AI, RPA, and MSP Automation - ERP138

This file was generated by Descript Todd Kane: Most MSPs are still treating automation as a nice-to-have. Aaron Bailey has been treating it like a core operational strategy for over a decade. Aaron is the COO at Webit Services. Before that, he was COO at MSP Bots and an RPA wizard. He's seen automation from both the product side and the ops side, and that's a perspective most people don't always have in the MSP industry. get into all things AI, RPA, automation, and general operations. welcome to the Evolved Radio podcast Aarin: Thanks for having me, Todd Todd Kane: So you are, uh, very close to my heart as a, an operator, uh, obsessed with improvement and efficiency and, looking for all the things that you can squeeze out of the tools and the systems that you're building. so you wanna tell us a, a bit about kind of your experience and your journey on, on that moving, I guess, from sort of the… You were a tech, obviously, then went into the management and operations side, and then went really heavy into, uh, automation opportunities. How did that come about? Like, what was sort of the, the aha moment where you're like, "I think we could do this better"? Aarin: Yeah, I think it was right around COVID. We had just hit COVID, and I was working, uh, at a similar sized or about a $4 million MSP north of, uh, where I'm at now in Elgin. Um, and, and it was before all these tools had come out. Anything had… Nobody had even heard of, of RPA in the industry. And I was sitting there looking at a ton of PowerShell scripts, and I'm like, "How do I string these together?" And it became one kind of principle of, I feel like if I could do multiple things at once or in a string, then I could potentially repair a machine or execute more complex tasks than I'm currently getting from, like, my RMM. Um, and, and you get limited to, like, the RMM of just being a quick job or, or maybe it's just running the component and, and there goes the job, and it's done, and you get what you get at the end. So I got really fascinated about stringing some of these events together, and it really was just tinkering with the PowerShell scripts. And as I kinda got really more into it, it d- it kind of evolved into Python taking over more of the scripting components of it and then, and then a GUI. And so I was a few years in, and that's when I met Daniel, and I was, like, completely enamored with him. He just had all these great ideas, and I'm like, "Daniel, let me show you what I'm doing. Like, I, I might be crazy, but I feel like there's something to what you're saying." Um, and so I ran it through the whole thing, and he's like, "Wow, that's exactly, you know, that's exactly where my head's at too operationally." And he's like, "Let me show you this platform." And so I was like, "Oh, I'd love to be a part of that." And it kinda just took over from there and, and just, you know, bunch of super nerds just nerding out even more about automation. It just, uh, it became a problem where it's like we're carrying a stopwatch 'cause if you get Daniel and I in a room, we'll just talk forever and ever and ever, and then we look back, and we're like, "Okay, what did we get done?" So it's just kinda been this evolving fascination. Um, and then of course when ChatGPT came out and everything and it kinda changed the landscape, it really changed how much I could get done. And I started looking at it of just, okay, well, how can I augment myself to get more of what I wanna get done, done? Um, and then just one thing after another and, you know, four endless sleepless nights after another, and I'm realizing all the cool stuff I can do. And I've just burnt, you know, 84 hours into an AI project, and, uh, I'm explaining to my wife, "It's gonna be okay." Todd Kane: This feels Aarin: It's all in the name of work. Todd Kane: Yeah. Like, do you have any hobbies anymore? Yeah, yeah, lots of vibe coding, lots of, uh, figuring stuff Aarin: Yeah. Todd Kane: AI. Yeah. Aarin: Yeah Todd Kane: Well, so let's, uh, let's go deep and nerdy then, 'cause, like, a bunch of things that you Aarin: Sure Todd Kane: I wanna understand better. Um, so I, I guess from my perspective, like, uh, like I'll, I'll give the background on this. I was always very cautious and a bit skeptical about the way that vendors positioned automation and, you know, the original, uh, what they called AI, which was, like, barely machine learning. It was a bunch of strung together if-then statements. a-a-and I think my skepticism was somewhat justified. Uh, my tune has changed a lot, which we'll get to. But, uh, w-w-w- the RMM kind of relying on… Like, the original RMMs were scripts inside the RMM, and then what I, I kind of told people based on what I sort of understood was, like, "Don't rely on internal scripts. Don't build your scripts in your RMM. Just have your RMM call PowerShell scripts." I still sort of understand that that's probably the default way to go about things as, uh, maybe for general purpose stuff. Is that correct, or would you say… Like you said, you kind of drifted more into, into, like, Python for, for more programmatic controls of this beyond PowerShell. What was the difference between those? Would you say, like, typical automations from R- from an RMM are still probably better done in PowerShell, or is there, uh, would Python be the new hotness, and if w- if so, why? Aarin: They're sort of different languages. I think I've always still had PowerShell that runs because in some way. I mean, it depends on what I'm interacting with. If I'm interacting with like the Windows system or something like that, yeah, I'm, I'm definitely diving into my, my, my PowerShell because it's readily available on the system, don't have to do anything special. I think what ended up changing for us is what we wanted to get out of the system was a little bit more than what a lot of people can offer. You get SaaS products and a lot of them are great. You can do a bunch of great stuff with them, but there's always something more that we wanna do as engineers that just doesn't fit into that white glove. Um, and so that kind of evolved into-- It started off as actually a, uh, a bot, um, when I came to Webit. Uh, it started off as just a, a bot on the Azura framework. Um, and so it was, uh, setting all that stuff up and then hooking that into the automations with Roost and then breaking free of the RMM entirely. The only reason we use RMMs is, um, we'll write scripts for monitoring if we're trying to monitor like a service or, or something like that. Um, and then we'll use it to push our quick jobs coming off of like Roost or something like that. And so, um, most of our actual computation doesn't even happen in the RMM. Um, I go crazier. I have SQL servers. I've got Python servers. I've got just server upon server of just stuff. It's, it's kind of looking at it more like a beehive of taking individualized components and, and essentially what is that component really good at and snapping it into a larger system, um, that all works in conjunction together. And so I'll have a little bit in Roost. I've got a little bit in, in like our thread system that we