Exploring AI in Project Management: Automation, Data Governance, and Future Trends with Mike Psenka

[00:00:01] Dave Sobel here at IT Nation Connect with another bonus episode of the Business of Tech. I'm talking with Mike Psenka of Movevilla and we're going to talk a little bit about their advance. Mike, thanks for joining me today.

[00:00:12] Thanks Dave, thanks for having me.

[00:00:13] I'm super excited to talk to you because in particular your product focuses on project management. What really intrigues me about the approach is how AI and automation driven it is. So I kind of have to ramble through some of my topics on that.

[00:00:26] And I think I want to start a little bit with, you've got some real experience now as an organization implementing AI, machine learning, automation, like whatever buzzword we want to talk about.

[00:00:37] What I actually want to hear from you is like, hey, with those experiences, what are the things you're finding that it's best at and the areas where it's best implemented?

[00:00:47] That's a great question. I think there are a lot of things, like you said, we've been doing this a long time. We're sort of not new to this space.

[00:00:53] Our AI is maybe a little different than everyone's hearing in the LLM space. We're very domain specific around this idea of project management and the directed acyclic graph.

[00:01:02] And so there were a lot of things I think we expected to deliver to the space. I think the surprise, and maybe it's not going to be a surprise to you because you were in this space.

[00:01:09] I think the surprise was how it's helped relationships, right?

[00:01:13] Okay.

[00:01:13] So at the end of the day, you know, AI and tech is amazing and like, oh, we're an AI and automation vendor.

[00:01:18] And, but this is not going to ever stop being a relationship business. Right.

[00:01:23] And so I think, you know, as we came into the space, we were in retail and finance and construction, and there were automation components.

[00:01:29] But, and those are all relationship businesses too, make no mistake, but this is fundamentally a relationship business about trust and fidelity of communication for their customers and their process.

[00:01:39] And so we were very fortunate and had a lot of customers and kind of went back to them and through our customer success process, like, hey, how did this AI and automation help you?

[00:01:46] Yeah. You know, we've reduced a lot of time. We saved, but we're delivering on time more.

[00:01:51] But several CEOs came to me and they said, look, what's been great for me is just our customer satisfaction scores around this stuff have skyrocketed.

[00:01:58] They said, well, is it because you're delivering on time? And the AI's like, well, yeah, we are more.

[00:02:02] He goes, but honestly, it's more important when we're not delivering on time because there was a supply chain problem.

[00:02:08] And what this has helped us do is manage expectations.

[00:02:11] So instead of us like scrambling at the last minute to go, oh, we missed a date or a problem.

[00:02:16] So let, so let's unpack that a little bit because that's the end result of the product you've put together and the effect it's having.

[00:02:22] But I want to go a little bit, then one step back in that because the technology is going to be good at some things and not good at others.

[00:02:29] So if I think about this from an AIML perspective, so tell me a little bit then about like to get that result, you've discovered that it does certain things well.

[00:02:38] Give me a little bit of insight into what that means.

[00:02:40] So I would say accuracy of timelines.

[00:02:42] If you were to distill everything away for them, when you think about project management, automating timelines.

[00:02:48] Yeah.

[00:02:48] If you have bad project timelines, it contaminates all this other business intelligence.

[00:02:54] It screws up scheduling, customer communications, resource forecasting, financials, profitability analysis.

[00:03:00] So we always say like the root of all of this is accurate timelines.

[00:03:04] If you have an inaccurate timeline, it screws everything.

[00:03:07] So let me unpack that a little bit because in there, what I want to understand is the technology implementation versus like what we were doing with humans.

[00:03:15] So if I think about it, if I think about a traditional human interaction with this, I put a project manager on there who's supposedly an expert in all of that.

[00:03:22] Yep.

[00:03:23] Is the use of technology the fact that they're able to see patterns better or is it an element of they're better at calculating timelines?

[00:03:30] Is it better of like, like what is the technology itself?

[00:03:34] Like if I think about AIM, again, AIML automation, whatever I want to call that thing.

[00:03:39] What is it better at doing than the human counterpart is it?

[00:03:43] So you have to let me know how much you want me to geek out on this and talk talk?

[00:03:45] I actually want you to geek a little bit here.

[00:03:47] Okay.

