ServiceNow, Inc. (NOW) Earnings Call Transcript & Summary

Hello, everyone. My name is Allan Sabol. I'm a Senior Director of Product Management here at ServiceNow. I lead our generative AI for HR and Employee Workflows. I'm joined by Kim Thomson, our Senior Marketing Manager for generative AI. We do have some forward-looking statements we'll be talking about, so I want to show our safe harbor slide. So today, we're going to discuss how AI is about to unleash the next wave of productivity across the industry that you're in and all other industries. This is the newest manifestation of AI is what we call generative AI. It's something you've heard a lot about in the news, and it's something that every business is thinking about right now. According to McKinsey, generative AI's impact on productivity, will actually add trillions of dollars of value to the global economy in the next few years. And this impact is going to be felt across every industry, department, every role and job within your organization. So getting the right AI strategy means you can actually unleash substantial productivity improvements. It will also help you transform the way that employees and customers experience your organization and your brand. And ultimately accelerate your agility as a business that help you drive business outcomes and stay ahead of your competitors. But it can be hard to really separate the hype from the reality. There's a lot of hype out there when it comes to GenAI, so we need to focus on what GenAI is appropriate for the enterprise. So who do you trust? Who do you listen to? In today's presentation, we're going to offer you our perspective on why marrying AI with a workflow automation platform like ServiceNow is a powerful combination for achieving your business transformation goals. So why is GenAI such a game changer whereas previously in previous iterations of AI, there was the ability to identify patterns. GenAI takes it to the next level by adding the ability to create net new patterns. And it does this in 3 steps: Step 1 is about understanding human language intent and being able to have the context to really understand what an employee or a customer is trying to do. The second thing it does extraordinarily well is synthesizing information from disparate systems, different sources, bringing that together into a comprehensive answer to whatever that employee or customers' question was. And finally, what it does exceptionally well is it brings together a natural language response and is able to bring back that information that is synthesized and give it to the employee in a way that makes sense, that's natural, that's very much human-like in nature. Now all of this is powered by large language models. I'm sure you've all heard that term, but not all large language models are the same. There are very powerful general-purpose LLMs that worked very well across a broad set of business use cases. And at ServiceNow, we allow our customers to connect a variety of these general purpose large language models for custom use cases that they might want to build on the platform. However, these general purpose LLMs are not purpose-built for HR and they're not purpose-built for employee or customer experience. That's why here at ServiceNow we built our own domain-specific large language model. The Now LLM understands the unique needs and tasks of the various personas who interact with the platform. So it understands the difference between, say, an agent and an employee and maybe a customer or an admin. Each of these personas needs to be handled differently, and they need different results and responses from the large language model, and that's what the Now LLM does. The Now LLM also understands the data model within ServiceNow and the workflows that we have in our platform. And that is extremely important to deliver the types of experiences that we want for our employees and other customers. And the result of all this is increased productivity and cost savings for your organization. And importantly, we've also built these guardrails around our domain-specific large language models to protect employee data, to enhance employee experience and to mitigate organizational risk. With Now Assist you get the best of both worlds. You get access to general purpose large language models and also you have access to ServiceNow's domain-specific large language models for HR and for employees. GenAI is already changing the game. This is not something that's 12 or 24 months away. This is something that's happening right now all around us. And what we're seeing across the world today is that there is extreme interest from a top level executives within organizations, Boards of Directors, they are really encouraging their organizations to adopt generative AI as soon as possible to drive down cost and improve efficiency. But we know that HR still has some concerns. This is a very new technology, and HR is always the champions and stewards of employee data privacy. And when we think about data security, bias, hallucinations, risk, these are things that are top of mind for HR practitioners and HR leaders. So HR leaders are working with their IT partners to find the right vendor portfolio that's going to help them mitigate risk while delivering GenAI solutions as quickly as possible. HR executives are also really trying to also figure out what exactly is the vision that they have for GenAI-powered employee experiences. And what we've seen across the world is a very consistent trend and what that vision actually is. What we've heard from our customers is that they want a unified conversational experience that allows employees to ask for help in their own natural language and receive a direct personalized answer to their question. They also want to be able to allow the employees to take actions from that exact same interface, that same conversational interface. But for HR, there's some unique things that must be true. It has to work for all personas across all interaction channels. And this means that we can't just support traditional chatbots. That's all from a customer's think of first is, hey, I'm going to build a bot. But we know from our user analytics and from research that many employees, in fact, most employees do not prefer to impact with chatbots, of most employees will actually start with the different channels such as search, if they're an office worker or if they're a deskless worker, they probably don't even have access to a chatbot. In fact, they're probably more likely to have access to only one channel, and that is their mobile device. So it's extremely important to meet employees where they're at and the channel that they're at and have that same conversational capability available wherever the employee prefers to operate. The second thing we heard from customers is that any kind of GenAI that's going to be effective at solving employee problems and answering employee questions needs to pull in data from other HR systems. That's why at ServiceNow, we are investing so much in bringing data in from other HR systems, whether it's HCMs or ATSs or point solutions, we're focused on trying to make sure there's enough data to answer as many employee questions as possible to really drive deflection and increase self-service. The third thing we heard from customers was it has to be proactive. It can't just only be waiting for the employee to have a problem. There are moments in an employee's life cycle where they need help, and we know that they need help, and we can now bring that to them through a conversational manner. Examples include something like benefits enrollment. An employee has not yet enrolled and the benefits enrollment period is closing, we might reach out and say, "Hey, do you have any questions? I noticed you have not enrolled in the benefits program." Similarly, during onboarding is a great time to reach out proactively to make sure the employees having a great experience and see if there's anything they need. Maybe they want to order an ergonomic desk. This is a perfect time to be able to let them do that through the conversational interface. The fourth thing we heard from customers was it has to support employees across all departments and all business services because, again, this is about employee productivity, and giving them one place to go. And that's what ServiceNow has been doing well for many years as we really cut across all the different business services in an organization, all the functional areas whether it's IT, legal, procurement, workplace that is what ServiceNow does. And what we've done now is just pull that into the GenAI conversational capabilities, so employees have one place to get all their answers. And finally, our customers told us that the LLMs need to know when not to engage with an employee. This is extraordinarily important from an HR perspective specifically. When you think about a case, a topic, something like sexual harassment or other employee relations-related issue. When you think about performance managing and employees is potentially underperforming. These are not times when you want a large language model hallucinating, and there will always be some level of hallucination in a large language model because they're probabilistic systems, not deterministic systems. And that means there's always that little risk. But that little risk is too big when we talk about something like employee relations. So what we've done at ServiceNow and what we encourage all of our HR customers to look for is make sure your technology or generative AI is smart enough to know when to disengage the LLM and hand off to a live agent for something like an employee relations case. And finally, I do want to point out that one thing customers are also extremely focused on is it has to work for HR. It has to allow HR to be autonomous. And the reason for this is because IT is going to be overwhelmed right now trying to figure out all the amazing things can be done with GenAI across the organization, driving operational efficiencies, running the daily business. With ServiceNow, we're trying to focus on helping our customers be able to get up and running faster and HR should not have to be dependent on IT to wait to deliver a solution a year from now or 2 years from now. These capabilities exist today, and our road map has so many more capabilities coming this year. So we certainly encourage you to think about how quick can you get value, how quick can you drive these efficiencies as you're making your buying decisions? So GenAI really can transform employee experience. In fact, it is happening right now. Our customers, as they've been deploying this have seen incredible results by not just in things like agent productivity and admin capabilities, but also really driving that employee experience and that case deflection and operational efficiency. So with that said, it's a very exciting time to be an HR leader or an IT leader.

