Bonus Episode: Air Traffic Control

 

Transcript:

Rich: [00:00:00] Running an MSP has never been more challenging. Your clients expect more, skilled technicians are hard to find, and every minute counts toward your bottom line. If you’re looking for practical ways to grow your business without growing your headcount, don’t miss the Kaseya ConnectEdge Virtual Innovation Summit.

This free virtual event is built for MSPs and packed with real world strategies to help you leverage AI and automation, streamline service delivery, improve operational efficiency, and increase profitability. Hear from Kaseya experts, explore the latest innovations, and learn how leading MSPs are using AI to reduce manual work, resolve tickets faster, and deliver a better client experience.

Whether your goal is to scale smarter, strengthen security, or give your technicians more time to focus on [00:01:00] high value work, you’ll leave with actionable ideas you can put into practice immediately. Registration is free. Visit Kaseya.com, and click on the banner at the top of the page to reserve your spot at the Kaseya ConnectEdge Virtual Innovation Summit.

And now, on with the show. And three, two, one, blast off. Ladies and gentlemen, welcome to another episode of the MSP Chat Podcast. This time, a bonus episode, sponsored by our friends at Kaseya. This is your normally weekly, this time twice weekly, visit with two talking heads talking with you about the services, strategies, and success tips you need to make it big in managed services.

My name is Rich Freeman. I’m chief analyst at Channel Mastered, the organization responsible for this show. I am joined by your other co-host, our CEO and chief strategist at Channel Mastered. His name is Erick Simpson. Erick, how you doing?

Erick: I’m doing well, Rich. The, the temperatures are moderate here today in Southern [00:02:00] California, so we’re getting a small little break before who knows when the next heat wave is gonna hit us.

How are you doing in Seattle?

Rich: Doing great, actually. We’re having very nice summary, but moderate… It’s it’s gonna be 76 today, no humidity. We’re doing awesome here in Seattle, which is of course why tomorrow morning I’m leaving town for the humid East Coast. That’s a whole other story.

I’m gonna miss out on all this good stuff here. But yeah, no, the weather here is great. And by the way, I’ll just drop a little hint. At the end of the episode, I’m gonna have a, a, a sort of summary one last thing for you here that when that next heat snap arrives might might give you some thoughts.

But that’s coming up later. For now, let’s dive into our stor- I won’t call it the story of the week, I’ll call it the story of the episode, this bonus episode sponsored by Kaseya. And this goes back… So in March, you and I attended GTIA’s CCF event in Chicago, and one of the highlights of that event every year is Carolyn [00:03:00] April, their chief researcher, gets on stage and shares her very latest sort of state of the channel research.

And one of the things that she pointed out that year is that there has been some slippage, some decline in partner satisfaction with vendors. Now, we’re not talking about, something dropping off a cliff. This is not catastrophic. Overall 74% of the partners GTIA surveyed said they are satisfied with their vendors, but that’s down from 88% just the year before.

And the percentage of people who said they are very satisfied dropped by more than half from 39% to 18%. So something has happened in the last year, and the question is what is it? And GTIA couldn’t really answer that definitively. They didn’t anticipate this decline, so there were no questions in the survey to see what was going on.

Carolyn’s theory about what was going on, and it’s a very logical one, is AI. And so AI is changing the world for partners. Partner [00:04:00] programs are have been a little sluggish in catching up to that, and therefore partners are a little less happy with partner programs and by extension less happy with their vendors I had an opportunity to speak to two people who really know the partner program landscape thoroughly.

One of them is Ryan Morris, a consultant. He’s also part of the team with us here at Channel Mastered. You know him very well. The other is an IDC analyst named Danielle Ebron. I raised the question with both of them. There’s evidence of slipping satisfaction with vendors. There’s the possibility that vendor partner programs are to blame, and a theory that says AI is the issue with vendor partner programs.

And they both it was different calls with the two of them, but they both said pretty much exactly the same thing, which is that AI is contributing to an existing problem rather than creating a new problem. That, in fact, there has been dissatisfaction with vendor partner programs for a number of years, and a number of issues contributing to that.

For starters [00:05:00] and this is something we talked about on the show with Ryan almost exactly a year ago in an interview we recorded at the GTIA ChannelCon event the old model where a partner just sold some stuff and got some credit for it in the partner program no longer really applies.

The life cycle of a deal is longer and more complex. Partners are contributing at different stages. They’re influencing deals that they don’t necessarily transact with a vendor. They want credit for that, and there are a lot of partner programs out there that don’t measure that, don’t reward that. This is an ongoing source of dissatisfaction spanning multiple years.

Another one the, traditional partner program was basically oriented towards resellers. And these days there are plenty of resellers still out there, but there are also MSPs, there are software developers, there, there are a whole bunch of different partner categories out there, and most partner programs split these functions out across different programs.

According to IDC data, the average channel partner today [00:06:00] wears 2.8 of those hats, plays 2.8 of those roles, and they want one program that measures and rewards them for all of that in one place and in one way, and there are programs out there that are doing that now but there are a lot of programs that aren’t doing that right now.

