How AI Will Reshape ITSM with Xurrent’s Jim Hirschauer

Hey guys, thanks for the throw. We're here with Jim Shaw, who's head of product marketing for Zern, and we're talking about, well, where ITSM is headed in the age of AI because, well, I think we all know big changes are coming, we're just not quite sure how. Jim, welcome to Shaw.

Hey Mike, thanks for having me. We've been on this curve, right? We've seen AI in the form of predictive machine learning algorithms so far in the land of ITSM, and, uh, some people love it.

Some people are dubious, and now we're on this curve towards generative ai, and that seems to be even bigger, but it's not quite clear to me where we are on this journey and, and how advanced are people and do we have the right set of expectations? So, you know, set the stage for us, if you would. Yeah, yeah, I think expectations are all over the map.

Um, from all the conversations I get to have at out at different shows and across all the different companies that we speak to, uh, there's a lot of hope, there's a lot of hype. Um, and at zurin, we're actually productizing things that to us really make a lot of sense. And in the ITSM world in particular, that all really revolves around productivity.

So we've been building AI capability into the platform for over a year at this point, and we really did look, uh, across the entire platform to determine what makes the most sense with this emerging technology, what makes the most sense to really help our end users be as productive as possible. So that's where we focused our efforts, and we have quite a, a, a large selection of various, uh, AI implementation within the platform that all sort of works together in various different ways to make sure that the end users are as productive as possible. Have you seen any specific use cases within ITSM that people are gravitating more towards in the age of AI that are, you know, maybe proven or places that you can be comfortable starting?

I mean, where are people succeeding? Yeah, yeah. So, uh, again, when you, when we think about it service management, there are few key areas where it really just makes a lot of sense.

So one of the, kind of the no-brainer areas for us was summarizing tickets. So if you've ever seen, uh, a ticket that gets logged, uh, often there are a lot of notes in that ticket. There can be many back and forth.

There'll be many different individuals involved in, uh, responding to a a single ticket. Uh, so just having the system summarize that and put that summary right at the top into a succinct few short sentences, uh, could save a tremendous amount of time when someone new has to come and look at that ticket. They don't have to read 75 different messages and spend all that time and then realize, oh wait, you know, there's actually nothing even for me to do here.

They can just quick, quickly look at the product summary at the summarization and, uh, and then go ahead and, and act on that if they need to. So that was one area where we're like, okay, this just easy, generative AI is a perfect use case for this, is really good at looking at that type of content and summarizing it into a few short lines. Uh, then we also looked at things like categorization.

So another area that companies struggle with, when people put a ticket into a request system, often they don't know how to route that ticket. So the tickets get dumped in and they get into this queue of routing, uh, that has to be done by someone. Some people on the back end have this massive queue that they have to work through, and they start flinging tickets around.

You know, it can take days, multiple different people involved in getting that ticket. And the fix for the ticket can take like 10 minutes, but you spent four days flinging a ticket around and then your customers really unhappy. So we knew that AI could, uh, look at historic tickets.

Um, within zunt, by the way, everything is a service. Everything is defined as a service within zunt, so it makes it a little bit easier for our AI to figure out, hey, you know, if someone's having problems with certain services, I know how to route this ticket even better. So our AI does a great job of classifying tickets and routing those tickets.

Um, and in most cases, it's gonna get to the right person the first time around. If it doesn't, it actually tells you why. It tells, it tells you why we made this decision.

The AI used some ticket, uh, from the past, and if that ticket was improperly classified, it gives you the opportunity to go back and reclassify that ticket so that you improve the system overall. So those are a couple areas. And then of course, like the big use case that, uh, most people are familiar with is really like a virtual agent where end users get to interface with the virtual agent, right on the, on the front end of the conversation, the virtual agent can look at past tickets, it can look at the knowledge base, it can make recommendations about how to solve the issue so that, uh, customer support agent doesn't have to get involved at all.

Um, and then if there's uh, no way for the end user to solve their problem directly with the help of the AI agent, then the AI agent can go ahead and submit a ticket on their behalf. So many more areas besides that, but those are, you know, a few of the key areas. I can see how we'll reduce the number of tickets or, and, and definitely resolve them faster.

I often wonder though, is tickets the right metaphor for taking care of these issues or is there something beyond tickets that we should be thinking about? Yeah, good question. So it's, it's the simplest term, uh, and that's why I used it.