[00:03:47] Okay.

[00:03:47] Cool.

[00:03:47] So the technology is very domain specific around something called a directed acyclic graph.

[00:03:52] A graph is connectivity between vertices and edges, right?

[00:03:57] Facebook is a graph.

[00:03:58] Search engines are graphs.

[00:04:00] Mapping softwares are graphs.

[00:04:01] Projects is a graph.

[00:04:03] Each task, right, is a vertice and there's an edge which is a connection or a dependency between those.

[00:04:08] And projects are called directed acyclic graphs, right?

[00:04:12] So you have a task.

[00:04:13] I've assigned you to update a firewall, right?

[00:04:16] That is a point in that graph.

[00:04:18] You have responsibility.

[00:04:19] You have a work estimate.

[00:04:20] You have a time.

[00:04:20] You have dependency.

[00:04:21] There are tasks that occur before that.

[00:04:23] But there are a lot of variables and components in a large-scale project.

[00:04:26] A lot of discrete math, management, and analysis of that that has to be tracked to accurately say,

[00:04:33] at the end of this project, we're going to deliver on April 12th.

[00:04:36] Okay.

[00:04:36] Well, how do we know that?

[00:04:37] But PMs can be great PMs.

[00:04:40] Make no mistake.

[00:04:41] We're not dissing them.

[00:04:42] But what is this better?

[00:04:43] Yes.

[00:04:44] But this, at the end of the day, its ability to analyze across a portfolio all of these DAGs,

[00:04:50] identify structural defects, flaws, process, capacity conflicts for individuals,

[00:04:55] by connecting and consolidating all of the disparate data.

[00:04:59] I got your tickets.

[00:05:00] I got your schedule.

[00:05:01] By the way, your mother's getting her hip replaced next week, and you're going to be out for two weeks.

[00:05:06] But there was a project that you were working on before that got delayed that got pushed into that two-week period.

[00:05:09] So now you're a day late.

[00:05:10] You're not going to be a day late.

[00:05:11] You're going to be three weeks late.

[00:05:13] And by the way, that's going to screw up these other three projects.

[00:05:15] There isn't a human being in the face of the planet, and certainly not in the world of MSPs, that can track all that.

[00:05:20] So what it's really, really good at doing is consolidating all of this disparate data, continuously analyzing it, just kind of like an RMM.

[00:05:28] Okay.

[00:05:28] So at the end of the day, it's sort of tracking, and a lot of our customers say that.

[00:05:32] They're like, this is kind of like RMM for project management.

[00:05:33] Well, the reason I'm asking the question to set the context a little bit is I know a lot of my listeners are struggling with understanding beyond the marketing.

[00:05:43] Everyone now claims there's AI.

[00:05:44] Okay, cool.

[00:05:45] What they're looking for is understanding which solutions are best applied by AI.

[00:05:51] And I want to make sure I'm reflecting back what you've just told me.

[00:05:53] It seems like what this is best at doing is this is doing really great at pattern matching so that it can see patterns either before they've happened before,

[00:06:02] or it's being able to pull in such a diverse set of patterns to them that a human couldn't see alone,

[00:06:08] or it's just simply too much for an individual to hold in.

[00:06:11] That's right.

[00:06:12] Is that a fair way of kind of reflecting it back?

[00:06:14] I think that is a good reflection of it.

[00:06:16] And the other really important point of this is we had to choose a white box interpretable AI of this.

[00:06:21] We couldn't afford to hallucinate and say, you're going to be 17 days late in this project, and you look at it and go,

[00:06:26] I have no idea why you're telling me I'm going to be 17 days late in this project.

[00:06:30] I think that's wrong.

[00:06:31] It's really interpretable because you have to go, and now I understand why, and now I understand how to mediate it.

[00:06:36] So I want to understand a little bit more of what you just said there.

[00:06:39] Tell me a little bit about what you mean when you say we've selected a white box,

[00:06:42] and the benefits of that are your ability to control this hallucination.

[00:06:46] So everyone knows about LLMs.

[00:06:47] They know about generative AI.

[00:06:49] Right.

[00:06:49] And they know about this concept called hallucination, right?

