- Hello, hello. I am very excited to do this once again. And I'm with Andy Krier this go-round. And yes, we're gonna talk about GenAI. Andy, please. - Hi all. Yeah, my name's Andy Krier. I head AI and ML for the ITSM product group. And I'm excited to be here with you, Chad. - Let's do this. We're gonna get right into it. We got 20 minutes out. We'll go fast. Of course, there's a little bit of safe harbor. We're gonna touch on some roadmap items around Now Assist. But with that, let's get into it. Andy. - Hey Chad, one of the top things that we continue to hear from customers, is you "Wanna increase efficiency." These are pretty obvious statements. "And we wanna improve the overall experience," right? We've been at this for a while and we're really excited about what Now Assist brings to the table with these two really key objectives. - Yeah, and you're gonna see that throughout these demos that we're gonna do, the efficiency, the experience, and really just looking at the user, both agents and end users. So with that, let's talk about some challenges. - All right, so again, we're hearing this all over the place. We introduced a virtual agent years ago to solve some of these same problems. Self-service is tough. It's really tough to help people get the answers they need the first time they look to make sure search is working well, make sure knowledge articles are giving the right answers. This has been a consistent challenge for us. Collating information, you've all heard the swivel chair effect, going between different systems, even going between different screens within ServiceNow. This is a challenge. It's frustrating for end users. It's inefficient, obviously. And then last, but not least, once you do find that answer, once you do finally solve the problem, do you spend the time documenting it? Do you put the resolution notes in? Do you write an article? You rarely do, because you've gotta worry about getting onto that next incident, helping that next person, and doing your job. So these three challenges are really what we wanted to focus on, introducing analysis for ITSM. - And just really building up those use cases. And so, what Andy was talking about is really taking things that are maybe mundane or really mire up the front-end process that makes you happy, requesting something, or agents that get frustrated, getting mired down in things. And I had the opportunity to speak to our Global Innovation Expert, Brian Solis, and he really touched on something that really hit for me. And he's talked about the platform. And yes, we're talking about GenAI a lot here on the floor, but the platform should also be talked about. 'Cause we saw a lot of companies going digital first, right? We saw Netflix, we saw AWS, we saw Amazon, they were digital-first companies on the forefront of everything. There is going to be a company that is gonna be AI first. And what you're seeing here is ServiceNow is putting in the work to be that platform to be AI first. And we're gonna continue on with that, right? - Yeah, that's right. I mean, we're gonna help you be AI first, right? We're bringing this technology right into your platform, right into your workflows, so you can start enabling it right away and taking advantage of this revolutionary technology, right? - Absolutely. Absolutely. So, technical capabilities, what are we talking about? - Let's get into it, Chad. Why use Now LLMs? Why do this on the platform? There's a ton of LLMs out there. You could plug these things in, there's APIs, you could do this, right? Well, you know, first off, definitely start exploring LLMs. This is the new technology, get into it. But there's some key benefits for having this on the platform. Obviously one single data model, one context, one intent, we can bring all of that into the prompt. We can synthesize that information. If any of you have been studying up and following this area, if you understand RAG, or retrieval augmented generation, these are concepts that we can do directly within the platform, either using advanced queries, machine learning capabilities, AI search. We can really augment those prompts so that they're super valuable and that output is really high quality. And the last part there, is we spend a lot of time making sure that these outputs in this generation is accurate, that it contains all the information it should, and that it's not hallucinating. And what that means, it's not making stuff up, 'cause these LLMs are really good at making stuff up. So we spend a ton of time with our prompts, tuning them, engineering them, making sure that they're telling you accurate information, which is super important when it comes to high-critical incidents, changes, those kinds of things. - And what you're gonna see here is like, to Andy's point, we announced this last year, there was a lot of skepticism. Even I had the skepticism around this. And to Andy's point, ServiceNow, the engineers, the product teams, they really put in the work to make this the best experience possible. Instead of talking about it, how about we just get right into it, okay? - Yeah. Let's get into it. - All right, here we go. Now Assist for ITSM, okay. So it's gonna be short, it's gonna be sweet, but we're gonna get into the details. And Andy, my subject-matter expert, he's gonna get in and really demystify some of what's going on beyond the scenes, okay? - Yeah. Yeah. I mean, it's like you said, it's a short sweet demo, but that's kind of the beauty of it, right? We wanted to make sure this was simple, that we meet people where they are today, we're not introducing new workflows, new screens, new things to learn. We wanna make this super simple and intuitive for the end users. - So analysis for AI search. Now, speaking with you, whether it's at the demo pods, search is a big ask for many of you. Why? To promote self-service, right? So with that, I'm showing you a pretty bland portal. Yeah, really