Transcript
Welcome to our webinar, More Results, Less Busy Work, the AI agent that works for you. For today, it is a recorded webinar, so you will be able to look at this on demand. If you have any questions, then please put those into the Q&A section, which is on the right-hand side of your screen. And there are also follow-up documents, so there are documents there under the Documents tab that you can have as takeaways with more information about what we're presenting today. So again, welcome to this session. My name is David Pickering. I am a Product Marketing Director for Avanti, based out of Australia. Welcome to you all. I hope it's been a good day for you. Mine's just starting. And we'll have a great session today around our new capabilities, which is around our agentic AI and our self-service agents. But before we get too far into this, I want to just ask a question of you who are attending and just see what's going on. So when are you looking to implement agentic AI in your organisation? Is it now? Is it six months? Is it 12 months? You're planning it, but you don't know? Or maybe you actually never want to implement agentic AI. So I'll just let you pause for a little bit while you sort of work out what your thinking is and what your organisation's thinking is. And I think we've got one person who never wants to implement that, which is, you know, these organisations are all going to be different. And of course, as agentic AI matures, that may change as well. But we've obviously got most people looking in that six to 12 month timeframe, with some people looking at it now. So not too surprising with those results. It's a great scenario. And I think that's true of the industry as a whole at the moment, that really it's still a maturing process, that agentic AI. And I think by the end of the year, we'll see a lot more people starting to use that in production and in serious terms. And there's a lot of noise out there about that. But let's move on to the next slide and sort of start to talk through what we're talking about here. So thank you for answering that poll question for me. It's fairly typical of the industry at this point in time. But when we look at what's happening in the industry, so particularly in the IT service management, it's really starting to change the way we go about IT service management. And when we look at what's happening in the industry, the global agentic AI market, you know, they're projecting that to grow from a $7 billion business to a $42 billion business. But that's across everything enterprise applications, etc. And what certain analysts are looking at is 40% of those enterprise applications will have embedded agents by the end of 2026. ITSM or IT service management is recognising that probably 33% of organisations will have that capability. And then we'll look at productivity improvements, which are quite significant when you're using an agentic AI, will probably fall into around about an 80% cycle. So we start to see there's quite a lot of activity happening across agentic AI, which will help you in that. But there is actually a path to agentic AI and there's certain steps that we've taken. Not too surprisingly, when we look at agentic AI, we really started this journey a long time ago. So in IT service management, we started with traditional automation. And that was, you know, we had rule-based workflows that would, you know, go and perform automation for us or activities or tasks that we wanted them to perform. But they were very much contained to what could they do, what could they actually behave. They didn't learn, they didn't have any way of changing without it being recoded. And that was really the start of this whole evolution when you think about it. And the concept of workflows was, let's go and try and reduce resolution times. And then we started to introduce machine learning and cognitive AI. And that's really been around for, you know, five to 10 years, potentially, where machine learning was doing that automated ticket classification or automated assignments and starting to learn about your analytics and what was going on. And then enhance that a bit further with other things like predictive analytics. So that was the first phase of really how we went about doing it. And then, of course, chatGBT turned up and we started getting into that generative AI concept. So generative AI, you know, has been in the ITSM solutions for 12 to 18 months, largely being used around, say, incident summarization, knowledge creation, creating coding for you or creating other aspects of that. And really, it's point-in-time type information. You ask it a question, it will go and generate the answer for you. But again, it's got no real learning about what it's trying to do. And then we start to move into that conversational agent AI, which is where a lot of that agentic AI focus is at this particular point, where using natural language work or natural language processing, you ask a question, it understands the intent of that question, understands what you want, has memory and has the ability to provide an answer for you. And it could be across multiple different tasks in different areas. And ultimately, what we want to get to, and where organizations are talking about this, but really it's probably 12 to 18 months off before we feel comfortable with it, is that autonomous agentic AI, where it's monitoring your environment, for example, it's looking at data that's happening, it's starting to answer things in its own right, and really starting to run its own entire agentic AI capability. And all of that is reducing the resolution