[00:06:51] Because within those neural nets, in between the nodes, and I've got a son who's getting his PhD at Berkeley in AI,

[00:06:57] you know, and it's upsetting when they say we don't know what's going on between those nodes in that process.

[00:07:01] But in that learning process, as they backpropagate information based on valuable outputs and go through those matrices,

[00:07:07] it can hallucinate.

[00:07:08] It can come up with an answer it thinks.

[00:07:10] And by the way, as one person described it, a lot of those LLMs are like the smartest dude you know that's a bit of a pathological liar, right?

[00:07:16] Right.

[00:07:16] It makes things up.

[00:07:17] So that's called black box, right?

[00:07:19] That's a black box, and it's not interpretable at this particular point in time.

[00:07:22] So it can deliver good value and information.

[00:07:24] But if we were to dump all of our data and try to pump it into an LLM and say, tell us what the problem is,

[00:07:29] it might hallucinate and say there's a problem or an issue because it misprocesses that data.

[00:07:34] So for us, because we're using sort of expert system, well-known discrete math, internal techniques, and heuristics and process,

[00:07:41] it's interpretable because we can look at a rule set about how these algorithms are processing the data and go,

[00:07:47] we know how it arrived at, we know exactly how it arrived at this answer.

[00:07:51] And we can backfeed step by step, and we can reveal that to the user to say, here's why it's saying this, these very specific issues.

[00:07:59] It's very traceable along the way.

[00:08:00] The downside to this is it's not general, right?

[00:08:03] Okay.

[00:08:04] It's very domain-specific, and that's what I'm saying.

[00:08:06] It's domain-specific to the DAG.

[00:08:08] That's it.

[00:08:09] You can't ask it, what's the best time of the year to visit Iceland and work.

[00:08:13] Sure.

[00:08:13] Okay.

[00:08:14] There are a lot of other things in the chatbots you can do.

[00:08:15] And that's a choice of right to the right problem.

[00:08:19] Right.

[00:08:19] And, you know, again, if I'm taking my quick step back, the reason I wanted to have this conversation is that my analogy that I've been using a lot with solution providers is, look,

[00:08:27] they will be playing the role of the sommelier in this.

[00:08:29] They need to match the correct model and solution to the problem.

[00:08:33] Right.

[00:08:34] But in order to do that, you actually have to dig deeper than just saying this is AI.

[00:08:38] I actually have to understand what the different kinds mean, and I want to apply them to practical solutions.

[00:08:42] You've done that.

[00:08:43] Yeah.

[00:08:43] So that helps us unpack that.

[00:08:46] It feels like that the next area here is you'd have a deep understanding of the idea of data governance around this.

[00:08:53] Yeah.

[00:08:53] Because you probably have given some real thought about the kinds of data that go into the model and how they have to be.

[00:09:00] I'm going to use organized loosely, knowing that the joy of this area is that not everything has to be perfectly structured.

[00:09:06] Right.

[00:09:06] But it does have to be organized.

[00:09:08] Give me a little bit of the insight into kind of what you've learned through this process about how the data has to be organized.

[00:09:14] Yeah.

[00:09:14] So a lot of this, too, and I think the cool part of this is that when you have the luxury of being domain-specific, you can then bring in organizational or process wisdom.

[00:09:24] Right?

[00:09:24] So you now don't have to rely, you don't have to be data-agnostic and go, well, we just have to let the model run against any particular set of data.

[00:09:32] You have the ability to go, let's get the world's best practitioners to help inform this process and help design it so it can become more efficient and reasonable.

[00:09:40] Right?

[00:09:41] So as it delivers answers, it delivers practical answers because the structure or the building of the algorithms in the process were vetted and refined by someone who was an expert.

[00:09:51] Okay.

[00:09:51] Right?

[00:09:51] Someone who understood and said, yes, you should prioritize these over these.

[00:09:55] This matters.

[00:09:56] Yes, this is going to get revealed.

[00:09:57] In practice, we don't care about that.

[00:10:00] It's less of a big issue.

[00:10:01] And so you get less noise in that process.

[00:10:03] Okay.

[00:10:04] Gotcha.

[00:10:04] And are there any particular sort of like insights you've had around that in terms of like, hey, if we knew this a little bit better, the data would be better organized?

[00:10:13] Or there's a technique that you think about to make that process a little bit more efficient?