currently run. We've got, uh, uh, some that's in a Python server just sitting in Azura. Um, but that's pretty much our primary instance of stuff running is that Python server which runs all our PowerShell scripts, uh, all our Python, everything comes from that server. But it, it's where it starts out is either on the RMM or within Roost, um, coming into that, and then it takes a direction based on whatever the code's telling it to do. Todd Kane: Okay, so it's more like extensibility than it was sort of like, like the actual call actions, right? Like PowerShell's Aarin: Correct Todd Kane: kind of controlling a Windows server or a Windows desktop, Aarin: Absolutely Todd Kane: more sort of global control and like functions that would call other stuff to do. I guess like, yeah, like you said, like it's almost in a way, like would you consider it sort of an alternate or a replacement of the larger functions of the RMM in a way? Aarin: No, I think they still have their own purpose. Um, I mean, you could probably dice and I'm sure there's crazier people out there than me that have probably tried, but I try to limit the amount of agents that, that are on my clients' computers in the first place. I mean, we've already got three or four different agents between systems that run on these things. So that's always been a big concern. Being able to leverage the RMM in that capacity doesn't really change what we're doing with it as far as the monitoring and everything else. It just, it's, it's kind of just, okay, I don't want that to be my entry point. I don't want my stuff to start in the RMM. It just needs to be a piece or a layer of this bigger system that we're building because inevitably, um, back to that beehive, as we start to snap stuff in and, and snap them together, um, we get better, we get faster, we get smarter with every piece that happens. And so it's just another layer, um, in that, in that onion that we're building Todd Kane: Yeah, 'cause this sort of goes to like, um, who is it told me this? I think it was Dave Sobel described RMM as a zombie product to me. I don't know if he got this from somebody else or if, uh, he coined that, but like I, I got immediately attached to that idea, and I still feel like it's somewhat true, but maybe I'm losing a bit of faith that like the, the, that it becoming a true zombie product is as near term as maybe I imagined it was. I think it's still got some legs, but like the longer term view of this I think becomes more like what you're describing of like global infrastructure, uh, and how these things are knitted together in some fashion rather than it being kind of built actually in the RMM. I think the RMM may be still as an attach point or maybe it migrates to like, uh, kind of Intune as backplane management, those sorts of things, right? But then, you know, um, you know, how are we managing SaaS platforms and, you know, third-party integrations, right? Um, uh, so, you know, it gets, it get- gets kind of complicated, but maybe as a quick aside, how do you feel about, uh, that, that description of RMM as a zombie product and like it will continue to live on, but like doesn't sort of carry the same gravity or importance in the MSP ecosystem in five years, say? Aarin: I mean, I don't, I don't… The world's getting really, really complicated. I mean, when I look at could I cut my RMM today and be just fine? Probably. I probably could. Would I have to do some extra work to get that to happen? Yes. Uh, but most of our clients have E3, so we can tap into the Microsoft system for patching. Monitoring, okay, yeah, that's a loss, but we also have Imibot, so that helps us with like RMM and, and, and patching and some other stuff as well. So we've got so many… I think we counted last time, we had something like 70 tools in our arsenal. Um, and I was just thinking, "My God, uh, we have a lot of stuff." Um, and it becomes why? And, and I, I agree. I don't see five years, maybe a little bit longer, um, but you just get one good idea that just shakes up everything and, and the cards start falling. I mean, we've already saw- seen threats come in. Or not really threats, I think fantastic products, but like Halo coming into the marketplace and really just taking on, um, that Autotask Connectwise relationship of them being the dominant. Now here they come, and you have NinjaOne making a play for it with their PSA and their RMM, and, and they're trying to pair them together of the best functionality that could possibly happen. And that's true when you're in their platform, absolutely. Like I'm not gonna be able to replicate the same experience in my Autotask platform as I would if I started to say, "Hey, I'm gonna throw in, um, you know, Automate or whatever, uh, as my RMM platform." It'd be a completely different experience. So I, I don't know that it's really the RMM is going away. It's reshaping into a different experience, um, into what it's being used for and how it's being used. Uh, the, the one thing that's really doesn't seem to be changing at that, that layer is the PSA as far as being a ticketing tool, but I think that's natural. I mean, what are you gonna do with a ticketing tool? Inject some AI into a 50-year-old framework? Todd Kane: Well, e- even like Thread, you know, like I, I love this, the work that they're doing over there and, and they have this, this call to action of death to the ticket. But, you know, realistically, there still needs to be some type of database of record of work activity and action. So is it a PSA? Like, where does the ticket live? Like, we still k- uh, we need the plumbing still, right? Aarin: You need the plumbing. If there's not a ticket, it never happened. Todd Kane: Yeah, that's Aarin: Even if it did happen and you don't know it, Todd Kane: Yeah Aarin: prove it. It's like if a tree falls in the wood and nobody's around to hear it, it doesn't make a sound Todd Kane: Yeah. So this also lends to, you know, delicate infrastructure. Like, as you kind of like gobble these things together and like pack modules onto modules and, and build upon sort of this, this, uh, this towering code structure, um, how do you think about sort of the risk of that being, you know, unstable or a lot of, uh, support effort or, uh, or it just being quite complicated and troubleshooting things that go sideways for some reason? If you add a module, then all of a sudden three other things break, right? Like how, uh, how do you think about scaling the complexity of the things that you're building internally? Aarin: I think that's part of what I've tried to do is keep everything into those separate modules and then, you know, we use a lot of RESTful API and other stuff to, to… That way if something does break, it's not really affecting the whole platform. We see it just in that individualized module. Um, but one of the things that I've, I've really tried to really focus the whole team on preparing for is, like, if you're not ready to start learning Python, at least in a, in a readability, you don't need to go sit there and code out the next app. But you need to be able to read Python. You need to understand what RPA and automation is, and be able to have a