an MSP? Doing it alone.

[00:27:35] The Small Biz Thoughts community has been the room where independent operators compare notes for nearly 20 years. Real peers, real numbers, real answers from people running businesses just like yours. Pull up a chair at smallbizthoughts.org. Interested in advertising? Head to mspradio.com slash engage. The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech.

[00:28:06] Thanks for listening. I'll see you on the next episode. Produced by Picture This Video. Part of the MSP Radio Network. Thank you.