[00:10:19] So I guess if I think I understand the question correctly, I'd say one of the things that we're looking at now, there are two things that people have to do to unlock a lot of automation around structured work, right?

[00:10:29] Projects that maybe they're not doing today.

[00:10:31] They're entering work estimates and assigning it to people and kind of putting skills into sorts.

[00:10:35] But if they don't have a duration, like, hey, it's going to take you five hours to install this firewall.

[00:10:40] I'm going to give you a week to do it, the duration.

[00:10:42] That's a week.

[00:10:43] And the other thing is the dependency.

[00:10:44] You know, we've got to get this thing ordered and delivered to you before you install it.

[00:10:48] Those two things, if an organization isn't doing that, then they really restrict their ability to unlock a lot of automations.

[00:10:54] I think one of the things we're discovering is, is there a way we can construct those dependencies and durations without making the user do it?

[00:11:00] So if they come into the system and go, oh, I've got this plan.

[00:11:03] I never did that.

[00:11:04] We can go, well, we can look at your old data through ML techniques and go, we think we can actually construct those dependencies.

[00:11:09] We can construct those durations to make it easier, right?

[00:11:12] It sort of becomes more plug and play for them.

[00:11:14] So they don't need a PM to go in and construct everything and design it.

[00:11:17] It kind of gets the 80, Pareto principle, can we get 80% of that done for them?

[00:11:22] Gotcha.

[00:11:22] Okay.

[00:11:23] So it's interesting, again, to look at that analysis.

[00:11:25] So you've probably got some way of thinking about making decisions here, like, hey, this is particularly high impact.

[00:11:30] I can get 80% of there and make a big impact.

[00:11:33] This investment would be significantly less.

[00:11:36] I'd really like to get, particularly because you're making these kinds of decisions on a pretty regular basis,

[00:11:41] give me a little sense of the way that you frame that and help think about your own decision-making process.

[00:11:47] Like, is there a process or a criteria or an analysis that you use to make decisions that you know you feel comfortable with?

[00:11:54] Yeah, so we have an advisory board on an analytics side, sort of the math and AI side.

[00:11:58] And then we also have an advisory board on the PM side and then even in the MSP community as well,

[00:12:04] where we'll go through and say, this is the rubric, these are the rules in the process,

[00:12:08] these are the things that we think we need to identify, these are the things we think we need to prioritize,

[00:12:11] and then validate it against those three things, right?

[00:12:14] Different purposes.

[00:12:15] One, the group, the math, the AI component of it, right?

[00:12:18] Right.

[00:12:18] Then it's traditional, very rigorous PM structural process for PMO, PM point.

[00:12:22] And then on the MSP side, it is the practicality.

[00:12:25] Like, you know, we know that this is vetted structurally and sound on a math focus.

[00:12:29] We know it's structurally sound on a PM.

[00:12:31] Is this important to you and how do we prioritize it?

[00:12:33] We sort of go through that.

[00:12:34] Our product team kind of goes through that process to understand those components.

[00:12:37] So when we deliver something, they're not going, well, that doesn't matter to me.

[00:12:40] That's cool, but it doesn't matter.

[00:12:42] Right.

[00:12:42] Okay.

[00:12:43] Gotcha.

[00:12:44] I kind of want to get your take on a trend that we're tracking.

[00:12:47] And I think you'd have a really unique perspective.

[00:12:49] As somebody, you're really, you know, project management affects all of this.

[00:12:53] But broadly, in a way, you're a delightfully disconnected from it.

[00:12:56] Right.

[00:12:56] Because you can apply project management principles across multiple industries.

[00:13:00] And you do, right?

[00:13:01] So you could be in manufacturing.

[00:13:02] You could be in, you know, anything that has a group of people that are trying to come together

[00:13:06] to accommodate.

[00:13:07] You can be not for profit.

[00:13:08] Structured work.

[00:13:09] You can be not for profits.

[00:13:10] And so you're engaged heavily with the MSPs.

[00:13:13] And one of the things that we're seeing right now is this kind of trend to platform versus

[00:13:19] integration.