conversation in, in, in what's going on in reality. Um, it, it helps as we start to go through this to have more people involved in that sort of, that sort of mentality of, okay, five, six years ago, everybody was jumping on the security train and everybody needed to be a security trained professional. Well, now it's like, okay, well, why wouldn't we do the same thing with automation? Um, and it makes more sense because, or it makes just as much sense as it did with security because the world ends up shifting. If you're not preparing your staff for that, then in five years I won't even be in business. So it's kinda just adopting and taking, like, I understand that AI is gonna happen. I understand that the world is gonna change. How can we better prepare ourselves, um, to be ready for that? And it's, it's kinda brings me back to that Ray Dalio quote of just, "You're going down a river and there's always objects in the river. Um, you can't remove the objects. All you can do is, is learn to go around the objects." And that's just kind of how the, uh, the AI and the MSP's gone Todd Kane: Or flow like l- flow like water, Bruce Lee, right? Aarin: Yes. Todd Kane: Uh, okay, so I'll, I'll maybe circle back to that 'cause that's another thing I wanted to, Aarin: Yeah Todd Kane: on here. Um, but the other, the other aspect that I find really interesting, um, I, I know a lot of the stuff that you're building, you build internally rather than using something off the shelf. Now, part of this is because you can, I suppose, and you have a bit more control over it, But, like, uh, what's, what's sort of your calculus on I could buy an off-the-shelf project, uh, product, but, you know, for example, one of the things I know you did was, uh, built like your own AI for triage. Like, that's generally beyond the capability of a lot of people. Maybe less so now with some of the, like the AI, uh, assisted coding tools. But, um, uh, that, that's, uh, that's something you chose to build internally rather than using a, like a, a software package that you could buy that do, does the same thing. So what… Like, what's your, what's your thought process or analysis of this feature exists in some of the platforms I could potentially purchase, but I could also just go build this myself because I know I can? Aarin: I think it comes down to, uh, looking at it from building at scale. Uh, you know, we've, we've done a considerable amount of just growth as, as far as a company, and, and we're about twice the size, but we've only added two bodies, um, across the whole stack. Um, everything has gone up and, and, and we… As we start having to deal with that, it becomes what delivers the most ROI coming back off of that. And that's an important factor because I think early on as I was doing automation, uh, I think I mentioned, I would go spend dozens of hours just trying to play around with it. And, and it really became important that if I was gonna sit down and do an automation, it was gonna have some sort of ROI or some impact on the company. It wasn't just for the sake of automating. And so as we start to build this stuff up, we start looking at what is the cost to do it versus the benefit that we're actually getting off of this stuff. How complex is this gonna be? Um, and then the second half of that is looking at does it fit what we wanna do? Because we're not building like every other MSP. I mean, I've come across a lot, as I'm sure you've come across a lot of MSPs, and n-not one single one have I ever looked at and said, "Wow, you operate just like I do." In fact, it's almost always the exact opposite. And so, um, and even when I was talking to Daniel yesterday, uh, we were talking about scale, and he's like, "Well, yeah, you go build it for the one company and that's great, it works perfect." And then the second, okay. And then the third, and then it's like, oh crap, we've gotta go figure out we missed this, this, this, this, this. Um, and so that kind of became just this whole, I don't wanna have to deal with other problems of other things that are going on. I don't wanna have to deal with, uh, um, being put into a box where I'm sure it might work for this group of MSPs, um, but we operate entirely different. You know, our entire triage and dispatch is automated, uh, with ninety-eight percent accuracy, and we haven't touched it in, in about a year now. Um, the triage we did move, uh, naturally over to Thread because, um, token costs and which became a completely different factor in all of this, of just how much token costs are coming in this. And it became cheaper to run this off of Thread at the time than it did to actually run this off of a large language model, where inherently it's actually cheaper, it's cheap enough to keep our dispatching on that system using the AI as kind of like a, um, this ticket got missed in the process, so it, it runs through a fail-safe check to say, "Did everything get done on the ticket?" before I actually punch it through and send it out to somebody. Which, and, and funny enough, that uses no AI. It's, it's actually all a roost workflow and a stored SQL, uh, process off an AI as our tenant. So, um, it's really is a true hodgepodge, but I haven't really gotten to like, this is what is the best thing for ROI or… I think the best thing that I've found is have a plan, know what you're gonna build, and know that it's gonna return some sort of ROI on it. Um, and that's kind of just been my guiding map Todd Kane: Yeah, so that's really fascinating, right? Like, um, I, I, I quote this all the time of, um, uh, especially being a person that consults with of different MSPs and gets a bit of a peek behind the door for a lot of them, and I always joke that everyone has the same predominant business model for the most part. Like, maybe you're a bit more of a VAR, maybe you're a bit more of an MSP, but generally, like, we're all trying to achieve the same thing with the same, same general business model. of them operate the same way, right? Like, which is Aarin: Yeah. Todd Kane: right? Like, the number of ways to actually do this and the differences that always sort of stem from sort of the, the, the, the, the lessons or the experiences of the owner, the psychology of the owner, the makeup of the team, the experiences of the team, all of those things influence how the operation actually sort of spins out and what it looks like. So there's, there's definite truth to that. But also your, your point about, um We build it for ourselves because we want it the way that we want it, not the way that the vendor chose to build it based on inputs from all of these other companies that don't operate the same way that we do. I think that's really, really interesting, quite insightful also that, uh, um, you know, you, you-- again, because you have the capability to take that on yourself, I think is a bit of a unique advantage as well. the token cost, I, I think this is a really interesting aspect, right? Like we're seeing this all over the place in the enterprise, right? So I'm glad you brought this up because I'm curious how this will eventually start to, uh, impact more of the MSP and the SMB space, we're seeing