Uh, inside of zurin, we, we have this term requests, and those requests can have workflows associated with them, and that's often what happens. We have, you know, at this, at the simplest level, you've got a request that is a very straightforward request, like, oh, hey, I need, you know, assistance with resetting my password, something like that. But there are often much more complicated requests that need to kick off a workflow like, Hey, I need, uh, to get more RAM added to this virtual server instance, something like that.

Um, and that can kick off an automated workflow where there are many different steps associated with the process to get that done. There are approval steps. There's, you know, the whole change control process that would be required.

Um, so you know, within zunt, we, most people are familiar with the term itil. Um, we, we subscribe to many different ITIL practices. We support, um, we're, we're certified on 19 of those practices, but we support, I think it's 27 last time I checked different ITIL practices.

And those practices are, are, are, are essentially a bunch of processes. So those are kind of baked into our system. And so we have these automated workflows that people can start out with and adapt them to their process that's unique to their company so that they can be as efficient as possible and productive as possible.

You know, I've pondered that issue myself in the, in the sense of idle, and I'm like, well, if the platform is now idle compliant, do I need an IL certificate or can I just say that it's in the platform? Yeah, you know, it's, it's funny, ITIL can be a four letter word these days. Uh, there's, there's some people that, um, are in the camp that, you know, they feel like ITIL is very rigid, um, and they don't want to be considered an ITIL shop, whatever that might imply.

Um, and then there's other companies that love itil. So, uh, we, we took that into account when we're building the platform. We put best practices in place as a framework to get people started very rapidly.

So one of the really interesting things about zurin is our typical go live time is 34 days. And if you look across ITSM and enterprise service management platforms, that is exceptionally fast. Most, I think Gartner had a stat that it was, uh, like six months is the average go live time for an ITSM implementation.

So with our average being 34 days, that's, you know, really, really fast, um, compared to that average. And the reason for that is because we give people a great starting point, and that's what we see. We look at ITIL as, hey, these are, these are really good processes that you can use as your starting point.

You don't have to be an ITIL shop, you don't even have to like itil, but I guarantee in most companies you have processes that you're following. And that within zunt, we have the bulk of those processes defined very, very close to what you're already following, and you just have to tune them and tweak them a little bit to get them to be exactly what your unique organization needs. We talked about AI agents.

Is there gonna be like kinda one Uber AI agent that I invoke and that's like the master butler from downtown Abbey, and then it goes talk to all these other agents to go do something? Or am I gonna engage with multiple agents who are kind of specialists in different areas and I personally am orchestrating them? How's that gonna play?

Yeah, yeah, good question. Uh, the still remains to be seen how the market will shape up with that, but um, I like to think of it as, yes, you really want to interface in one place, ideally, and that's, that's what we built at Zern, by the way. So we built a bunch of underlying AI capabilities that we felt were a foundational capabilities that the AI virtual agent, which is really just, it's just an interface to talk to our ai, which is pervasive throughout the system.

Um, without those other underlying functionalities, the AI that you're talking to would be fairly useless. And so it really is this kind of, this snowball effect of all of these different capabilities. I mentioned a few of them.

Another one would be like our, um, AI automation builder. I use that all the time within our system. I don't know how to build automations manually in zunt.

I have no idea I joined the company six months ago. But we have the user interface for it where you can actually just tell, uh, the virtual agent, Hey, I want this automation built. And you describe it in your plain language, and then it goes ahead and builds the, the code, the pseudo code for you, and you save that and it executes.

And it's, it's worked amazingly well for me. It takes me less than five minutes to build any of the automations I ever need. So I like to think of it like, yes, that you, you talk to the AI and the AI will route your request to the appropriate functionality under the covers without you having to go into a million different places and use all of the different features.

One of the things that I find when I talk to people that they're kinda rep having a little difficulty wrapping their heads around is the degree to which they can trust the AI agent. And I ask this question because they've all kind of figured out that a gen AI is probabilistic. So it's, it's guessing what's next.

And, but a lot of the workflows are deterministic, right? They're supposed to be done the same way every time a hundred percent of the time. And if the AI agent does it right, eight outta 10 times, they're like, that's not good enough.

So how does the AI agent kind of get trained to the point where it is, you know, at least close enough to being deterministic that people will trust it? Yeah, that's a really great point. So there's, there's two different things at play there.

There's the AI itself where you're interacting and interfacing with that, having a conversation with the ai. There's also the AI invoking a workflow, right? So once in er, once our AI invokes a workflow, it will help populate the information into the workflow that's required within a given workflow.