[00:13:20] And what I mean by that is, you know, some of the strategy is we're going to try and

[00:13:24] take this all in together and do a single platform to run your business on versus a strategy of

[00:13:30] we're going to pull best in breed components together into a single integrated communicative

[00:13:36] whole.

[00:13:37] Yep.

[00:13:37] So you think a lot about the impact on orgs by the people and the management.

[00:13:42] Kind of give me your sense of this cyclical trend and what your impression of the two

[00:13:47] strategies are.

[00:13:48] That's a great question.

[00:13:49] So I guess the first thing I'll say is it's cyclic, right?

[00:13:52] I've been in tech for over 30 some odd years.

[00:13:54] We see this pendulum swing towards best of breed platform.

[00:13:57] I think it's really, you know, the idea of platform is really important for spaces or verticals

[00:14:02] or segments that might not be very technical.

[00:14:04] Okay.

[00:14:04] Because they don't want to learn new software.

[00:14:06] They don't want to learn new tech.

[00:14:07] It can be overwhelming and it can stress and strain the system.

[00:14:09] Well, let's be really clear.

[00:14:10] This is a super technical environment.

[00:14:12] They can accommodate and manage best of breed in that process.

[00:14:15] And they kind of need to because they're looked to for that particular process.

[00:14:18] Right.

[00:14:18] I do think it's interesting in this regard to say if there was ever an industry that could

[00:14:24] tolerate best of breed, it's this one more than any because they have to constantly learn

[00:14:30] new tech and new process.

[00:14:31] Yeah.

[00:14:31] It's a whole stop.

[00:14:32] The other thing I'll say is, at least on the business side of the project, is that all

[00:14:36] the other major platforms, when we think of it in regards to our own business, we look

[00:14:39] at it and go, you're missing the point.

[00:14:41] Like, they're all looking at project management like people looked at it for 30 years, which

[00:14:45] is, let's build a really nice collaborative list where we all share it and get along with

[00:14:48] it, as opposed to work as a programming language.

[00:14:51] And if you don't think about work like a programming language and debug it and manage and compile

[00:14:55] and distribute it, you are never going to be able to unlock automations.

[00:14:58] Full stop.

[00:14:58] It's not possible.

[00:15:00] One of the things that the community doesn't know at large is just how bad the odds are.

[00:15:03] We talk about this sort of this probability dragon.

[00:15:05] And for a typical MSP that maybe, let me say small, like only has 20 projects in their

[00:15:10] portfolio.

[00:15:11] And they've got, say, 40 tasks in those projects that are on the critical path.

[00:15:15] If their engineers and their customers all deliver their work on time 90% of time, the

[00:15:20] probability that all those projects are done without a hitch is 6.03 times 10 to the negative

[00:15:25] 37.

[00:15:26] The probability that the sun won't rise tomorrow morning is like 3 times 10 to the negative

[00:15:32] 17.

[00:15:33] So it's a billion...

[00:15:33] And I'm going to trust you.

[00:15:34] The math on that.

[00:15:35] Yeah, it's like a billion, billion, almost a billion, billion, billion times more likely

[00:15:39] that the sun won't rise tomorrow morning.

[00:15:41] Then they're going to not have problems in their projects.

[00:15:42] And so this is my issue when we talk about going, oh, can we just improve our project

[00:15:47] management a little bit?

[00:15:48] No.

[00:15:48] You have to fundamentally look like this is a programming language.

[00:15:53] You have to address it autonomously, like an RMM.

[00:15:55] And a lot of our customers will say that.

[00:15:56] They're like, we need to have this RMM component for projects because the fail...

[00:16:00] And this is interesting.

[00:16:01] Like, you think about the failure point of an endpoint and how much that costs them when

[00:16:04] they manage it.

[00:16:05] Every tech...

[00:16:06] Most MSPs have as many project tasks as they have as endpoints.

[00:16:08] And every one of those tasks has a much higher failure rate than an endpoint.

[00:16:11] And they're not monitoring it.

[00:16:13] They're going, oh my God, there's an edge explode.

[00:16:15] So last question then, because I want to get something up and asking a ton of leaders that

[00:16:19] are watching the space that are interested.

[00:16:21] Give me a sense of, you know, we're rounding out 24.

[00:16:23] Yep.

[00:16:24] I'm not going to ask you to make five-year predictions because that's useless.