It is an exciting time to be an HR leader, but when we start thinking about GenAI in next slide, with anything new comes a level of uncertainty, whether it's small level of uncertainty or big ones. We see all kinds of questions, right? Especially if you are not somebody who's been thinking about and following AI for years, this is new to you. You might have questions ranging from what is in LLM, what are these large language models to more challenging question like, how do I make sure that my information and my company's information, my people's information is safe. What about compliance? What about -- what are the data models? How -- What does that look like? What about risk? Risk is a huge factor. How about what can AI even do for HR? Where does that come in? Isn't this an IT problem and not an HR problem. The questions are huge. And so there's this huge level of uncertainty, and the big thing where I find so many of us sitting right now is where do I even begin. When you have this many questions on where to get started? How to get started? What things should you even think about? What questions should you ask? It's how do I even get started? So what we wanted -- aim to do today is try and give you a little bit of guidance on that. So when we're thinking about a responsible AI strategy and what it will do, it should do some of these things here on the left side. So it should ensure data privacy. It's going to create responsible AI policies, you are going to create those policies, right? Or it's going to produce unbiased GenAI results. It's going to help you avoid these data hallucinations, again, Allan was talking about hallucinations and times when you really, really don't want it hallucinating. It's really important that we can try and figure out how to mitigate that. And then in the end, what you're going to get out of it is it's hopefully can make your people more productive and provide personalized support that otherwise you can't necessarily sustain, it's not scalable without something cool like GenAI coming in and doing that for you. It's going to reduce manual workloads, things, tasks that are in place you don't necessarily want to be doing because it's manual and redundant and it could be done by AI. And it's going to support employee needs. At the end of the day, it's going to reduce legal risk, not increase it, and it's going to also protect your employees. And so what does that kind of strategy look like? And we're going to go through all 4 of these pillars, and we're going to talk about each one individually. So no need to start screenshotting now. I promise at the end, we'll have a big checklist. You'll be able to take a nice little screen grab or take a photo of so that you can actually take action on these. So no need to do any of that just stay tuned in here. So employee data and organizational IP should be secure. That's super important. And then we want to make sure that there's transparency and that's going to come from both vendors and to end users. And then for security and infrastructure, we want to make sure that all those requirements are being met. And then lastly, we're going to make sure that these are custom fit to your organizations. So let's like jump in and see a little bit of these individually. First, employee data and organizational IP is secure. Why is this important? Why should we care? Okay. Before we get into this little news article, let's take a second and think together. When you are driving to work, have you ever gotten there or gotten through a piece of your commute and then like, oh, I don't remember getting from point A to point B. I think pretty much everybody that's listening has had this experience where you know you're out there, you out there safe, you kind of don't recall what happened in the middle. So I've just kind of proved to you -- we're proving to you that all of us do things without thinking about them all the time. So when you look at stat like this, 6% of employees take sensitive data into GenAI tools like ChatGPT. We automatically sometimes assume like, well shouldn't we just know better. Why would you do that? But we all do things that we're not thinking about, even if we're trained to not do those things. So sometimes against all of their best efforts and all the things we do to make things like that not happen, human error happens. And so let's talk a little bit about things we should be asking or things we can ask our vendors to help mitigate some of the risk that comes with people doing things like pasting sensitive data and through GenAI tools. So a few things we can do, we can ask our vendors, how is data collected, stored and used right? So what data is being pulled in? And then how is it being stored and used, you want to make sure that's nice and responsible. How is PII data cleansed and sanitized and people are putting it in the system, you want to make sure that, that's cleaned up and taken care of before it goes anywhere else, right? So that when accidents happen, there should be a way to capture that and kind of help reduce that risk again. And then are your vendors and partners screening their vendors and partners? So all of this is things that are your vendors when you're vetting them, you should be asking them these things. Are they taking these steps to help mitigate risk when it comes to employee data and keeping that information and the IP secure? The next one is transparency provided from the vendors and for end users. And we start thinking about things like bias. Now how is bias related to transparency? Sometime or other, we've all written some kind of research paper, right? And you want to have the best sources. So if you -- we all went to the library at some point and looked at books and tried to find our sources, right? And then the Internet came along and it was so much easier to find those sources, right? Because you have like infinite information, but