time. So the more we get to that right-hand side, the intent is let's reduce the resolution time. So let's try and fix things before anyone notices them. But it's also about improving our productivity, and we'll talk about that on the next slide, is how does that improve our productivity? What does it do for us? Let's talk about that a little bit further. So what we're seeing is, you know, with AI, there's become more of an urgency to try and get things so that productivity can improve. And when we think about that, the shift to an AI-first employee support really starts to change how things happen in your organization. So when AI is capable of resolving those first-level calls, you get 30 to 35 percent of repetitive low-complexity tickets basically being solved. And in some cases, you could get 80 percent, depending on the type of queries. So it can be quite significant, but you're looking at that 40 percent as a starting point and potentially much higher. So you're starting to get the employees asking questions, getting things resolved when they ask them. And if you didn't have that employee-facing AI, which is scalable, so more employees, the more it's still going to handle, then your resolution times will rise. Because this can't handle more tickets when there's more employees, for example. And they will struggle to keep up with that growth. And obviously, when you start to have poor resolution times, poor support, employee satisfaction drops. And then the agents start to face increasingly complex work. So this starts to sound a bit weird, right? But as Tier 0 and Tier 1 issues are auto-resolved, human agents then have more time to work on those complex multi-step cross-system issues. So they're working on something that's much more in-depth, much more, takes a bit longer to handle. So that then raises their average handling time. So they start to have to handle things, they take longer. So whilst they might have historically been able to resolve things in a day, it might take three or four days now because it's a more complex type of thing. They need to have a more deeper domain knowledge because they've got that ability to learn. And then the cognitive load and burnout because you're starting to work on complex heavy things might start to impact some of those human side of things. And then, of course, we will have that efficiency cost and SLA pressures. There's always going to be there in the industry. Ticket volumes continue to rise. And IT budgets, unfortunately, tend to stay fairly flat. So AI helps you improve that time to resolution. So that's helping your organisation get to that point, improve your productivity. You start to meet SLA compliance well and truly exceed that, but your cost per ticket starts to get reduced because you're answering more tickets with AI. So they're reducing that cost per ticket. And the adoption shifts ITSM from that reactive firefighting to that predictive automated operations. So we start to change how we're working. So where does AI help? So what can AI do for your employees, the IT, the business? So what if we could eliminate the employees' frustrations of dealing with, say, your service desk? So with the conversational self-service agent, we can – it feels as though you're just texting a friend type of concept. You're having a chat to them. You're having a conversation. You get instant help with the issues that are impacting you now. And there's no more sort of, okay, what service catalogue request do I need to look at? What portal do I need to jump on? So if I'm calling them, I have to go and hold. So we're trying to get it so it's instantaneous as it can, but it's very much a conversational type approach. And what if we could also change how IT work and they can escape that ticket treadmill, reduce those repetitive pieces of work that they work on now, get AI-powered guidance and spend time on meaningful projects instead of tasks that, you know, they can be resolved under intelligent automation. And then also from the business side of things, we want to drive digital transformation. So get those productivity gains, lower the cost per ticket, and have that continuous service delivery and free the IT team to focus on strategic innovation that will accelerate their business outcomes. So all of those are what we're aiming to do when we talk about, you know, an AI answer. So what I want to introduce you to is our Avanti Neuron's AI self-service agent. So this agent is there to take the end-user side of things. So how do we help the employees at this point? So it's a conversational AI agent that enables an employee to find answers to troubleshoot an issue that they might have, or maybe it's a service request. Handle those high volume of, you know, basic tickets with knowledge, and it could be knowledge within ITSM, it could be knowledge that's external, and we'll look at that in a minute. And the concept here is, you know, have that organisational capability that can handle incident creation, service requests, knowledge, and, you know, just getting typical ticket updates and statuses. The outcomes or the intent of this is to lower your ticket volume. So the service desk, our human agents, they get less tickets, but the front-end AI agent is probably handling a lot more. Get faster resolutions because AI is handling that, it's not going to a human agent. Get better data quality because it's a structured format on how it presents that information. Get higher employee satisfaction, and then obviously reduce your service desk workload because those repetitive tasks have been handled by someone else. So overall what we want