Chad? This is the portal? - That's the portal. - That's it? - Sorry. Yeah. But I'm sure the same portal's out there. Maybe if you guys have a better portal, please show us. But this is our portal. - Well, here's thing, though. You could spend as much time as you want on your portal. You should. You should make it the best experience possible. But where are users gonna go first? They're gonna search, they're gonna go look for something. And it's great to bring that one thing back to, you know, they're gonna put a search string in, they're gonna get an article back, they're gonna go through. And typically, they'd have to go read the whole article, find the pieces that are relevant to them, understand it. Hopefully it's written in a way that they can comprehend easily. But you know, a lot of times it's not. The beauty of generative AI with AI search, is not only do we bring that article back, we bring back the part of that article that answers the question, right? And that's really critical, 'cause I don't have to click anywhere else, I don't have to go anywhere. I can literally get that answer and those resolution steps right here without having to start a chat or open up an incident or anything. - And there is an opportunity to dive in more. As you can see, there's a hyperlink there to find out if it's truly what you need or if you wanna find out more. We don't limit that. We definitely just make it easier, right? - And that source is really important, 'cause that's gonna build trust with the end users. It's gonna, again, make sure that people know the LLMs aren't just making this stuff up as they do. This is actually coming from a real, genuine, vetted source. - Right. And you'll see a little bit more later on, how these vetting sources come to life. All right, let's talk about VA. Some in the audience, it's been a pain. I've talked to some in the audience. I love VA. - I love VA. Show of hands. Who here has implemented VA in your environment? - Okay. And who regrets it? (audience murmurs indistinctly) - Oh, we got one, two. All right, so let's talk about VA with GenAI. There are complexities, garbage in, garbage out, right? And what we've seen with GenAI, and I'm gonna actually let Andy speak to it. It's amazing. So I don't wanna steal your- - Well, I mean, let's talk about the before and the after. A lot of you have done this already. You have to define the intents, the utterances, the topics, the NLU models, you've gotta train them. There's a lot of work that not only goes into setting that up, but then maintaining it over time. What we see with generative AI is that it doesn't need any of that. It's this huge, large language model with billions of parameters. And what do you really then start to focus on are the primitives, your catalog items, making sure that they're clean, they're not fully customized with a bunch of DOM manipulation. You're focusing on your knowledge articles, making sure that they're written well, so that generative AI can parse that and bring that information out. So you really get back to basics, you focus on the data that's in the platform already, and then you allow generative AI to take advantage of that. - Absolutely. So something as hard as wifi in a plane. So wifi in a plane, it's pretty general, but it's specific enough where it could trip out normal machine learning, right? So we get a generated answer, and then we can say that "I'm all set," and just go from there, right? And that's just it. - Yeah, that's really cool, 'cause you're seeing AI searched, not just in the search in the portal, you're gonna see it in the virtual agent. When you've got your virtual agent integrated to Teams, you're gonna see it there, on mobile. It's one solution that's gonna solve on all those different channels. - One thing I wanna top off on too, is using the NLU, it's so limited, right? But that wifi search or something like that, you can get very specific. You can use acronyms, things like that. And GenAI will start parsing that out to give you the right answers to the right questions too. - Yeah, let's take a look at that when we get into catalog ordering. - Okay. - 'Cause that's a really interesting area where people will say all sorts of things, maybe not even know what they're asking for, but the LLMs do a great job of figuring that out. So here it is, catalog, another big ticket item for many of you, getting a good catalog experience, we've added it to Now Assist, to this experience. - Yeah, and if you've done topics before, you know you have to create topics to fulfill these catalog items and these orders. Now you don't have to do that anymore. If you've got your catalog items with minimal customization, there are a few parameters that we'll walk you through in the product, but for the most part, it can parse your entire catalog and make those conversational. - Absolutely, and speaking of conversations, you can go from this poor portal that we have here and you can have that portal experience, but your catalogs can be in a portal experience, as you can see here. You can request that, but then it becomes omnichannel, you can go across channels to get that same experience. And so instead of doing the form, I'll go jump back into the VA and type the rest of the questions there. That's right. So in this case, you can see I am in the product department and I need access to dynamics, right? Everybody needs access to applications. This is really common to get these kinds of requests, right? And these are simple catalog items. These are great for the VA, they're great to be conversational. In this case, we find the catalog item right away. And what's cool is that as we're getting started, I've already indicated that I'm in the product department. As you can see, that's one of the fields on the left-hand side here. But I don't have to provide that information. Again, it's gonna parse that and go right to the next question. - Where are you? All