then the owners became you have to make sure that the sources that you're saving, they're reliable sources. And you better hope that the thesis you based your whole paper on or your students or whomever is writing it, you hope that, that thesis wasn't based on information found on Wikipedia, right? So why? Because you go to Wikipedia, it's got great information sometimes. But other times, somebody can go and we know how this works, they can rewrite Wikipedia with all kinds of bias, all kinds of information that's not totally accurate, but it sits there on the Internet and it pops up in the source when you're starting to do some research. Our large language models are kind of like that. It's not being fed the right data, it's not a reliable source anymore. So then when we ask it to compile it and write paper for us and create a thesis, and it's thinking about it, and it could spit out all kinds of biased things that we don't want it to do. When we talk about that in an HR use case, that's a big, big problem. So in this article, we're talking about UNESCO saying that there is pervasive gender bias among GenAI tools. So let's figure out how we can make sure transparency helps us bridge that gap with bias. So again, when we're talking to our vendors, what are we going to ask them, what are their underlying models? What models do they have? Who built the models being used? And what are their terms of use? So what's going into that? What are those terms of use? How are they kind of making sure all that safe? What data will you use to build those models? And what was done to reduce bias? It brings us back to the whole Wikipedia thing, right? What did they use as a source? Is that source reliable? How is it going to continue to be fed in, and how does it work towards reducing bias? And then is documentation available for GenAI skills? Security and infrastructure requirements are being met. I promise I won't give you another story on this, and I think we all know why it's important. Security is always going to be something that's right at the top of our minds. As HR people, we think about security and making sure that our people's data is safe, the things that are private, stay private. But then you've got the biggest industry analysts in the world saying, breaches are going to happen. GenAI is going to lead to breaches, and it's going to lead to fines in 2024. So we know this is an issue and what can we do about it? So let's look at a little few more questions for -- where to start with us. First, we can ask where our data models hosted? Where are they hosted, where is this information being kept? Where does the information flow? Don't forget, information is not a one-way street. It can go north than one-way, it goes in, it goes out. It can go in multiple different directions. So where does the information flow to and from? Which -- who are their sub vendors? Who are the other sources? What is the information connected to beyond the vendor that you're picking? And then are regulatory conditions being met, and are there sub vendors in compliance? So again, if the information is going in and it's pulling out, if it's going to sub vendors, are those sub vendors also in compliance? Are they making sure that the same level of standards and security is being met because they're also now the guardians of your information. It's not just the one vendor that you've picked if they're working with other vendors and sub vendors. And then lastly, custom fit to your organization's needs. There is no company quite like yours, right? It's not one size fits all. We can't just cram everything in a box. With something like GenAI, part of the whole point is that it's customized, that it fits your needs. It fits your end users' needs. But no organization is quite like yours. If I took a sampling of a bunch of companies in your same industry, I would be willing to bet you could at least think of one to two use cases for your organization that aren't going to work for somebody else that does the same thing or is in the same industry as yours. So you need to make sure that what you do implement is going to be right for your organization. So the questions on this one are a little bit more for you, what actions you can take internally to get started. This might even be the best starting point, as that starting at the end here and thinking about your organization's specific needs. So do you have a responsible AI team? Is that in place? If it's not, it's time to go get one. So figure out who you can put on that team. And then who establish the safe AI policy. And it's not just you. It's not just going to be HR, it's not just going to be IT, it's going to be all the different parts. So that leads us to have you consulted other parts of the business? Have you talked to legal? Have you talked to risk? How about engineering, ops, product? There are a number of other people within the organization who should be consulted to see how to build the best policies that work for your organization, work for your people, keep everything safe. Keep data in the right places and make sure that at the end of the day, you're getting is back to that original slide, right? Where you're reducing the legal risk and protecting employees, but you're getting all kinds of productivity and personalized support. And all of the other amazing benefits that come with GenAI because let's not forget, while some of this is overwhelming, and it's a bit scary, at the end of the day, GenAI is so worth it because it can really lift that burden off of your people and kind of supercharge what you're getting out of the organization at the end of the day. And so here's that tech list that I promised. This is everything we just looked at in a nice little snapshot. So these are all the 4 pillars we visited today. And some questions you can ask as you go ahead and vet those vendors. With that, I'll hand it back to you, Allan.