And so these are ongoing sources of dissatisfaction with vendor partner programs that spill over into dissatisfaction with the vendors. Here’s where AI enters the picture. We’ve been talking about this a lot on the podcast. AI is compelling MSPs to become consultants to their clients, to be that virtual chief AI officer to identify workflows in need of optimization, deliver solutions.

It’s a whole new kind of business model that MSPs are gradually awakened to the fact that they need to adopt and we talked about this in the context of the PAX8 Managed Intelligence Provider idea at their event not too long ago now the partner programs out there, they’re really good [00:07:00] at delivering product training, whether t- sales or technical.

They’re pretty good for the most part at acknowledging the fact that a lot of their partners are service providers and giving them enablement around providing services. Very few of them yet Pax8’s actually one of the rare exceptions, is helping their partners understand and adopt one of these new AI forward consultative outcome-based business models.

And so that is a source of dissatisfaction with partner programs and with vendors. It is contributing to ongoing dissatisfaction. As Danielle said at one point, you can’t blame the whole thing on AI. It is a piece of the story, but it is a bigger story. And then the q- the question is what do vendors do about it, Erick?

Because it is very easy for us to identify these issues, but as we will recall from an interview we did in this show with the people responsible for Cisco’s new partner program, which does reflect the whole complex deal cycle and the 2.8 hats [00:08:00] everyone, it is a modern partner program, but it was quite an undertaking to build that.

And there, there was pushback after they did. It’s hard for the vendors to do that. And so to overcome this, this satisfaction issue with the partner programs and just with the vendors, more of these companies are gonna have to take on this difficult process.

Erick: Yeah, a couple things come to mind Rich. I can’t remember when we were at that event, if Carolyn April actually s- let us know what the samp- like the, the demographics of the folks that answered that survey.

Was it across all GTI members? Was it, was there segmentation between MSPs versus resellers? Because I think there might be a little bit of a, a difference there in, in response. Do you recall?

Rich: It was definitely a channel survey, not an MSP specific or reseller specific. And yeah, it stands to reason if they were to do the, the breakouts for us, you would see some [00:09:00] variation.

Erick: Yeah. I would imagine maybe MS- because MSPs, are different and want different things than resellers or other types of indirect sales channels. So that’s one question. Be interesting to to maybe noodle on a little bit. And we know from experience, Rich, we know from experience that the larger the organization is, the harder it is to modernize it to what today’s MSPs and partners expect from a channel program.

So it’s tough moving the ship the bigger it is, right? To try to make that, that three-point turn maybe. Maybe it takes six points sometimes the larger it is. It would also be interesting, to know what vendors or what categories of vendors. In your analytical mind, Rich, I’m sure this would be great, juicy data for you to parse and go we can infer some things here.”

So those are some of the questions, but there’s no doubt that everyone is playing catch up Rich? To try to [00:10:00] lead this new AI kind of movement and opportunity and meet the demand of the end customers, because I think that’s where it starts, is like the, the business owners and these businesses want to adopt AI.

They’re adopting AI. Their users are adopting AI, whether they fess up to it or not. And how do we as an industry take advantage of an industry made up of distributors, vendors, partners, every flavor of partner, the more consultative we need to become. I was having a conversation earlier with with a client of ours and, it was a conversation around The more consultative that the sales conversation has to be and the more strategic the deliverable needs to be, MSPs need help getting to that level of maturity, [00:11:00] right?

It’s no wonder that we see organizations like Pax8 come out with the MIP program and other vendors kinda leading with this AI first initiative. We have to do more than just build AI into the platforms and the applications that MSPs are using to run their businesses. We have to teach MSPs how to become these consultative VC A- AIOs, EIEI AIOs, if as it were, in order for them to to more quickly adopt AI themselves and then convert that into more than just a, a cost-saving or an efficiency improvement or a process improvement platform for themselves, but also to monetize that by delivering these services with the appropriate security and guardrails and guidance for the customers that are demanding this of them.

Rich: Yeah, and that is the essence of what Ryan and [00:12:00] Danielle were talking about basically, is that the partners need to make a change to a consultative model which historically most of them have not embraced before. They don’t quite know how to do it. They’re looking for help to their vendors, and very few of the vendors are providing that right now.

And so really the, the moral of the story for the vendors is it’s not that people hate you and they hate your products. They are hungry for something that you’re not giving them. And the way forward it’s not an easy way forward, but the way forward to fix your SAT issue partner SAT issue, is to help them with this transition.

And, frankly, Erick, it’s gonna be part of the job is going to be on the side of the vendors figuring this stuff out. We did– We had a conversation with Rex Frank at the Pax8 event and he and his team over there, and they developed the Managed Intelligence Provider curriculum.

It was a year of really hard work to kinda figure out what to teach the MSPs who wanted to become MI. There are a lot of vendors [00:13:00] who are gonna have to go through that, but that is the only way that you fix that piece of the problem. And then the other thing you gotta do is give people more recognition for the various roles they play in the life cycle of a deal and consolidate the different partner identities.

So again, easier said than done, but that’s that’s what vendors– what partners want and what vendors need to do. As long as we are talking about what partners are doing, need to do with respect to AI your tip of the week deals directly with that basically in terms of how that consultative dialogue with a client begins.