Um, it usually asks you for that information so that you can tell it how to fill in those fields. It will also try and fill in some of the fields on its own, and you can, you have an opportunity to correct it if need be. Um, so once that workflow is started, that is deterministic, like you said, that's, that workflow is gonna continue, the AI's not gonna manipulate that workflow.

It's not gonna change it on you because it is a process that needs to complete you. I don't want the AI to, um, you know, remove an approval step that's required. Uh, so that's not going to happen.

We don't allow our AI to do that. But on the front end of that conversation, it's really important that the AI doesn't hallucinate. We know that hallucinations are still a big challenge within the world of generative ai.

We need the AI to be secure, so it's not gonna leak our private conversations out to third parties. And we've taken an approach where, um, you know, in, in order to minimize any risk, uh, of, of any of those, um, we use AWS bedrock as our backend of our ai. And what that means is, is those are specially trained, uh, instances of Claude.

There are instances that our, our, our own for our own purposes. So they're used only for zurin, they're trained as support agents. So this really helps to minimize hallucination.

So if you ask the ai, you know, how do I make a peanut butter and jelly sandwich? It's gonna tell you, I have no idea. I'm here as a support agent for you and I, if you ask me, uh, support questions, I'll happily answer your support questions.

Mm-hmm. So that's one way to prevent the hallucination thing from happening on the front end. And then from a security perspective, none of that data that's being transmitted, there's a little bit of data that gets transmitted, transmitted to, um, our ai, but none of that data is used to train the model because the model comes pre-trained.

So we have a lot of guardrails in place from a security perspective and, and from an anti hallucination perspective that we see very little in the way of hallucinations. They happen sometimes still, I don't think there's any AI system I've ever used that doesn't occasionally hallucinate. Um, but we've, we've mitigated that for the most part with the approach we've taken.

Do you think in time we will, you know, you've heard the conversation about pets versus cattle, you know, do the AI agents become pets or do we treat them like cattle and they just come and go as they need it? So first of all, I like to be very nice to my AI that I interact with just to be on the safe side because, you know, none of us really know what the future holds in this aspect. Um, I like to think of the AI as much more than a pet.

I, I really, I use AI on a daily basis. So AI can do so much for us. Um, I see tremendous productivity gains on a daily basis.

So I'm definitely a big believer, uh, in the future of ai. I also like to be positive. I like to think on the positive side of this conversation and think AI is gonna benefit us as a society more than it would hurt us as a society.

I hope that that stays true. So, you know, just to cover all my bases, I, um, I'm very nice to my ai. I ask things with a please and I say thank you to my AI and try and be very human as if I were an interacting with a, with an actual human being.

On a slightly more serious note, um, do you think that in the age of AI we might be able to knock down a lot of these IT silos that have been built up over the years? Because so much of what we wind up doing is toil, and so much of that toil is trying to integrate the various, uh, silos. So might we get to a point where we can just flatten all this a little bit?

Yeah, I, I hope so. This is a really difficult conversation, actually. So I, I spent a lot of time in IT operations.

I worked in IT ops for 15 years, um, as a systems administrator, as an architect. Uh, I've done a lot of different things in it and I still have lots of conversations with IT practitioners, and I ask them regularly about how they feel about allowing AI to make changes within their environment, because ultimately that's what that, you know, the question you asked that comes down to that, how comfortable are we allowing AI agents to roam within our networks and make changes that, you know, in order to flatten that, that structure, they're gonna have to do it on their own. We're gonna have to put that trust in an AI agent.

So I don't see that happening now, right? There's still not nearly enough trust, and I think that's probably the right approach for now. We're in, in the infancy of this technology.

Let's face it, it's developing rather quickly, and that's great, but most people that I know are still not ready to let AI agents and, and we're talking, you know, agentic AI at this point, right? Which is like the kind of like the holy grail where the AI takes action on our behalf and can kind of do everything for us. Um, I, I don't think most companies are quite ready for that.

There will be little pockets of that in order to start building that trust and to build the systems, uh, to learn the lessons in areas that where we can minimize risk. Um, but it's gonna take us quite a while, uh, as an industry overall and as human beings overall, to really trust AI to that point where we're, we're allowing it to, uh, break down those barriers and, and do all of those, you know, make all those changes and do all those tasks on our behalf. All right, folks, you heard it here.

It people aren't going away anytime soon, that's for sure. But you might wanna make a list of all the things you don't enjoy doing and a list of the things you do like doing and give the ones you don't like doing to the ai. Hey Jim, thanks Pete on the shot.

I appreciate it, Mike. All right. And back to you guys in the studio.