Okay. Thank you, Kim. So I'm going to finish up by just saying that at ServiceNow, we have our own responsible AI team. We have an internal team that looks at how we're deploying and using our products and other products for our employees, making sure that we are protecting our employees. But our product itself, the Now Platform has a responsible AI team and their job, what they spend all day doing is trying to figure out for every feature we build, are we meeting these needs like what you saw Kim take you through today. It's very important that everything we build is compliant, that's safe, that it protects employees and it puts the right guardrails in place to make sure that we reduce legal risk for our customers. So we do stay on local state and federal laws and also some of the international things that we're seeing that are happening in Europe and the Americas. So you can rest assured with ServiceNow that we are definitely doing our due diligence when it comes to responsible AI. And I suspect you probably have a strong AI team that you can draw upon in your organization to make sure you don't feel alone in this because you're not in most cases. And it will help you feel about moving forward and really getting the benefit out of generative AI. So last thing I'll mention is just when you combine the power of the ServiceNow platform with the transformative nature of generative AI, it really does drive business productivity, employee productivity, employee experience and cost savings. So very much encourage you to get going on your GenAI journey, and feel confident you can do this with the help of your organization. Thank you.

All right. I think we've got some time to take a few questions. So I'm going to pull some questions from the chat. Allan, I think this first question maybe you can address for us. So one of the questions on the checklist was actually -- was to ask vendors about documentation available for AI skills. Can you elaborate on that? And what documentation should they have?