Erick: Absolutely, Rich. And, speaking from experience as a former MSP, I will tell you that the last thing that you want to hear from a client is, “Hey, we went with this other organization because you are not meeting this need for us,” right? So it dovetails perfectly into the conversation that we’re having [00:14:00] about AI.

If we want to make certain that our clients feel like we are leading them in their business growth journey then we have to be doing that. And the first step in that, Rich, is to ask every client how they’re currently using AI. There has to be that conversation now, and if you have not had it with your clients yet, please begin, because someone is.

We know that there’s a lot of compet- that there are a lot of competitors out there that aren’t even MSPs that are reaching out to our customers, speaking as a former MSP, and our clients, and introducing these AI forward services. So we have to be leading that if we don’t want to let comp- competitors into the henhouse.

So asking clients how they’re using AI is the first step, but trust but verify, right? So you have to inspect what’s going on because in my [00:15:00] experience outside of AI, any other technology, we know that, that- Individuals will do what they think is right. In the absence of process, they will do what they think is right to get their job done.

And if an individual thinks that they should be using ChatGPT on a free subscription with no guardrails to get their job done without having any kind of guidance or process or procedure or accountability, they’re probably gonna do that, and we shouldn’t blame them. We should be identifying what our staff needs to do their jobs and then delivering it in a, a structured and secure manner.

This is the business owner now thinking, right? But do they think of their MSP relationship in the way that they should in this specific instance? They may be the primary violator, the business owner, the business leader, right? These are the folks that kind of fall prey more often than not when we see these surveys come back, Rich, of [00:16:00] who fell for the phishing email.

Eh, it was, somebody in, on the executive team, right? So it’s gotta be a conversation that has to be had at the highest level, understanding and identifying not only how they’re using it, but what is it that they’re trying to achieve. What does success look like for that organization in that, leader’s mind when leveraging AI?

And then what risks are inherent in that process. So developing a simple kind of an assessment, and we know lots of vendors provide their partners with AI assessments that they can run with their clients just to get a feel for where things sit. Just a governance assessment, a checklist so that you can actually confirm or, in probably more cases than not, reveal to the leadership how many employees are actually using different versions and flavors of AI, and the [00:17:00] shadow AI, the AI that’s being injected into the applications and the platforms that are being used in the environment.

We have to look at everything because only then can we leverage that conversation into a governance conversation and reveal what the risks are for the business, and then what we need to do to step up the security and the guardrails to ensure that we don’t have data leakage and exposure, not only to the internal staff of company data, but to training the AI model kind of things where, you can do a Google search and find things that have been put into these, these models.

I just read an article the other day about, some data leakage where it was not supposed to leak things, we talked about the OpenAI Hugging Face situation that’s going on right now, right? It reminds me of that j- in Jurassic Park, remember when what’s the term he says?

Life will find a way. AI will find a way if you leave it to its own druthers. And, employees have to be trained [00:18:00] and governed in terms of, helping them not violate any kind of privacy requirements and things like that. And then just schedule a str- a strategic AI discussion.

This is the first step to really saying, we are gonna become your VC-AIO. We’re your VCIO, your VCTO, VCSO, maybe. VC-AIO is a way for you to again add more strategic value to the conversation, having more conversations in the boardroom, and then delivering strategic value to that business owner in a way that they may not have perceived you to have been delivering it prior to that

Rich: Yeah, it you were absolutely right before, Erick, in that your tip of the week dovetails perfectly with what we were talking about before, because we were talking about how MSPs need to make the transition to having a consultative relationship with their clients and that this is difficult to do, and this is why the partners out there, MSPs and beyond, are hoping vendors can help with that.

So it’s a process. It takes time, [00:19:00] basically, to make that transition, and what you’re pointing out is it’s a dangerous idea to just wait during that process and not talk to your clients about AI at all. Go ahead and initiate that conversation. Now, you may not feel like you’re ready to really do sophisticated things around AI with them, but there are things you can do and things that they will need to do to prepare themselves for those more strategic AI solutions, and governance is a perfect example of that.

So you can come in, and you can do a discovery inventory, basically, all the A- AI, shadow AI, authorized AI that’s u- being used across the organization. You can lock out the stuff that shouldn’t be used. You can set up the guardrails and the pol- This is all valuable work and it keeps you engaged.

It gets the AI conversation going. There’s a security conversation that needs to be hav- had. You can have that conversation now. The third piece of what you were talking about, the AI strategy discussion, I would just say, look, if you’re not feeling confident yet to have that conversation, maybe that [00:20:00] third piece comes a little bit later.

But don’t wait. Don’t wa- There are people calling into your accounts now, and as, as we’ve also talked about on the show, there are plenty of end users out there calling out to people who are not you and saying, “I want AI solutions. Help me out.” So you don’t have time to wait, and there are things you can do today that will add value to the client and keep you in a strong trusted advisor relationship with that customer.

Erick: Said, Rich.

Rich: Folks, Erick and I are gonna take a quick break on this episode of the MSP Chat Podcast, sponsored by Kaseya, and when we come back on the other side, we are going to be joined by Kaseya’s Paul Burke. He’s a VP of engineering over there. He is responsible for their digital workforce, basically the, the AI technology that automates the work MSPs do out there.