at the enterprise level, there, you know, I'm sure you've heard the term token maxing, right? There's a few, a few companies. I know a guy who works at a very, very recognizable, uh, uh, uh, enterprise SA- uh, SaaS company, they have a token leaderboard. And the CFO was in a meeting with the leadership team and suggested like, "Hey, do you think we're creating, uh, an incentive that doesn't align with what we're trying to achieve here?" And everyone kind of understood what they were, they were suggesting, but they got completely shut down and the, the, the ethos from the leadership team was essentially let them cook, right? But at the same time, we're also seeing a lot of enterprise companies that are now drastically slashing the AI coding budgets because they have, they have dwarfed the actual, uh, human labor costs that they were supposed to replace, Aarin: Yeah Todd Kane: So there's some interesting efficiency economics that are not necessarily working out the way that people anticipated with utilizing AI coding, and I'm curious sort of how this will eventually impact some of the SMBs that are trying to, uh, you know, do some of the stuff on your own. Um, I think like you have a more traditional coding background and can do some of this. You're, uh, kind of by yourself and string these things together. Uh, how much has that shifted in using like, uh, um, you know, vibe coding or AI pair coding? How much of that is your, is your workflow now versus before? Aarin: Well, I think it's, I think the industry has got it wrong. AI was never, in my mind, never meant to replace humans. And the more that we try to go force that down humanity's throat, the more lost the companies are gonna be. I mean, Salesforce went and laid off, what, 4,000 jobs, and then next thing you know they're hiring them all back. Um, AI's always been an augmentation. It's, it's, it's like taking something that can only handle… It's, it's the world of data. Our data in the world is so complex. There's so much data. We can't physically, as humans, handle that. So we need AI to be able to keep up with the amount of complexity that life is, is, like, produced. But it's not like, oh, let's go replace humans and, and we're just gonna have robots walking around. Like, that's, it's not gonna happen. Um, but companies think that that is a good way to get ROI. Oh, let me just go replace it with this magical robot in the sky that's gonna make all my problems go away. They end up making them worse because, yes, the AI will kiss your butt, it'll tell you you're right, it'll tell you you have a great idea, even if you're like, "Hey, I think I'm gonna go take over the world with, uh, a slingshot." "Yeah, you should do that. That's a great idea. I think you should do that." I'm like… So it's, it's fascinating to see all these companies dive so heavy into this, thinking that it was gonna fix all the problems, but it was not ever meant to do that. It was meant to augment. And so I augment my workflow, I would say about 40% of the time, because I can do more than I could without it. And when we're talking like, oh, let me go through and structure out everything and all that, like, I'll, I'll just make the AI do that. But I still have to go back and check it because what library did it use? What did it do here? What did it do there? Um, which is always the biggest thing working at MSP Bots. It was one of the biggest things that everybody looks at of, like, what third-party libraries are you using when you're coding in your system? Because if one of those gets attacked, it's, it's just an entryway into your app. And so data security became just so prevalent. And so I get really paranoid in some of the vibe coding in just using any of the AI stuff for specific functions. Like, I won't use it for anything authe- authentication based. I won't use it for anything, like, security based and that nature. But I will use it to check the security, which is, is, it sounds weird, but it's like checks and balances of yourself. Todd Kane: Yep. Aarin: Um- Todd Kane: to that point, like Mythos was like that… There was a lot of… I, I was talking about how like, like Mythos is legit and like maybe it's AGI and it, uh, its impact on security w- it was pretty dramatic. And I initially got a bunch of pushback from people saying, "Oh, you know, like, GPT 3.5, they said they couldn't release it because it was too dangerous. Like, this is a marketing tactic." I'm like, "No." Like I, I agree with what you're saying, but if you look at the data, like it, it was able to find thousands and Aarin: Yeah Todd Kane: of vulnerabilities in, in open source software. Like it was legit, right? So I, I, I Aarin: It's scary. Todd Kane: It's really scary. It's a totally Aarin: It's scary. And I read today that the, um… Yeah, that's the other thing about, like, the AI is just, like, how do we view the AI in the US is different from how they would view it in China or different from how they would view it in Russia from, you know, just business use cases, even down to environmentals. Like, we're more like, "Oh, let's not go kill a bunch of water and power plants and build all this stuff." Like, we're protesting out there. They'll just keep building and building and building, and some of that has been seen in the, in the manifest of the number of attacks that have come across recently. And we just see more and more of those AI attacks coming across and across and across, and they're just, they're just Rolodexing, looking for those vulnerabilities, anyone they can find, and then right in, and then somebody's gonna have a real bad day. Todd Kane: Yep. I mean, we won't, we won't sidetrack. This is something I, I probably wanna have someone on to talk about, uh, to go a little deep on this. But like the, the, the whole idea that like, uh, a data centers are gonna steal all the water and, and things like that. Like, we had this conversation on a, on a group chat, um, that like, has anyone attacked the almond industry, right? Like see how much water they take, right? Like a lot of these fears I think are misplaced in application. Like I get why people are fearful of it, but I, and you, you see all these layoffs and stuff like that, and Aarin: Yeah Todd Kane: people that are coming out of college are somewhat Right. in kind of having some hate on for AI. But I think sort of the, the why is a maybe mi- misapplied and the actual impact that it will carry, I think is misapplied as well. Aarin: I think the impact is huge. I was, I was talking to my wife's cousins and, and they're 20 years old respectively in, in variances, and their biggest concern is, "I don't feel like I need to save. I don't need to do anything because AI is just gonna come get my job." And I'm like, "That's not how you should be feeling at 20 years old." Todd Kane: Yeah. Yeah Aarin: And first off, it's not coming to get your job. I mean, I think that's just another far-fetch in the industry of just it's gonna replace all humans. Okay, well, what would humans do then? Like, that sounds like a very dark dystopian future that I don't wanna be a part of. Can we get to Mars already? Todd Kane: Yep. Yeah, universal, uh, universal, what