Yes, sure. So first and foremost, when I hear you said skills, a lot of people don't know what that means. What is a GenAI skill? Let's start with that. So a GenAI skill just means it's a unique use case for which GenAI is being used. And an example of the skill here at ServiceNow, for example, one of our first features that we built in Now Assist was something that was called case summarization. It was just the idea that an HR agent could look at a long, long case with all sorts of work notes. You can hit the summarize button and get a summary. That's a skill, case summarization. So that's an example. So for a new skill like this, whatever skill it is that your vendors' providing. The vendor should be able to show you how the LLM was actually trained for that skill, they should be able tell you where the data came from to train that skill. I mean, was that data obtained in a way that you're comfortable with? Like where did it come from? And is it quality data that you can trust, where you feel like it's not going to introduce any type of bias based on your organizational situation. So the other thing you should be looking for is documentation that will help you understand any potential risks from that skill. So what is it about the skill? Is this skill, the type of skill that could introduce hallucinations? Could it introduce bias? How are those hallucinations being mitigated? What's being done in the product to try to mitigate that? I'll give you another example using that same skill we just talked about. With case summarization, it was really interesting. Our very first feature that we put in Now Assist, when we built the feature and we used our large language model and we tested against general purpose language models, and we did all of this testing. We had the same thing back from all the LLM. So what happened was we took a long complex case and hit summarize, we got a beautiful summary. It made sense, completely clear, super helpful for the agent who's trying to get ramped up on a new case, save them 30%, 40% of the time to get ramped up. What we didn't expect was if you take that same case -- or not that same case, but a different case comes in and if it only has a sentence or 2 as it's a new case, you hit that same summarize button, the LLM felt compelled to just make something up, and it just created a completely fictional case summary. And we were just amazed by that, right? But that is what happens with these probabilistic systems that are GenAI, is they do hallucinate if guardrails aren't put around them. But this is where the importance comes in, is what is the vendor doing to put guardrails around the large language model to protect against those types of things. So in the documentation, you should see things like, oh, well, we took our case summarization feature, we said anything less than X number of characters, we will not even offer summarization because it's just going to hallucinate. Anything more than Y number of characters, we will definitely show the summary button because you're going to get a really good result. So those are the types of things we can mitigate and the type of things you're going to be looking for when we say the vendors should be able to show you documentation. That's the type of thing you should be looking for.

Awesome. Can you talk a little bit more about probabilistic systems? I think this is a new concept for a lot of people listening in. So I'd love to hear you talk a little bit more about that.

Yes. So what happens with generative AIs, and sometimes when people see us the first time, they're like something is wrong, there's a bug. It's not a bug. The idea is, remember, we said that was unique about generative AI compared to traditional AIs, it's actually generating something new. It's creating something. And because of that, it's going to be different every time. So if you ask a large language model, the same large language model the exact same question 5 times in a row, you're probably going to get 5 different answers each time, because, again, it's probabilistic. It's at the -- in that moment, it's taking the information available to it, it's interpreting it, and it's giving the best answer that it can determine in that moment based on the context. The user type and any other training or guardrails that you put around it. So you always get different answers, and that's very different than a deterministic system, which kind of traditional AI would do. You could do like ML classification, you can say this is an apple -- sorry, this is a fruit. This is a vegetable, this is a meat or whatever. That's something you could do with traditional AI. In this case, a large language model will have to look at something and interpret with all the information available to it whether it's a fruit or a vegetable, but it might also be able to generate much more. It might be able to say, this is a yellow fruit that has a peel and is about 6 inches long, right? I mean that's a banana. So it you can do those types of things. It can do much more than what a traditional AI capability could do. But you're never going to get that exact same phrase every time you ask the question, you're going to get a slight variation on it because it's constantly interpreting with new data. And it's always going to give you a slightly different results. So that's what I mean by that.