He is a super interesting guy as expert as anyone you will find on this extremely relevant and timely topic for MSPs. He’s gonna be with us to talk all about that right after [00:21:00] the break, so stick around

Welcome back to part two of this episode of the MSP Chat Podcast, sponsored by Kaseya. And this is our spotlight interview setup. And folks, this is a good one. I’ve been looking forward to this one for a while because we are gonna get into the weeds of AI and RPA and automation for MSPs with someone who knows a ton about that subject matter.

I know him as the father of Kaseya’s digital workforce. His name is Paul Burke. Paul, welcome to the show.

Paul Burke: Thank you so much. I’m looking forward to the discussion. Glad to be here.

Rich: So Paul, I gave a very brief and deeply inadequate introduction there, but tell folks a little bit about who you are and your role at Kaseya.

Paul Burke: Yeah, sure. So I have a deep experience in, in hyperscale systems. I came to Kas- and AI, and came to Kaseya really to look at the way [00:22:00] the industry was transitioning for the MSP, and how could we really get ahead of that, given the breadth and depth of the MSP portfolio that Kaseya runs, and leverage this agentic workflow and these agentic workflows to really benefit the customers.

And so I came on board to do that. We’ve done that, and we’re very excited to release it, and really thrilled about the results.

Rich: So there is a a tendency in the industry to lump together AI and RPA, which you folks at Kaseya refer to more often as deterministic reasoning. But, AI, RPA, it’s all automation or hyperautomation.

It’s all basically the same thing, and obviously they are not the same thing. So from your perspective, talk a little bit about the differences between those two things and why MSPs need them both to automate effectively.

Paul Burke: Absolutely. So we wanna use probabilistic [00:23:00] systems where the input is ambiguous, and we wanna use deterministic systems where the output needs to be defensible.

Applying one where the other belongs is the most common architectural misstep I see, and it’s usually invisible until the solution is in production. So what do I mean by this? A language model is a judgment engine. You hand it a ticket that says, “

Rich: Hey, the email’s being weird here,” and it infers

Paul Burke: meaning.

Nothing else does that well. But it’s probabilistic by construction, so if we ask it twice, you get two different answers. That’s the trade-off. You get judgment, and you give up consistency. A deterministic system has the opposite trade-off. The same inputs, the same output every time at almost no cost. What it cannot do is handle a case nobody anticipated.

This is the challenge with traditional RPA, and that is why classic MSP automation breaks, because rules are brittle at the edges, and the edges are really where our technicians actually live. And the reasonable expectation, the one I hear most often, is that AI [00:24:00] just replaces the rules. It does not. They solve different halves of the same problem.

Reading a ticket is a language problem, so a model belongs there. Ranking 12 technicians by skill certification workload is a math problem, so math belongs there. Ranking should not hallucinate. Availability should not be guessed at

Erick: So Paul, to make this more concrete for our listeners, Kaseya uses what I think is a cool term, reasoning mesh, involving four separate agents during ticket triage. What do each of these agents do? How do they work together and how do they come up with the correct outcome w- through that interaction?

Paul Burke: Yeah, great question. So when triaging tickets, we have four agents, two paradigms, and one coordinated decision. Each agent is an expert exactly one job, which is what makes each agent accountable. [00:25:00] So the impact agent makes the call a technician makes in the first three seconds of reading a ticket: priority, issue type, subtype, and severity.

It’s a language model because understanding what a user means is a language problem. The skills assessor agent maps the problem to the specific skills and certifications needed to solve it. Also a language model because getting from nobody in the May- Miami office can print to print server administration, endpoint deployment policy, and branch network troubleshooting is a semantic reasoning problem, not a lookup table.

The smart assigner agent ranks every available technician against those skills, their certifications, and their performance history. It’s deterministic because ranking’s math, and math should not hallucinate. The capacity planner agent checks who is actually available right now, time zone, shift, current workload.

It’s deterministic because availability is measurable, and measurement should not guess. So the language agents establish meaning. The deterministic agents turn meaning into [00:26:00] a defensible decision.

Rich: So the language agents never touch assignment math, and the math agents never interpret a ticket.

Paul Burke: Now, there is a part of your question, that your question left out, and there’s a fifth agent, and this is the key one to remember, the critique agent.

It classifies nothing and assigns nothing. So what’s its job? Its only job is to challenge the other agents. When the impact agent says, “This is a P1 outage,” the critique agent pushes back, “Are you sure? What signals did you weigh? What did you miss?” The critique agent runs on a different underlying model than the agent it critiques, and this is deliberate because if the critic and the writer share the same model, they share the same blind spots.

And this approach is a meaningful part of why we see 95% classification accuracy in production, and this is measured against what technicians actually accept. Every decision has already been argued with before a human ever sees it

Erick: I wanna jump in and just respond to what you just shared because it [00:27:00] follows, what my experience is.

I came from the enterprise building out call centers and service desks for Fortune 1000 organizations, and the agents that you just described are taking everything from identifying, classifying, routing, assigning. But one thing that is new to me i- to hear is this critique agent, which is very unique.