is it? Not universal bacon, um, uh, universal base income, universal, Aarin: Basic, yeah, universal basic income Todd Kane: Yeah, but there was a second one that, uh, I'm blanking on the term, um, where it's, it's like a full version. Like, it's not like Aarin: Okay. Todd Kane: like 60K a year, Aarin: You just get a full paycheck Todd Kane: everyone gets like 160K a year. Uh, yeah, anyway, whatever that term is. All right, so, uh, yeah, we won't solve the, the AI debate necessarily right now, Aarin: Not today. Speaker: Tired of fighting the MSP fires alone? The Opsleader Pro group connects service delivery professionals who understand your daily challenges. From KPIs and workflows to career planning and team management, Opsleader Pro has systems for you to use. Join operations leaders from successful MSPs who are sharing real solutions for managing client expectations, optimizing service delivery, and making your service delivery team as effective as possible. Opsleader Pro, 'cause your service desk deserves more than just survival mode. Visit opsleader.co. That's O-P-S leader dot C-O to apply to join the public community. Todd Kane: so scaling, I, I think it's interesting, like you noted, like you guys essentially doubled in volume with only, uh, adding two additional headcounts, so like a, like a pretty, uh, asymmetric, r- ratio for labor to, to work. Um, I mean, you would, y- I think you would credit that to a lot of the work and the, the automation work that you guys have done. where would you say is sort of some of the highest impact? And, and maybe this is sort of a twofold question of what did you find, and therefore what would you reme- recommend people investigating within their business saying like, "Well, shit, this sounds great. Aaron, you know, uh, was able to have this huge impact on, on the business and how they manage the, the workflow." What are the areas that you found the most value in? Um, you know, triage probably, um, being a, a big component of that. What are the other areas that, that were a really good, uh, source for you guys to automate work? Aarin: Yeah. The biggest thing came down to having a really detailed, um System across the board. I mean, between, you know, the vCIOs, the projects team, we have a field team, we have a help desk, we have a central services. Not every single one of those teams even existed when I started a few years ago. Um, when I started, there really wasn't much of a coherent system that was kind of there. Like, they had systems in place, but there wasn't anything that really tied stuff together. And, and so I looked at it and saying, "How do we tie all these things together? Let's make it simple for the, simple for the employee, uh, easy for me to, to kind of have visibility into, and that it turns over really good results for the client." And so as we went through our, our entire system, we started really just gutting everything and rebuilding it because I'm really kind of OCD about that. Like, everything needs to function in a specific way. And that's not uncommon, but I think the layer that I took it to, the granularity that I stepped into is a lot different than what most people do. They kind of scratch that surface level and then they're like, "Okay, that's good enough. Let's move on." And it's like, okay, well, that may be good enough for one to three million dollars, but that's not gonna carry you past three million dollars. Once you get past that stage, you start talking about a different maturity model for your businesses that is needed. Um, and therefore, your policies have to evolve in some of that. And so having some of those granular policies and playbooks that say, "This is what happens, when it happens, and how it happens," has been the biggest lifesaver of anything. Because at the end of the day, we're still a people business. We still have people to support. I still have people that call up and say, "Hey, I wanna talk to a tier two right now." No automation in the world is gonna fix or solve that. They want a human. By having really good finite documentation and playbooks, it, it really helps figure out, okay, where is everything running? Good. Everything runs great there. Now let's take it in reverse and look at what doesn't run great. Let's look for the anomalies. What sucks? What's bad? What's terrible? Let's focus on that. And so we shifted focus in just saying, "Good is good. We know good is good. Let's look at bad." And so our whole life for the last year has just been living in what is the bad? Where is the bad? Where's the problem? Where's the problem? And it may not be something huge like, oh, the house is on fire. Sometimes it's like something stupid like, oh yeah, our dispatching automation wasn't kicking off, or something happened like it wasn't tagging a ticket correctly. Okay, let's go do this and this and this. So as we start to build more stuff on top of that, it kind of grew. But without that base foundation, we never would have gotten off the ground. And that's pretty much the biggest key component to everything. It's even… You can't dispatch anything without a roadmap. You gotta have a blueprint of some-- or sorry, not dispatch. Automate without some sort of blueprint, some sort of document that says, "This is what I'm building and this is how I'm building it." And if that doesn't exist in your operations today, putting that automation in and saying, "This is how it's gonna work," isn't gonna fix that problem. You're still gonna-- You're probably gonna have more problems after you get done with that that you don't know how to fix because you implemented something without knowing what the outcome is gonna be Which is a huge part of anything, of just what is the outcome of what we're doing, what we're trying to accomplish. And so we've kinda tried to keep that mentality as we go through and build all this stuff, but that foundation has been always near and dear to my heart. Uh, which is why, you know, when we were talking a few years ago, I sent you so many documents, and I'm like, "Here, here, here's 1,000 documents." I'm And you're probably like, "Whoa, this guy's nuts." Todd Kane: friends, Aaron, because the I, I think this, this is an unexpected answer, I would think. Like, uh, like if anyone would have guessed, like, what you would say if I asked you that, it wasn't good documentation and procedural understanding, right? Like, that's not the automatic answer when people are like, "How do you automate something?" It's like, "Good documentation to understand the process." Like, wait, wait, where's the cool techy stuff? Also, I, I think that's an incredibly important base layer, is what Aarin: Yeah Todd Kane: actually trying to achieve? What's the end goal, and how do we actually achieve that and then iterate the process internally? Once you have a good process, then you can automate it, and I think that's where people kind of go sideways on this, is they're trying to automate a procedure that doesn't really exist in the first place, Aarin: Absolutely. Todd Kane: it's good and it produces a good result? Like, uh, let's just keep stabbing at it, basically, right? Aarin: Yeah. Yeah. And I've been fortunate enough to where because we have gotten… I mean, when I came in, we were taking in, I don't know, somewhere around 100, 120 tickets a day. Um, we've since now, we take in maybe 20 a day. Um, so the, the volume of everything has gone down. Some of it's also, like, just fixing the business and getting everything into where we want it to go and how we want it to go, lining that up with what is our vision and our EOS say that we wanna do, where do we wanna go with all this stuff? And then taking, you know, what Eric is, is, you know, him being the visionary of saying, "Hey, here's where I wanna go." And then it's my job to kinda say, "Okay, you know, I like… You know, you gave me 1,000 ideas. 