to do is have that Avanti Neuron's AI self-service agent improve your service delivery, decrease those resolution times, and enhance your user experience. So they're what our goals are, that's what we want to get the benefits to and how we can help you. Before I go to the demonstration, there's a question from Dustin, so where does the AI agent get its knowledge set for resolving the first level tickets? So you have the ability to configure this, and you'll see this in the demonstration, but you have the ability to configure how your agent gets information. So at the moment, it's internal, so it's internal to ITSM, but you can also add external data sources. So you tell it what external data sources you might want it to go and look at as well. And then we'll be extending those knowledge sources to things like SharePoint and Confluence in the near future as well. So really trying to give you both internal and external knowledge. Let's go and meet your digital teammate, let's go and have a look at your Avanti Neuron's AI self-service agent. So let's have that. We're just going to run a video of a demonstration of that solution. We are going to demonstrate our Avanti Neuron's AI self-service conversational capabilities. And this is the first phase of our agentic AI framework and agentic AI personas we're building on that. So it's a self-service one. Typically a user at this point would either send an email to the service desk or jump onto the self-service portal and go through the process of navigating forms and knowledge articles, et cetera. What we want to do is try and simplify that and make it more conversational for the end user. So I'm just putting in my laptop freezes when I try to submit my expenses, which is probably a fairly typical type of end user type question. They don't know whether it's a laptop or whether it's a web browser or whatever they're using. So now the agentic engine framework engine will start to work out what type of intent is that and what else is it doing? Now in this particular case, we're showing our thinking in detail. In a production environment, you're probably not going to go and need that much information. And one of the things is because I've mentioned my laptop, or if I'd mentioned I'm having trouble printing, we'd look at what assets are associated to that. So in this scenario, it knows I have a couple of assets that are linked to me. So which one is that going against? Because what we want to be able to do is link the incident to our assets. So we start to understand that. So I'm going to pick the MacBook Air as my asset that's got an issue, and it's going to come back and start to give me, just ask a couple more questions because it doesn't really have enough information to just give me a knowledge article. So it's asking when the freeze happens during expense submission, do you see any specific app or browser and what's happening at that time? So I'm just saying it's in Google Chrome and it's when I access the expense hub within our workday. So it's taken that information. Now it's going to do some more analysis and it's asked me another question. So before it can give me that knowledge article, what else do I need to know? So it's asking when Chrome freezes, is it workday's expense hub? Does it happen every time? And what else have I tried to do? So I just said how it's happening and we haven't tried one of those options you gave me. Now what it's going to do is it's got enough information or it thinks it's got enough information to go and create a knowledge information for us. So give us back different ways we might be able to resolve this issue. Now I've got our knowledge response. So the knowledge response isn't just a knowledge article in its own right, it's not a single knowledge article. It's a combination of different information. Summarise the issue that we've got and then we're talking about, okay, do I need to disable the writer extension? So if I've got that particular extension loaded, I need to remove it because that's found it as a knowledge article. Then it's pulled up different knowledge information and we can see through that all the different options we have and at the bottom we have what resources we use to try and resolve this issue. Now if it fixes it, I can say yes and we say goodbye. But in this case, I'm going to say no because I want it to go and escalate this. Now we're going to give it two escalation paths. One is I could go and chat to a live agent if there was one available or I can go and create an incident. In this case, let's go and create that as an incident. And then what we see here is a draft of our incident details. So what's our summary and then our description of the actual issue. So it's giving me a nice summarisation of what's happening and what the information is. If I'm happy with that, I can just go, hey, it looks good. And we've created our incident. So success, incident has been created and we've got our incident number. So that's gone through and now it can be handled by the service desk team. And we're going to look at our incidents within ITSM. We can see I've got my incident I've just created. We've got our title, which was what we had before. And then we've got our summarisation of what's occurring in there. So the issue is isolated to me, MacBook Air located in Australia. So everything that's been done that we talked about and gathered the information, it's been classified automatically. So it's going as an application error for the desktop service team and we've linked the asset. So that's a key factor of that. We asked about that asset upfront and we've linked that