I think that’s a gap that most MSPs don’t have a role or an awareness of during incident management in their service desk. So I’d love to hear your feedback on that and how you guys came up with that. Was that feedback from partners? Was it something that came in from, you know, general ITIL? Where did that come from, Paul?

Paul Burke: So this came from the fact that when we use language models, they’re not deterministic. Because they’re probabilistic, what we said is this is a probabilistic model, therefore, we need an oversight on that model.” Just if you have a junior [00:28:00] technician, you want someone more senior to oversight their work initially when they come on board and to make sure they’re making the right decisions.

And what we learned through this experience is we’re able to drive much higher degrees of confidence in the results by adding this critique loop. And so think about it. We started, we took the LLM output, the probabilistic output, and we said, “We need to be able to judge that output and critique it just like we would an intern.”

And so we treated it like an, a, a new intern into the MSP, and we treated the critique like a senior inter- a senior MSP technician that is evaluating and critiquing that work, and that was our approach.

Rich: Yeah, that’s an architecture I run into a fair amount in these sort of sophisticated AI models, and sometimes what you’re calling the critique agent is referred to as a judge or a ref.

But the idea that there’s something looking at what the AI is doing and validating that it’s what it should be doing shows [00:29:00] up a fair amount. It’s a really interesting concept. Speaking of interesting concepts, so w- we talk a lot in the industry and even on this show about the difference between built-in and bolted-on AI when it comes to a, a platform of the kind Kaseya has.

Talk a little bit a- about the difference between built-in and b- bolted on and how it plays into this concept that you refer to as the integration tax.

Paul Burke: Absolutely. So built-in versus bolt o- bolted on is an interesting question and is how you opened, because of about two weeks ago, every serious vendor in this market claims to be built in.

When everyone says the same word, the word stops carrying meaningful information. And so from my perspective, built-in should mean the intelligence comes to your workflow. If built-in turns out to require you to consolidate onto someone’s platform before the intelligence works, that’s not native intelligence, that’s bolt-on migration.

The integration tax is what you [00:30:00] pay when your intelligence and your data live in different places. And so how does that break down? You pay it in context loss because every integration is a translation, and every translation loses meaning. You pay it in fragility because connectors break and schemas drift.

You pay it in technician hours spent teaching a tool things your platform already knew. And you pay it in compounding nothing because the vendor’s model gets better across their whole base while your specific history never becomes your asset. And I do wanna call out a side note here that agentic AI does not reduce that tax.

It actually raises the stakes. So an agent that suggests on bad context will give a bad suggestion ideally that a human catches. An agent that acts on bad context takes b- a bad action at machine speed, which is really why I think the useful question is not who claims to be native, built-in versus bolted on, but it’s how you validate it.[00:31:00]

Erick: So you have a longer-term vision, Paul, for where AI tooling of the kind we’re discussing is headed based on what you call a control plane. What is a control plane in this context, and how will it impact MSPs?

Paul Burke: Absolutely. This is something I’m very interested in. So I see the control plane as a governing layer to all MSP service management, not the front end to Kaseya.

The control plane is the dispatcher for your digital workforce. It decides which digital specialist handles what each one is allowed to do on its own, and where it had to stop and ask. So it governs the work. It does not sit in the middle of the work. So when I think about the control plane, it consists of identity, intelligence, and orchestration in one layer that can drive any tool in your stack, ours or anybody else’s.

And why this distinction is important [00:32:00] now is because the connector is being commoditized. For 20 years, integrations were a moat in this channel. Now the MCP protocol is standardizing how any agent reaches any tool Which in my opinion raises the real question. If connectivity is commoditized, where does the durable value go?

It goes to two things which a protocol deliberately does not provide. It goes to knowing which signals matter and how they relate, and governing what an agent is actually allowed to do. And so the second one is really what the control plane is. And I think it’s why the strategic question for MSPs is shifting from which AI features you buy to who is positioned to be your control plane.

Because that’s a much harder things to change later than a point tool. And so when you look at the impact on MSPs, it changes what the job is. Right now, adopting AI feels like evaluating tools. In a control plane world, it becomes org [00:33:00] design. You’re deciding how much authority to gel it– to delegate to which specialist in which client environment with what oversight.

And you’re reviewing that the way you would review a team’s performance. And so the binding constraint stops being can this be automated and becomes how much authority am I willing to delegate and can I prove it was used correctly? And so that’s how I see a control plane.

Erick: Reminds me of a term I use a lot when I talk about these kinds of things in a service desk like dispatch and things like that, and I call it air traffic control.

Paul Burke: I love that analogy. It… And it’s very much the same thing. And so when you think about a control plane, the great news is with this interoperability, if you think of the air traffic controller is the MS– in front of the MSP, it’s the any tool system in the ecosystem provide it’s plugged in can l- be leveraged by the same intelligence and orchestration.

And that’s the job of the control plane [00:34:00]

Rich: So this this may get at something that you’re just talking about there, in terms of what enables the air traffic controller, the c- the control plane to do its job well. When Kaseya Intelligence is dealing with a ticket, there are basically three things that it does.

It triages, it reviews, and then it learns. And arguably, the learning piece is the most important of those three. So talk a little bit about that learning p- piece in the process and how it enables an MSP to accumulate a proprietary training data set.