right, so now it's asking which region I'm in. So I'm gonna put in "Not applicable." So, "N/A," I'm being- I'm surprised you didn't say "San Diego." - Well, San Diego's the best, so. We're putting in "N/A," right? The reason for this is to show that GenAI can be savvy enough to start picking up what you're trying to answer, right? - Right. Yeah, you didn't need to define that in intent. You didn't need to define that as an option. It's just gonna figure it out. And so this becomes a true conversational catalog experience, thanks to the underlying platform in GenAI. - That's right. It'd be pretty easy to trip up NLU on that one, right? - So we'll go ahead and close this out. It's for a webinar. - Now, really quick though, let's be real here. Sometimes you have catalog items and record producers that have a lot of fields and a lot of options, like 20 different things you gotta fill in. And it's really complex, and it's gonna take some time. Those aren't always gonna be great to be conversational. You don't wanna make everything conversational. It's important to think about that user experience. How are we making it more efficient for them? - Right. So pick your battles. Go for the simple stuff first. Get that stuff enabled. And then maybe take a look at some of those really complex catalog items and see, do they need to be that complex in the first place? And if they do, the form might still be the best place to do it. So, you know, don't try and boil the ocean here. Go for the, I'm gonna use all my CEO talk, low hanging fruit here. And, you know, do what makes sense for your end users. - And on top of that, I think just to close this out, we see that it's captured a lot of this information, so it becomes auditable as you're going through these conversations. So to Andy's point, find out what's working and make it better, right? Don't get it over complex. - Chad, can you give access to Dynamics Now, please? Can we finish this up? - Okay, here we go. We're good. All right. - Awesome. - Let's move on. We have the requested items. We talked about the self-service, right? So really just making that front-end experience, those CSAT scores, a whole bunch better. - Yeah. - But what about the agents? - What about the agents? - Well, before we get into this, I'm gonna just share a little pet peeve I have. - And I think you've all been there, where you call in, you're talking to a virtual agent, whether it's on a chat or on a call, they're asking you questions like, "Hey, what kinda computer do you have?" You know, maybe they're asking information about your user account, trying to validate you, and you try and go through the troubleshooting, but it doesn't quite work out, so you need to get to a live agent. And what's the first thing they do? "What kinda computer do you have?" They ask the same questions. - Like, "Ugh, I just answered all these questions. You really don't have that information?" - "Let me transfer you." (Andy laughs) Ask the same questions. - "What kinda computer do you have?" This is extremely frustrating for everybody and we wanna solve that problem. Now, we track all the transcripts. You can always go back and read the transcript as an agent, but you don't have time. It's easier just to ask the question again and give that poor experience. But within the Service Operations Workspace, within our chat capabilities, what we can do is actually summarize that entire chat conversation that they've had with a virtual agent or with another agent before that, and we can bring out the key parts of that conversation that are important. So you don't have to ask the question again. You can pick up right where they left off. Now again, this isn't just efficient, it's obviously gonna save some time, this is a better quality, this is a better experience for those end users, reducing that frustration and making them more productive in the long run. So really just summarizing everything and not asking those same questions, mean time to resolution, right? You're seeing a lot of that. Just really getting through quickly. - Speaking of quickly. (Andy chuckles) - All right, so incident summarization. We're talking about the same types of things here. Getting up to speed quickly, not repeating those steps, not having to go back and do things over and over again. Incidents can get very, very complex, right? You got a lot of work notes, you got a lot of information that you're trying to parse through, you've got related records, you've got history there. It takes a long time to get up to speed on those things. - And one of the things that, hearing from many of you out there, Service Operations Workspace has been a catalyst to getting all that information in a single pane of glass, right? But think about some of your agents that maybe are new to ServiceNow, don't like the window experience, but they've got a bunch of tabs. Well, really just trying to focus in that information using GenAI. 'Cause as we see here, we have impacts, we have summaries, we have activities. All this various information can be summarized very quickly. Yeah, so instead of reading through the entire work notes, looking through all the related lists, trying to figure it all out, this summary is gonna get automatically generated with all of that information. And this is one thing really cool. We're not just putting it all in and saying, "Hey, LLM, summarize everything here for me." No, we get really specific with the prompts. So first, give me the issue, what is this all about in the first place? And then give me the key actions taken. This is what's so cool, is that it can actually parse through that entire history and pull out the things that were important that happened. Now, this is something, again, we've spent a lot of time getting these prompts right, we spent a lot of time getting the input data right, so that it can produce these kinds of outputs. And you know, I think that's really