asset as part of that process. Now we also do more than just incidents and knowledge articles. We also can do service requests through the conversational AI. So I'm going to, I need a loan laptop for a conference next week. And now the AI will start to look at what potential options I have for that particular question. Now it came back. It's looked at all the different service catalogue request options that we've got, and it's given me a list of four there. So we'll just cancel out because none of the match in this case, I'm going to say, yep, the laptop computer loaner looks good. And then we come back with the service request offering and then the different fields that we want it filled in. I'm just going to quickly enter in what I want. And then we just need to get, fill in the last piece of information and then we'll go off and create that service request. All of that looks good. We're happy with that or we can edit anything. We do want to edit. We come back with a standard form that we had before. So we don't have any having to do any extra work on top of that. And I'll submit that. And again, my service request has been created. Now we can also check the status of any incident or service request simply by asking a question of my agent. And then we have all our instance. Now we could have asked about a service request as well, but we can say what the information is related to my incidents, what the status is. So the final part I just wanted to show us, we want to make the agentic AI pieces easy to configure and easy to set up. So you can see within our AI configuration hub, we've got our generative AI configuration capabilities and we've got our agentic AI configuration capabilities. And I'm going to come into here and I can determine what I want to do. So a few things to note, we've got analytics there so we can start to track, you know, what's happening from our agents, how long did they take to respond, how many people interacted. We want to be able to configure these easily. So both internal knowledge sources and external knowledge sources. So we can go to an external knowledge source and have that data within there that would pull back information that's relevant to what we're talking about. And then how do we drive that intent? So this is how we tell the agentic AI, this is what you should look for and how you should work when you get that information. And the same with anything else that we have. So we want to make it nice and simple, you know, whether we want them to skip the knowledge articles, what the default template is, whether we want to bring back device information because we could say, look, we're not going to show devices for a particular thing. And again, our intent. So we've configured that to make it easy for everyone to be able to configure and set this accordingly. I know Alex asked a question there, which was, are questions addressed by the AI agent tracked within Avanti to the user asking for assistance? We track the metrics, so what type, how many questions were asked, what did, how quick were they responded, were they handled by the AI agent or were they escalated, but we don't track the actual questions that have been asked. It's a good question. I'll ask our R&D team whether they have some thinking on that one as to whether we should or shouldn't be tracking the questions as well. So what I want to talk about here is the agentic AI framework. So we've built an AI framework that we can extend, that we're going to build upon to enable us to run pretty well any of the Avanti solution across this agentic AI framework. We're starting with ITSM, but what we're doing is look at it from the point of view of there's a persona-based AI agent. So we looked at the self-service one at this point in time, but what we want to do is have those personas-based AI agents that handle different things, so self-service, ITSM self-service, but it could be self-service for something else, for HR, for facilities, all sorts of different areas. But then we're going to build out other personas, so like for service desk analyst agents. So that would then be as a service desk technician, you would start to have an AI agent that assists you with trying to resolve issues. You can have that administration agent where you want to configure or create something within the ITSM solution. IT and operations could start to look at, I need to do endpoint management, how do I handle endpoint management, asset management, security, et cetera. So those personas will get built out over time. And then what you saw at the bottom level, and I'll talk the middle level last, is we have those task-based AI agents, and those task-based AI agents will again be built upon to handle the particular tasks that they've been asked to do. So a knowledge search, incident creation, the service request, summarization of information, the question and answer, which is getting the status of tickets, et cetera. And then we'll build out more of those task-based agents, for example, like executing automation for patch management, for example, or other areas. So we build out both of those persona-based levels and the task-based levels. In the middle of that is the agentic AI framework that's going to handle all of that. So the goal definition, what are we trying to achieve with the questions that have been asked and the personas that have been asked? The large language models that we have there, the memory, so to remember those conversations, the reasoning. So if it's the question that's been asked, it needs to work out the intent of that question and then needs to decide, is that question a