900 of them are great. I'm gonna ignore this other 100 and hope you don't kill me for them. But let's go focus on these 900." Um, and so you take all this stuff and put it together. Todd Kane: so yeah. Aarin: It is. It is. Uh, it's also learning when to say no. Todd Kane: Yes, yeah. " Aarin: No, we can't do that today." Todd Kane: but that, that's a really important aspect as well. Like, uh, the vision- the visionary and integrator relationship is, I think, too often integrators are afraid to say no, right? Uh, because they're like, Aarin: Yeah Todd Kane: you know, that's what the owner told me. Like, I, I, I don't, I don't necessarily own this place," right? A lot of, uh, integrators and operators are not necessarily owners. They can be, but in a lar- large majority, I don't think they are, and that, that's a difficult dynamic of being able to sort of politely and professionally put back and push back and say, "I hear what you're saying. I, like, are you okay if I pick these three and I'm gonna drop the rest?" Right? Aarin: All Todd Kane: that feels like a dangerous conversation when it's entirely practical, right Aarin: Well, it should be. I think that's majority of the problem across, across… You know, if we look at why can't an MSP grow, it's, it's likely something to do with something, somebody at the top has hit a lid some way, shape, or form. Like, they can't get out of their own way to allow the company to grow. That's why they say getting off of owner-led sales is the hardest thing to do. It's, it's, it's a hard thing to do, and I don't blame them. I mean, from the visionary end, it's like, this is their baby. They built this bad boy. This is everything to them. Companies have to grow. Companies have to be healthy, and we always have to consider what that, uh, that outcome is gonna be, whether it's good or it's bad, no matter what layer it comes from. And, and I've been really lucky that Eric's been a great, great mentor, and he's been a great friend. And, and it's kind of just we hear each other when we're talking, and we're going through all these processes, and it's like, "Hey, what do we actually wanna get through? Like, get out of this." And, and when we sit down and line it out, we're pretty much identical. And it got really scary because we used to think so much alike I wouldn't have to talk to him. Um, and I feel like that's a really great VI dynamic to be able to have that sort of sync, but it doesn't mean that we're always in sync. There were definitely periods where it was like, "Okay, hold on. This or this or this." Um, but as you get through those, some of those, those things and start to work through it, you get stronger as a, as a dynamic, and you build something that's really impenetrable when you look at it from can something bring this down? No, not if our dynamic is actually working. Because if it's working up top, it'll work down below Todd Kane: Yeah I think this is a, an important aspect, I agree, of like the professional dynamic of being able to disagree professionally and still align on we're trying to achieve the same, the same thing. We disagree on approach. Let's kinda talk about that, and at some point I may go, "You know what? I don't necessarily disagree, but I understand sort of why we're making this decision, and we're gonna run forward with this." And I'm not gonna say, "Hey, I told you so. I told you this wasn't gonna work." It's like, "Ugh, okay, we tried this. Didn't work. Like, let's go in a different direction. That cool? Cool." Right? Like, like I think people make massive mountains out of molehills in all of these subtle sort of disagreement dynamics between, uh, a visionary and an inter-integrator, when as far as I'm concerned, like that's necessary, right? Like I often joke, you've heard me say this in meetings, is a meeting without a little drama is not a meeting worth having, right? Like there has to be enough disagreement for you to align on and push back and have sort of the, the, uh, that subtle disagreement, uh, on approach that there's other people are heard, and there's other dynamics and other voices in the room. That's the point of, of those discussions and having that Aarin: Yeah Todd Kane: different perspective. Because to your point, like if we align on everything, we're, like there's no point in having a meeting Aarin: Nothing's gonna get done. Todd Kane: yeah, like we all agree. It's like, like, okay, we're gonna talk about this thing. Yeah, yeah, sounds good. Okay, well, why are we here, right? Like this absolutely could've been an email, right? Aarin: Yep. Todd Kane: Yeah. All right. So Aarin: Yeah Todd Kane: what, another piece I wanted to, to hit on here is, uh, sort of user-facing versus, um, versus backend automations, right? So, um, I'm curious to you, uh, how you think about what the, the, the backend automation, how, like how much of what you do is just purely underground plumbing mechanics, uh, is, is invisible to everyone. It does some fancy stuff and produces something eventually for either an end user or an internal, um, um, uh, sort of member of the company. And what I'm thinking about here is like how does AI and automation continue to show up for both the end users as well as the internal users within the company, right? Like you mentioned the VCIOs, and we're starting to see a lot more, uh, automation, data collection, and AI enablement for, for the VCIO group. um, I'm struggling to wrap this into a consolidated question, but I think you know where I'm going here. Aarin: Yeah, no, absolutely. Todd Kane: th- the, yeah, the, uh, the underground backend mechanics Aarin: Yeah Todd Kane: user-facing interactions with the automations and the AI. What, what's, what's sort of your take on that dynamic? Aarin: No, this has been actually pretty funny. Funnily enough, been pretty, uh, hot recently on the topic as we kinda go through stuff of just client-facing versus the plumbing, and I think it, it kinda sparked off of, um, funny enough, watching the Super Bowl and, and seeing that almost every single ad was AI, and by the end of it I just wanted to throw up 'cause I'm just like, "I'm done. I'm done with this. I don't wanna… I-- Where's the funny Budweiser commercials?" Like, what happened to us? Todd Kane: ads were pretty funny though, like poking Aarin: The, Todd Kane: GPD Aarin: it's… That's about it, though. Like, it's, it's everything else was so fake. And so as we started to go through, I started to