Paul Burke: Absolutely. So we’re not fine-tuning a frontier model on your tickets.

We’re composing two different assets: a platform-wide intelligence layer based on a corpus of experience, and a per-tenant intelligent layer that is the MSP’s. That distinction matters both technically and for the MSP risk posture. Now, [00:35:00] triage is the mesh we just talked about. When we get to review, this is where technicians accept, reject, and refine.

Now, in this context, review’s not a courtesy step. It’s the data generation step, and this is what most people miss. Every accept is a label positive, where every reject is a label negative with a human correction attached. You cannot buy that or scrape it. It only exists as a byproduct of your team doing their job.

And then what learn does is learn turns that into an asset. When a technician resolves a ticket, we summarize what the problem actually was as opposed to what the user recorded. We extract which skills genuinely solved it as opposed to which were nominally assigned, and that all feeds back into the intelligent layer– intelligence layer, excuse me.

These two assets compose, and so Kaseya has two decades of aggregated, anonymized ticket history behind the platform. And the reason that matters is not the size, [00:36:00] which is over a billion tickets. It’s that the specialist is useful in your first week instead of your third month, so you’re not starting from zero.

And then on top of that sits your tenant’s own layer, and this is the broad experience of your MSP, and that’s what gives you a competent starting point. Your data is what makes it yours. There is a catch in this, though, and I think it’s really important. A learning system is only as good as the resolution quality it learns from.

So documentation hygiene, which is the least glamorous discipline is actually becoming a competitive variable. So two MSPs can run identical software, and one will compound faster purely on resolution discipline. So the data and data maturity becomes a really key aspect in the learning process

Erick: So speaking of data, Paul, a lot of people believe that data is your moat as an [00:37:00] MSP, right?

So you have a little different perspective on that. What, from your point of view, drives a competitive moat for MSPs if it’s not data?

Paul Burke: Yeah, from my perspective, I don’t see data as the clear advantage. Every MSP has data. I see context as the clear advantage, and by context I mean knowing which signals matter, how they relate,

Rich: and what happened the last thousand times

Paul Burke: this pattern showed up.

And here is the number I’d put in front of MS- every MSP owner. In Kaseya’s 2026 State of the MSP Report, 48% say AI and automation is their clients’ single biggest need. Roughly 13% have turned that into revenue. That gap is too wide to be a sampling artifact, and I would encourage anyone to check it against other channel surveys out there.

Everyone in that gap has data. Data was not the constraint. I also don’t think that is a capability gap. I think it’s a trust gap. And [00:38:00] MS and– And MSP’s entire business is built on trust. Your clients hand you the keys to everything. So when you put AI in front of them, you’re being asked to warrant the behavior of a system you did not build.

Most MSPs are not willing to do that yet, and honestly, they’re right not to be because most of what has been sold to them cannot support that warranty. So context is what makes the decision correct. Trust is what lets you sell it. You need both, and the second one is harder. So I see data as the table stakes with context as the differentiator.

Trust is the product. Nobody ever bought a managed services contract because their provider had good data. They bought it because they trusted somebody. And

Rich: so three reasons data alone is not a competitive moat would be, one, everyone has it,

Paul Burke: so it’s table stakes rather than differentiation. Two, unrefined data is a liability at first.

It’s all storage cost, and it’s a breach surface, returning nothing until something converts [00:39:00] it into context to be able to make a decision.

Rich: And thirdly, data deprecates because your 2019 ticket history describes an environment that most likely no longer exists.

Paul Burke: So what actually drives a moat is the speed of your learning loop, not the size of your data archive.

An MSP that turns resolve ticket into a better next decision within hours beats an MSP with five times the history and no learning loop, and the gap widens every month. The advantage lies in the derivative. So a competitor can buy the same tools and hire the same people with similar data. They cannot buy 18 months of your compounding intelligence loop

Rich: We’re gonna have to pause this again. What’s going on? Sorry, folks

Data in the form of context, proprietary data, the MSPs data, that’s not necessarily what differentiates, but it is part of the [00:40:00] equation for them. So arguably, in this larger conversation we’re having is consolidated data, that, that initial data layer you were talking about. T-talk a little bit about the advantages MSPs can realize by virtue of having access to a platform like Kaseya’s that is connected at a native level with RMM and PSA and security and backup and has that sort of common foundation.

Paul Burke: Sure. So the most valuable signal in an MSP almost never lies in one tool. It lives in the correlation between tools. So to share an example, when a ticket comes in saying SharePoint is slow this morning, a system with only PSA data sees a low-priority complaint from one user and routes it that way. Now, if we give the reasoning layer visibility into the entire stream of events, where RMM shows a sustained memory pressure on that host starting overnight, backup shows a job that failed on the same host at two AM, and [00:41:00] secur-security shows an endpoint agent that stops reporting on that same host in the same window.

Individually, the– they’re problems to be solved. However, together, that’s a completely different ticket with a different priority, a different skill requirement, and most likely a different conversation with that client. So no AI can infer data from what it cannot see. And this is not fixed by adding four more integrations because it would be reassembling at the edge a stream that was never designed to be reassembled, with timing gaps and schema mismatch at every seam, et cetera.