cool. Because again, now I don't have to go do that again, I can understand exactly where this is left off. Expand that, imagine a major incident, right? This is gonna have one person come in, then three people, then 10 people, and every one of them is gonna have to get up to speed on that. - Yeah, and with that, I mean, just think about that resolution around that, but the summarization. Think about early in career teams, right? Someone that's just using it just now, early in career. they're gonna get up to speed so quick because it's gonna help demystify some of the jargon that you see within that, right? - That's right. Yeah. - And then the other fun part is always resolutions. - Yeah, that's right. That's right. So I don't have a problem with resolution notes. I just type in "Resolve" and I go on to the next one. It's pretty easy, you know? - Yeah. That that works well. - It's really helpful for the next person that gets it too. - Yeah, so when you look at your poor knowledge bases and things like this, or maybe a sales playbook that you wanna do, or a resolution automation, you have people like Andy just saying, "Resolve." - I'm lazy. What, what? - What? Do you wanna teach me to be different, Chad? - Yes, yes. No, but in all seriousness, we know that a lot of teams are under the guns just to get to an end result, right? And so it's very easy to skip steps. One of the things that we really try to do is look at that resolution cycle. - That's right, that's right. And again, we talked about the key actions taken, what we can do in the incident summary. We can go one step further and look at the actions that actually resulted in a resolution and put that in the resolution notes, get this drafted. Now, like anything with an LLM that's producing content, you're gonna wanna double check it, but we're trying to get 90% of it done so you don't have to start from scratch and think of it yourself. This is really important. And the other cool thing about this incident here is it's integrated with Teams. So we've got a Teams chat transcript in here as well. I mean, how often are all of the resolution and all that troubleshooting happening in an external channel? And then that information is completely lost. Integrate Teams into ServiceNow, get those transcripts into the platform, and let LLM start to take advantage of all that data. - And the work behind that really came from you, right? It came from you, the customer, because many of you work in Teams. So again, showing that ServiceNow is taking this greatness and then bringing it to a channel that you use, a communication hub that you use, and still bringing in that goodness into the platform so it can be meaningful. Speaking of that, we take resolution notes. What can we do from there? - Yeah, I mean, resolution notes are awesome for the next agent that's gonna see a similar incident like this and try and solve it. 'Cause they can go in and see exactly how that was done. But that's not really what we're trying to drive for. We wanna prevent that incident from ever happening in the first place. So, we're gonna continue on this path. We've got the fully-documented incident, we've got the resolution notes that were generated. Now we're gonna create a knowledge article so that this can show up in search, this can show up and get summarized in a portal or in a virtual agent. And this incident never happens again, right? So again, we heard this yesterday in the keynote. It takes 30 minutes, typically, to draft and get a knowledge article into a good place, right? We can do this in seconds. We can take all of that information, again, with really clean, great prompts, and have a great article drafted. And again, I got a starting place now. I can go in and add to this, I can change it if I need to, but I'm not starting from scratch, which is really, I mean anybody writing, that's the hardest part. - And nor do the agents have to be editorial experts. That's not their job. They're here to take care of issues. They're here to take care of major issues. They're not here to be publishers. The other thing is they're not there to help clean your knowledge catalog. This is what Now Assist can really help with, right? - That's right. And this is really just the first step. As you can imagine, you've got redundancies, you've got knowledge gaps, you've got quality issues, probably throughout your knowledge base. If you don't, that's amazing. I'm so proud of you. That's incredible. Most of us do. And that's where we're really seeing this going in the future. And speaking of the future, just kinda wanna flash this slide up, for a number of reasons. One, for you to see what we're planning here. For two, we have a lot planned. This is just through the end of the year. And this is gonna continue, 'cause there's so much opportunity here. So coming out, literally this week is our May release. We've got a bunch of stuff that we covered today. In Q3 we've got huge things like Change Summarization, Prompt Designer, Prompt Studio. And then in Q4 we're looking at expanding that story around change with Change Assist Skills, being able to query a change right from the Now Assist panel. We've got capabilities on Knowledge Assist to take care of things like knowledge gaps, like we talked about. There's just so much opportunity in this space, and we're just gonna continue to invest and bring more innovation to your (indistinct). - And then really becoming that platform AI trust center that you can. So that is our time, unfortunately. I truly enjoyed the time. I loved seeing all of you. If you have any questions, or you wanna see this in action, we have the demo pods over there. You can see Now Assist with ITSM, and any of these use cases, from self-service onto agents. If you do have questions, we can take questions on the side here as well. But I do, I wanna thank you very much for spending your time with us. - Thanks all. (audience applauds)