notice that, you know, what is people's reception to AI gonna be? Because, you know, my dad is, is very, we'll say, uh, stuck in 1950, we'll call it. So he just got his first iPhone last year. He's had a flip phone up until then. Um, I mean, it's, it's crazy how much he lives… Yeah, never- doesn't email, doesn't computer, doesn't believe any of that stuff. I, I'm surprised he has a debit card instead of just a checkbook. So Todd Kane: up to be such a nerd coming from that lineage then? Aarin: It's, uh, it's, it's, um… I was a wild child and, um, yeah, I don't know. I got really into it Well, I started off in film, surprisingly enough, and then when I got bored of that, I was like, you know, I'm… My grandfather was really into computers, so I think it kinda just was, Todd Kane: Same Aarin: just took over and just ended up being one of those things to where I realized I'm really good at it. So, like, what am I doing? Like, am I really gonna go make a blockbuster hit in the middle of Oklahoma? Probably not. Um, so that just kinda died on the vine there. But, um, Todd Kane: Or I derailed Aarin: no. Yeah, you derailed me. Now I forgot where we were going with all this. Todd Kane: internal versus us- Aarin: Yes. Todd Kane: backend versus, Aarin: So Todd Kane: yeah Aarin: yeah, it became like this, this, this kind of… When we were talking with our clients, they started to get frustrated, and they're like, "I don't like this AI." And I'm like, "Well, what about it don't you like?" They're like, "I don't like that it's talking to me." And I'm like, "I… Wait." They're like, "Am I… Is it talking to you, or is it saying something wrong? Like, explain this a little bit more to me about what you're saying." And they're like, "Well, it just says some weird stuff." And as we started to go through some of the systems, we realized the AI had been hallucinating and sending out random public-facing notes just at random to just random people about random things. Uh, it didn't even have to do with the ticket. And I'm just like, "Okay, that's concerning." Um, so we kinda reshaped our strategy and, and really the only time that AI hits a client-facing ticket is that initial triage bot and then a reminder bot. All the other AI is, is completely on the back end. It faces either the techs or it just functions on the back end autonomously. We've tried to break away from just shoving so much down our clients' throats because we've learned from past events like security. And having gone through security becoming a thing and a big push in the industry, the more we shoved it down people's throats, the more they got frustrated, hated us, and left. Um, I've got something like a 95% client retention rate over the last two years. Like, we've lost two clients over the last two years. Um, and those were clients that had already left and came back and left again, but it was all pricing. And so it's, it's people want to be treated fairly. They don't wanna have a bunch of stuff shoved down their throat. They don't wanna be told what to do all the time. Like, it becomes a partnership. And so I look at it in that regard of just I don't wanna put any more AI down their throats than they're able to handle. Otherwise, they're just gonna go somewhere else. And the downfall of that is not everybody in the MSP can or is implementing AI in any capacity, and it's just as easy, if not easier, to find a, a, an MSP that doesn't do any of that stuff than it is to find one that actually is fully all into that at this point. We haven't fully made that swing or that adoption. So it does play a huge factor of just how much do I actually do and how far do I actually take this? But for my guys, I mean, we're all nerds. We all joke about it. Um, I've turned the AI off for them when they've complained. Um, and they're like, "Whoa, bro, this is too much." And then I get it. Sometimes when it's like every 10 minutes and it's reminding you of something or it's doing this, like it could be a little distractive. So we've learned to tune it on the back end even as well. But it's just kinda case by case Todd Kane: Yeah it can be quite overwhelming, right? Like even when Aarin: Yeah Todd Kane: I, I'm working on a project with, with AI and like I'm asking it for something and then it gives me like just this massive mountain of work to do and I'm like, know what? Maybe this isn't that important to me," Aarin: That's why I asked you to do this, AI. Todd Kane: Yeah, exactly. Um, yeah. Okay, um, other piece I wanted to ask about this? Um, uh, now I'm blanking out. Oh, yes. The, um, so the other aspect I, I find fascinating and I'd love your input on this, is, is, uh, voice AI. and like I've seen a, a number of companies experimenting with this where like call answer is actually done by AI and this is incredibly polarizing for people, uh, for the reason that you're noting. Like some people are Aarin: Yeah Todd Kane: "I, like, like screw off with this AI thing. I don't wanna talk to this. Like give me a human." but I can't figure out exactly what it Is Originally I assumed it was gonna be generational where like, you know, sort of 50 plus type thing like they don't wanna talk to an AI but like the 20s like they don't e- they don't wanna talk to a person, right? So like maybe it was generational. That is actually not the split that I'm seeing. I'm seeing it more on just sort of a client base by client base difference where I know this one organization they went to completely no human interaction with, you know, ticket ingestion, right? Like tickets are submitted entirely digitally, uh, either by the phone or by, by email or, or chat and no humans accept incoming tickets and I was like, Aarin: Is this Xfinity? Todd Kane: with this." Aarin: Are we talking about Xfinity? Todd Kane: No, no, no. Well, yeah, Aarin: not the internet provider. It's definitely not Xfinity Todd Kane: are, they are an, they are an MSP, and they, they've had zero pushback. Like they've had a few people that are like, "Hey, what is this?" Uh, like, "Is this legit?" And like then they're like, "Okay, cool, whatever," and they, it's gone really well for them. And I Aarin: That's good Todd Kane: nec-necessarily the case for everybody so, uh, you know, I'm, I'm, I'm sort of curious how this plays out to your point of like we can't push this on people that are not necessarily ready for it and like we're barely getting past the point of convincing some owners that like live answer isn't as important as they think it was or think it is. You know, as, uh, like 10 years ago you wanted live answer and they can't get past the fact that like, no if you call someone back within 10 minutes just as good but that's really difficu-difficult to convince people. So you know as AI starts to show up in a lot of these other places in more of a user-facing capacity I think it's gonna be incredibly variable on how successful It's gonna be for no parameters that actually make a lot of sense. What are your thoughts? Aarin: It's easy to reduce labor in that avenue. I think you can make some really good justifications, like you said, client by client. Um, you know, it's, it's back to my Xfinity joke. I mean, uh, it is