So reliable cross-domain correlation happens when the data already lives together. This is what we mean by a unified data layer being the first component of Kaseya Intelligence rather than an afterthought. So aggregated history across thousands of environments is not a bigger version of one MSP’s data.

It’s a different kind of asset. It’s what lets the platform recognize a failure pattern in your specific tenant that it has never encountered before. [00:42:00] And that’s the part an individual MSP cannot build for itself on any budget, and it’s the honest argument for a platform over a point tool. I would like to call one co- common ownership.

One interesting thing, though, about common ownership is that common ownership of a lot of products is not the same thing as unified data. Plenty of vendors own broad portfolios where every product still has its own schema and its own database. And that’s the integration tax paid internally and passed on to the MSP

Erick: So Paul, everybody’s got a perspective that deals with AI around the human-in-the-loop conversation, right?

What’s your position on that at Kaseya? And then I’ll have a quick little follow-up thought.

Paul Burke: Cool. So human-in-the-loop is not a feature, and it’s not a permanent state. It’s an authority you grant one workflow at a time, and you have to be able to take it back. Our [00:43:00] position is that human-in-the-loop is graduated authority with full responsibility.

And so the industry has collapsed effectively three different things into one phrase. Human-in-the-loop means a person approves each decision. This is where trust starts, and it should be the default. But it’s temporary because if you stay there forever, you’ve not automated anything. You’ve added a review queue to a review queue.

Human-on-the-loop means a system acts and a person supervises an aggregate, sampling, watching trends, getting alerted on the unique cases. This is where the sustained value actually lives. And then we have a third, which is human-in-the-design, means people sit at the guardrails and boundaries and revisit them on a case, and this one never ends.

So when we look at taking our triage specialist as an example, it follows this position. A new triage specialist starts in human oversight by default, not opt-in. It shows its reasoning because this is not a black box and should not be run as one. [00:44:00] When an MSP has validated accuracy on their own tickets, they flip it to run in the background, and it can go back the other way anytime, with the reason being the platform should earn trust, not demand it

Erick: I like that human on the loop distinction and the human in the design.

And calling back to when we were talking about, contextu- contextualizing the data, and you made a comment about making better business decisions. And I’m starting to feel like everything that we’ve been talking about and what you’re delivering is helping business owners and technicians and engineers make better business decisions.

That’s the capability it feels like you’re building here. So I’d just like your thought on that.

Paul Burke: Yeah, no. So th- that’s spot on. So because if you look at the evolution of where we’re headed, right now our question primarily centers around one agent and one technician. Within two years, it will be many agents holding onto privileged access across hundreds of client tenants, [00:45:00] increasingly over open protocol.

At that scale, human in the loop cannot mean a person watching every action because there’s not enough human attention in your company to do it. So it really starts to mean the business benefit, which is policy, which is identity per agent, authority scope per workflow, every decision observable and rollback that works.

So oversight has to become an architectural property rather than a staffing model, and this is really the ability from a control plane perspective, tying it back, to drive solid business decisions at scale

Rich: Paul, super interesting conversation. I really appreciate you taking some time out to to have it with us here.

For folks who would maybe like to get in touch, learn more about you, or learn more about the topics that we’re talking about here where would you direct folks?

Paul Burke: Yeah, right now f- for me, I would direct folks to, LinkedIn if they wanna connect and start to have some of these [00:46:00] conversations.

Absolutely happy to do that. And I think that’s a great starting point.

Rich: Okay. Once again, Paul, thank you very much for joining us here on the show. Folks, Erick and I are gonna take a quick break now. When we come back on the other side we’re gonna share some final thoughts about this interesting conversation with Paul Burke.

We’ll have a little fun. We’ll wrap up the show. Stick around. We will be right back.

And welcome back to part three of this episode of the MSP Chat Podcast, this bonus episode sponsored by Kaseya. And thank you once again to Paul Burke VP of engineering at Kaseya. It, there are probably about a dozen things he said at least that we could kinda dive into here, and we shouldn’t do that in the interest of time.

I’ll just point out for myself, as somebody who thinks a lot about data and the importance of data in AI, and specifically in the world of managed services and systems of record [00:47:00] and all that kinda good stuff Paul has a wonderfully nuanced take on data that just introduces some complexity to it in terms of the value of the unified data layer that a company like Kaseya can provide and the value of that proprietary data set that belongs exclusively to an MSP, and the power that comes from combining them.

We could go down that whole rabbit hole, but it helped me certainly un- understand, think better, more clearly about the role of data in AI and in managed services. It was great having a bit of a conversation there about trust and the role of trust in this whole conversation. It echoed stuff we heard from Pax8 before in terms of that being the key differentiator for MSPs.

You’ve already got, if you’re doing your job well, you’ve already got your end user’s trust. That puts you in the, The lead spot gives you an advantage in terms of being a, a AI advisor to your clients. It’s all about trust. You’ve got that. And then I particularly enjoyed, we were talking about built-in versus bolted-on AI, and he [00:48:00] got into his definition of what it means to be built in.