service request question or is it something that might be handled by an incident or whatever the case might be, action frameworks, et cetera? And then the two on the left-hand side are quite important. So integration, so we need the ability to integrate to other areas, for example. So external knowledge, for example, Confluence, SharePoint, that's integrations that we need to handle. So what are those integrations that the persona or the task might need to access? And then something that's really important as well, which we're building out, it's not in the first phase of this self-service thing, is that interoperability using a MCP server, which is basically a server that enables other agents to talk to that. And where you can think about that is if you're using Microsoft Copilot, you could sit there and go, I want to order a new laptop. And if you're using ITSM, that Copilot would go through that MCP server and perform the same things we just saw from the self-service agent, but you're using Microsoft Copilot or whatever other agents that you've got at the front end. So that interoperability enables that agent-to-agent communication and the ability to talk outside of just purely the Avanti agentic AI framework. So let's have a look at those task-based agents and walk through those a little bit. Incident creation, we saw how that worked. It's the automated creation of those. It's conversational. It gives you the ability to do device mining. So we saw how, when I asked the question, it came back with what's my server, what's my assets are for that, so laptops or printers or however, and the ability to go out to that live agent. So it's about creating a structured incident that's probably got more information than what you get from an incident that's been created from an end user who sent you an email, for example. The service request, again, using that conversational service request, I wanted to loan a laptop. So it looked at all the different service catalog options, whereas traditionally someone would have had to jump into the service catalog and try and work that out themselves. So it gave a good intent of saying, we think this is the most relevant to you, and if it's the only one that comes back that's relevant, then it will just present a request. It's integrated with your service catalog that you have now, and it's also all the workflows that you already have configured for that. So if that workflow for a computer loaner that I submitted before required to get approvals and workflows for fulfillment, none of that changes. It still generates a service request. It still has the same backend processes that are performed. The knowledge search, we saw how it looked at the internal knowledge source within ITSM and how you can also extend that to external data sources. So if we've had Workday or Google support or Microsoft or whoever we had configured for that. But we give you the ability to say, I want these data sources available, but I'm not going to open it up to the entire random world of data. It's going to be trusted data sources that I've defined for it to use against. And then the question and answer, at this point, it will give you the status and you can look up your particular instance. We are enhancing that as we go through the stages of development to get that to update records. So you can say, look, I've solved this issue, you can close it, for example. That's coming in the near future. And then the configuration, a key aspect of that is we want you to have control of what you want to turn on. How do you want that to work? So device mining, do you want them to mine your assets and come back with that as part of the query? Do you want live agent with the ability to talk to a live agent? And when we talk about the live agent, we didn't show that, is that it will use the agent chat that you have already there for a service desk analyst. So it's not an additional, you don't have to jump into this tool to chat with them. It's just using the standard ITSM agent to talk in the background. And then the analytics, we want to give you information that enables you to identify whether you're actually getting anything out of this. Are you getting an ROI out of the tool? If you're not getting an ROI out of the tool, is it a process issue? Is it a communication issue? Is it just a change, organizational change? Or maybe there's areas that need to be resolved. You don't have enough knowledge to have that knowledge search actually answer the questions properly. So the agentic AI framework is, as we've seen, the first phase of that, that's self-service agents. But it's also part of our bigger AI capabilities. So we've already got those generative AI and machine learning capabilities from instance summarization, knowledge generation, write assist, dashboard creation, et cetera. So how does that work now when you look at it from an end-to-end perspective of I want to create a ticket and how do I go across all those? So this is where it's AI driven value across every stage of that incident lifecycle. So when we're looking at the first stage, which is we want to deflect any tickets from being created at all, that's where the self-service agentic AI agent comes into play. It's the employee self-service agent and that will help your employees solve their own issues. So they look at it, they get the resolution, we never need to assign a ticket at all. So if we do need to assign a ticket, this is where our AI for ITSM comes into play. So we get incident classification, automatically classify the ticket, assign it to an appropriate user. We want to understand that issue. So