something that, you know, for myself personally, when I call Comcast, which is a, a major… One of the main internet providers across the United States, there's like three of them or something like that, and then a bunch of other smaller ones. But it's AI. I, I get a bot every single time. I have to jump through hoops to try to get support, and I'm like, "I just wanna talk about…" You know, maybe it's like my speed is slow and I wanna upgrade. Like, I wanna give you more money and you're making me jump through all these hoops. So a- I think majority of America, while they love the fact of AI, it scares the hell out of them, and they're not ready to just be a part of that m- that, that wave, and we have to be respectful of that and just, it's a really cool nerdy tool, and all the nerds are really geeking out about this. But, you know, between the energy that it, that it takes and, and the, the cost, I mean, I was reading that it took something like $192 million for them to, to, to go from, um, what was it, 5.0 to 5.2 with ChatGPT. So this is two hun- $200 million just to make a small minor version update to the product that didn't really… I mean, when 5.2 came out, I think, I think my nine-year-old is smarter than that thing at that point. Uh, and then so then they kind of freaked out and released 5.5 and, and now you have what you have. But it is getting extremely expensive and everything else. How much smarter do we think we can get? And that's the looking forward of just saying if people don't tolerate it now, and it's gonna take 20 more years to get to where the, where we think it should be today, and you're gonna drag people th- they're gonna get numb to that. They're gonna hate it even more. And so it's kind of just been like a very slow just integration in that regard of just back to how much do we let it talk to clients. Very minimal because we don't wanna burn them out. We don't wanna scare them off. We want them to be comfortable with it, and we want to also as their partner walk through this A- AI journey as they have it. And, and they're gonna have questions, and we wanna, you know, be able to answer them and help them, um, kind of get their RPA in for their business or whatever it may be that they wanna get done. Uh, but it's also knowing the fad of just, hey, we don't just go all in in AI just because somebody said, "Hey, AI is the future." Um, it's, that's, that's craziest. Todd Kane: Especially because like I'm really curious how this shakes out, especially over the next sort of two to four years because, uh, I think we're starting to see the recognition that we're in the subsidized era of AI, right? Like originally Uber, when it came out, like you were-- you-- half of your Uber ride was pay for- paid for by VC dollars. we're sort of in the same era of AI where like, people were panicking about like the restriction changes for, for like Claude all of a sudden like felt more expensive. They got less token usage. And part of that was just a simple physical limitation of compute, right? Like they had literally run out of compute and had to s- find some way to save it. And then, you know, uh, xAI to the rescue. Since they have no clients, they were able to fork off ev- all of their, their Aarin: Imagine that. Todd Kane: to, to Claude. Thank you. Yeah. Uh, but, uh, I, I think, I think it's dangerous to rely on this stuff too much with the expectation that. it might get a lot more expensive in the future. And then, you know, Aarin: Yeah Todd Kane: point of this is not cheaper than humans, right? Uh, but you know, it might become a lot more expensive in The future as the VC dollars start to sort of retract from, from helping to subsidize this. So, you know, um, you know, buyer beware Aarin: thirds of the world of comedy that's tied up in AI right now , Todd Kane: Yeah, Aarin: which is a dangerous game Todd Kane: yeah, 'cause you, you talked about like the cost of, of, of forking to the new model. Like I saw some of the… I, I think these are still estimates, pretty good estimates, but, you know, like for example, whatever the number was, I'm, I'm pulling this out of my head. It's, it was something to the effect of like a quarterly or an annual loss, either way, you know, pretty dramatic of like negative $14 billion in EBITDA, right? Like not a business, you know? So Aarin: No, it's not. Todd Kane: Yeah. Aarin: it just goes back to the whole point of just what is the outcome of this? The same thing with anything that you're doing in the MSP, you're doing in operations, you're doing in automations. The most important factor is what is the outcome of what you're doing? And if it's absolutely not defined in any way, shape, or form, then you're really just selling people a pipe dream. AI's the future, go buy it. Like, yeah, okay, um, Todd Kane: Or Aarin: prove it. Todd Kane: Yeah. Aarin: For what? Like, Todd Kane: Yeah. Yeah Aarin: prove it. Like, okay, cool. I can, I can automatically categorize my emails and I can do all this other stuff. That's very limited. I can't get AI to do my emails at all. Like, it won't do it because it doesn't know who my clients are, what I want, what I do, what I like. It takes me longer to tell it what I'm doing, and I'm just like, "Screw it. I'll do it myself." Todd Kane: Yeah, Aarin: so it's, it's just become more of an augmentation of what is the most boring stuff that we have that costs us money in the industry? How can we just focus on that? And as the technology gets better, we look at new ways as time goes on, but we're, we're still very cautious to not just go throw all our eggs into that basket. Um, and I was even with a peer group and he's like, "Hey, I was thinking of just doing an AI business unit." And I'm like, "Okay, well that's, that's pretty cool. Tell me about that." And he's like, "All they're gonna do is AI." And I'm like, "Okay, so, um, you're gonna build up this, this million-dollar business?" He's like, "Yeah." And I'm like, "Okay, so if AI goes the way the, uh, the World Wide Web did, and we have a crash, what are you gonna do when half your company's, uh, bankrupt?" And he said, "Well, I never thought of that." And I'm like, "You gotta think about stuff like this. What is the outcome of you having an only AI? You have to mix it in with some RPA. You have to have a little bit of flavor in there." Like, I've been preaching that for a while. RPA is the superior of the two right now because it doesn't need a graphics card to process, and it's not hallucinating telling you, "Yes, I love you. You're correct." Todd Kane: Yeah. But I suppose we could use some AI models to help us build the RPA, you know Aarin: You could do that, yes. Prompt engineering, all that stuff. But I mean, I guess it's how many bottles of water do you wanna kill that day? Todd Kane: Depends how thirsty you are or how, how Aarin: That's right. Todd Kane: is. Okay, cool. This has been awesome, Aaron. Um, I mean, we could jam all day, but, uh, Aarin: Oh, absolutely Todd Kane: we'll, we'll give you some, some, uh, s- your day back here. Uh, so it's been fun. Appreciate your input, Aarin: I appreciate it, Todd. Todd Kane: uh, been a blast, man. Thanks. Aarin: Likewise