And he said “Built-in should mean the intelligence comes to your workflow,” basically instead of you feeding your workflow into this other technology, this other product, this other add-on. If the, the tool that you’re using has that AI functionality built in, it’s coming to you. It’s fitting into how you do things.

A- again, a lot of different things that we could talk about, but I really enjoyed that conversation.

Erick: It was a great conversation. It made me think differently about, things that I had assumed before. And a couple of things that really stood out for me, Rich, was, we’re we talk a lot about the value of the data and, he really made us think differently.

It’s not just the data, it’s the context, how you can contextualize the data. That was a good aha moment for me. And then, just the other point where we started talking about, having disparate solutions, like having your Goldilocks [00:49:00] bolted together solutions, and then each of those different applications in that bolted on kind of framework that you were just talking about, then e- each one of those becoming its own unique potential vulnerability point for an attacker, right?

And leading into the rest of it. So that was interesting to me, too. And then I also appreciated, the, the confirmation that my moniker for, kind of dispatch and oversight and all that is like air traffic control. I didn’t think about that additional, a skill or agent that says, “Oh I also have to verify all these other things.”

It it’s pretty amazing how complex you have to build a, A process with AI to replicate what humans do, right? So when I think about that conversation compared to how fast I think we’re going, I’m like, yeah we’re probably still a long ways away before we can actually replace a human because there are so many things, and only folks like Paul [00:50:00] and his team and really smart folks are thinking that way.

We’re just there, talking about the opportunity and the risk and all this stuff, but man, behind the scenes when you think about what it takes to build that, it’s pretty astonishing.

Rich: It is remarkable, and it’s remarkable the degree to which companies like Kaseya, and I keep thinking of some of the security vendors we spoke with at the RSA Conference, the degree to which they can hide that complexity from the user so that there are like, f- five agents basically taking place in ticket triaging, and you don’t really need to know that.

You don’t- you’re not seeing the interaction. It’s just doing that triage work for you. Yeah it’s it’s a remarkable thing the the complexity and the speed with which the AI can do all this stuff without you even really being aware of it. So yeah, very interesting time.

Erick: Yeah, hearing, hearing that last comment of yours made me think of that’s what MSPs are fighting all the time. We don’t see you guys anymore, so you know, what are you guys up to? The same thing here. It’s like that’s the magic of it, it’s making it, transparent to the [00:51:00] users.

Interesting stuff.

Rich: Folks, that leaves us with time for just one last thing, and here, Erick, is where we come full circle. You were talking at the top of the episode about how the temperatures are reasonable in Southern California where you are right now. They might be spiking back up again.

Here is an exciting job opportunity that, among other things, might help you beat the heat a little bit, because it turns out that the folks at Snickers Ice Cream, and specifically the folks at Snickers Ice Cream who are responsible for Snickers Ice Cream Minis they’ve got a job opening right now.

They are looking for somebody to be their chief online chill officer. This, Erick, I’m reading from the press release, is a paid two-week mini experience designed to bring or help bring more positivity, fun, and chill to people’s feeds, all while bringing to life the brand’s humor and product truths.

You do, and y- there, there are some requirements in here. They, y- they’re looking for somebody who’s fun, who’s good on camera. It’s apparently gonna be a lot of social media work involved in this. It [00:52:00] just lasts a few weeks. And I am guessing, it doesn’t say, but I am guessing, and this is the only reason I might even theoretically be interested in this job, there’s probably some free ice cream in it for you, too.

Now here is the thing, and this is coming to you straight from your friends at at Channel Mastered and the MSP Chat Podcast, because this show is scheduled to go out on August 4th, which means if you are watching or listening on the day this episode became available, you have got until 11:59 PM Eastern Time tonight to apply.

And there, there’s a link in the show notes. You can go and get that done. You’ve got just hours to go if you want to be the chief online chill officer for Snickers Ice Cream. So get to it, folks.

Erick: Let’s hope that it’s it’s compensation plus ice cream instead of ice cream in lieu of compensation, Rich.

Rich: And if you get the job, cut us in for some minis, please. Just do the right thing.

Erick: There you go.

Rich: Folks, that is all the time we’ve got for you on this special bonus episode of [00:53:00] MSP Chat, sponsored by Kaseya. We thank you very much for joining you, as do the good folks at Kaseya. We will simply remind you now, as we always do, that this is both a video and an audio podcast, which means that if you are watching on YouTube but you’re into audio podcasts, go to Apple, Google, Spotify, wherever you get your audio podcasts.

If you’re listening to us but you’d like to check us out on video, go to YouTube, look up MSP Chat. Wherever it is you find us, please subscribe, rate, review. It’s gonna help other people find and enjoy the show too. This show is produced by the great Riley Simpson, part of the team with us here at Channel Mastered, where we help vendors build, grow, optimize thriving MSP channels.

You can learn all about the many ways we do that at our website located at www.channelMastered.com. Channel Mastered has a sister organization called MSP Mastered. That’s Erick and his team working with MSPs to help them grow and optimize their business. You can learn more about that at www.mspmastered.com.

So once again, thank you very much for [00:54:00] joining us. We’re gonna be back later this week with another regular episode of the show. Until then, we will simply remind you, as we always do, you can’t spell channel without MSP.