[00:16:27] Right.

[00:16:27] What I'm actually more interested in is you're thinking about the next 12 months.

[00:16:31] Yep.

[00:16:31] What's the thing you're keeping an eye on most of note, like waiting to do something,

[00:16:37] whether or not it's change, evolve, pop?

[00:16:38] Like what's the thing you're keeping an eye on as you think about 25?

[00:16:42] So I think obviously one of the things that's most interesting to us on some of the Gen AI

[00:16:46] stuff and utilizing some of those tools and technologies is interpretability components

[00:16:50] around that.

[00:16:51] Right?

[00:16:51] So as those things evolve, where you place those components to be agents of trust and

[00:16:56] credibility for answers.

[00:16:58] Right?

[00:16:58] So I'm sure you use the LLM tools that are out there today.

[00:17:02] Extensibly.

[00:17:02] Yeah.

[00:17:03] But there are components where you have to fact check it or make sure.

[00:17:06] And there are certain areas where you wouldn't use it because you go, well, I don't know.

[00:17:09] I can't trust this business decision because it's not ready for that.

[00:17:12] So I think one of the interesting things and sort of the people that I'm talking to that

[00:17:15] are in the know in the space on the academic and on the industry side is kind of to get

[00:17:19] a better understanding of how you can improve in a lot of those tools, their ability to

[00:17:23] consistently deliver answers with high fidelity and communicate that.

[00:17:27] So that's an interesting thing for us as we look because there is a merger between this

[00:17:32] idea of white box, interpretable and black box.

[00:17:35] They don't have to live separately.

[00:17:36] But when you integrate those technologies, you have to make sure when they interleave

[00:17:40] because the customer can't go, well, this is white box.

[00:17:42] It's black box.

[00:17:43] It's interpretable.

[00:17:44] It's not.

[00:17:44] They have to go, you've given me an answer and I have to have a high level of

[00:17:48] reasonability that this answer is correct and you have fidelity and consistency with those

[00:17:53] answers.

[00:17:54] Right?

[00:17:54] So that's, I think we're interested, you know, it's sort of a mouthful, but I think

[00:17:57] that may be in the next year, that may be five years away.

[00:18:00] Okay.

[00:18:01] I mean, because you see how rapidly this has changed.

[00:18:03] Oh, sure.

[00:18:03] And we can't predict it.

[00:18:05] So I'll be interested to check in with you down the road a little bit to see how the

[00:18:08] progress is.

[00:18:08] Mike, this has been fascinating.

[00:18:10] Thanks for joining me today.

[00:18:10] Thanks so much, Dave.

[00:18:11] This is great fun.

[00:18:12] Appreciate it.

[00:18:12] Thanks.

[00:18:14] Are you ready to get your brand in front of the tech leaders shaping the future of

[00:18:18] managed services?

[00:18:20] Here at the Business of Tech, we offer flexible sponsorship opportunities to meet your

[00:18:24] needs, whether it's live show sponsorship, podcast advertising, event promotion, or custom

[00:18:30] webinars.

[00:18:31] From affordable exposure options to exclusive sponsorships, our offerings are designed to

[00:18:36] fit businesses and vendors of all sizes looking to make an impact.

[00:18:41] Prices start at just $500 per month, making our packages a fraction of typical event sponsorship

[00:18:49] costs.

[00:18:49] Be a part of the conversation that matters to IT service providers worldwide.

[00:18:55] Join us at MSP Radio and amplify your message where it counts.

[00:19:00] Visit mspradio.com slash engage today to explore all the ways we can help you grow.

[00:19:08] The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines, posted

[00:19:15] at businessof.tech.

[00:19:16] If you like the content, please make sure to hit that like button and follow or subscribe.

[00:19:21] It's free and easy and the best way to support the show and help us grow.

[00:19:26] You can also check out our Patreon, where you can join the Business of Tech community

[00:19:31] at patreon.com slash mspradio or buy our Why Do We Care merch at businessof.tech.

[00:19:39] Finally, if you're interested in advertising on this show, visit mspradio.com slash engage.

[00:19:46] Once again, thanks for listening to me.

[00:19:49] I will talk to you again on our next episode of the Business of Tech.

[00:19:56] Part of the MSP Radio Network.