summarize the incident, you know, we've got good data if it came from an agentic AI, but maybe there's people been looking at it. Summarize it, quick catch up, make sure we understand what's happening. If there's multiple issues coming in, so multiple people are starting to look at that, using incident correlation to look at, do we need to escalate this as a master incident or a major incident? Do we need to escalate this to a problem? Incident correlation. As I'm working through an incident, I want to resolve this. I don't want to spend time writing up nice emails and everything like that. Maybe I'll use the email assistant with AI to grab all the relevant information from the incident and generate that draft for me. And then we want to continuously improve and also to enhance the agentic AI knowledge articles, we want to generate knowledge articles. So using AI knowledge generation, generate knowledge articles that then can be used to back in through our deflects, through the agentic AI, and then analyzing everything using the capabilities of our AI assisted dashboard creation. So all of that brings together both the agentic AI and our AI for ITSM. Just the last couple of slides. So our vision for AI at Avanti is, whilst we've started with ITSM, it's empowering across all of our tool sets. So we're looking at more than just ITSM, we're looking at having that AI across the autonomous endpoint management, IT service management, we've seen already the start of that exposure and network security. Using the Avanti Neurons platform to use not only AI, but the orchestration capabilities and the workflow and the automation to enable that to improve your have intelligent risk management, accelerate your productivity, and bring together that security and operations so that we've got more data. And ultimately, it's building out a platform that's embedded with AI and designed for outcomes that are going to improve your enterprise capabilities, drive digital transformation, have that cross product orchestration, so it's not just ITSM, it's everything within the platform. And with MCP servers, it can be coming from Microsoft Copilot, for example. Having an open ecosystem, that's where our MCP servers come into play. And then our responsibility and predictability. So having the ability to measure what we're doing, get true ROI out of that outcome, and looking at that. So that is everything for you. I hope you've got some good outcomes. I know there's been a few questions, are there any more questions out there? I've had the two questions. Another one from Alex, to identify assigned CIs to a user, must the Avanti CI be used? You will need to use the Avanti CI at this point, so that you'll have the CI within the Avanti CMDB, and you'll have the assigned user against that user. So at this point, that's the way that will work. As we build that out and start to expand our data sources below that, I think you'll probably see that might change a little bit as well. Okay, there's no more questions. Thank you very much for your time today. One more question. Dustin, how customizable is the response from the agent? Is it a closed box, or can you add something to change its behavior and responses? So you can. So you can have, within the configuration, there were a couple of basically open text sources. So one was how you, the intent, or wanting to understand the intent of the question. So it's like, what words would it mean for me to answer that as a service request, for example? How do I understand it's a service request? In that scenario, you can extend, change the text of that to include other areas of how you want the knowledge to do that intent. So it's really telling it what to do. Things like, you could narrow it down and say, look, I only talk, you know, when I'm answering a question, and I only use Okta, for example, not Microsoft Azure AD, you can tell it, this is what we use, and this is how we do it. We can start to filter things a little bit. So you do have some flexibility there to drive how the AI wants to work. SharePoint, the knowledge isn't linked to SharePoint. We're building that out. So SharePoint's the next on our list to be able to go and grab knowledge out of SharePoint. And then after that, we'll be grabbing data out of Confluence, which is another major one. Dustin's saying, ask one question at a time. It effectively, depending on the answers, it will potentially go through and ask one question at a time. So you can also drive how many questions do you want it to ask. So I think at the moment we default to, I think, three questions. So it will sort of look at, try and get enough information out of three questions. But you could say, I'm going to allow it to ask ten questions, and it will keep asking until it's got enough information. You can also limit that to say, look, I only want two, because it's going to frustrate the employee if they have more. So it does actually ask different questions at different times, and depends on the responses you get. If the end user gives really good answers, it's probably going to be able to answer, you know, not have too many questions. If they're giving one-word answers, it's probably going to have to ask more questions. What about Snowflake? Is it a knowledge article source not on the plan at the moment, or is that Smokeflake? I haven't heard of that one, but I don't think that's on the cards just at the moment. But that may be something that longer term we might be able to do at Snowflake. We might be able to do something later at a time, but it's not on the roadmap at this point. Thank you very much for attending.