Let me grab another question here. Let's see. You mentioned that a vendor, sub-vendors should also be compliant. Now this question adds, imagine a vendor wouldn't tell us who their vendors are, how can we ask the company we are working with to provide information on other vendors that they're using?

Yes. So I think the two things that come to mind are again documentation, right? Because that will be in the documentation, should be in the documentation to who they're working with and where they're getting their data, where they're going get where they're getting other things, which then leads to the next thing I was going to say was just transparency, right? The real question is about transparency. So the vendors you work with should always be willing to show you who the other vendors are that they're working with, who are the suppliers? Who the people who are giving them what they need to build their models? And now what your responsible AI team can take a look further into those other sub vendors, if you will, if they want to. And often times, they won't need to. Most of -- if you're picking a good vendor, they're going to have reputable large language models, foundational models. And that's actually a really good example of something you could be looking for is like what is the foundational large language model on which this GenAI capability is built. So obviously, there's lots of different large language models out there that you could potentially use. And many of them are very reputable, obviously, Azure, OpenAI and there's an LLM that Meta provides. It's actually quite good. So there's so many different models out there. But the question is, what if they're using a different model? What if it's a model that you've never heard of that is from a company you don't necessarily trust or is not a company you're comfortable using. You need to understand where that foundational LLM is coming from that the vendor is using to create that GenAI skill. So that's one thing I would definitely look for. A trusted vendor would be willing to tell you the answers to those questions and document that so that you know what's being used to build out the models. Those are the kinds of things I think that you should probably focus on.

So again, with that one, transparency is key. And if it's not transparent that might be a red flag is that what you -- would you agree with that?

That's exactly right, yes.

Okay. And I see another question, I can grab this one. Who should be establishing safe AI policies? So again, this really depends on your organization. It's always smart to go ahead and make those decisions as a group and not trying to do that unilaterally. Like a different parts of your organization are going to have different pieces of knowledge with and it comes to what should be added into a GenAI, into a safe policy. So you're going to want to consult those various different groups. So like legal, risk, compliance, HR, of course, and many of the other parts of your organization. So the best thing to do is really start by forming kind of a task team and bringing in people who are really interested in this and want to make sure that they're a part of building this out together. Did you want to add anything to that, Allan?

No, no. I think that was great.

Next question. Can you guys tell us more about ServiceNow's approach to GenAI specifically for HR. I'll hand that one back to you, Allan.

Sure, sure. Our approach to GenAI. Well, as I mentioned in the presentation, the approach we use is we use domain-specific large language models. And I talked a bit about those. But I'll just say, these are so important because they are purpose built for the enterprise application, right? They're not meant to plan your next vacation to Hawaii or get out of some other -- those are consumer applications. If you think of ChatGPT, go in there, you do some fun stuff in there, but then you would never provide that to your employees as is -- as ChatGPT . You do something to enterprise that up, well, we've already done that, right? We've taken these different foundational LLMs. We've built them to be enterprise-ready. And specifically, we've done that for HR and employee experience, right? So that means we've essentially built and we continue to build new features that put guardrails around the LLM, right? That really make sure that it's protecting the employee, the employees' data, enhancing the experience that they have, reducing risk for your organization and really going after bias and hallucinations. Linking another practical example of this. I think it will help kind of drive home the point. I mentioned that case summarization that we had for agents. So we talked about the fact that this case summarization gave you a nice little summary and that's great. But when we think about something like an employee experience, something similar, where perhaps an employee reaches out and they want to learn more about a sensitive topic. There's no out-of-box large language model that understands the concept of case sensitivity, right? So when an employee starts to engage with a topic like that, we want to get that over to a live agent, right? That's the thing we would want to do right away. And so when you put guardrails around the large language model with a domain-specific model, like we're talking about. You can actually detect and understand when something is, in fact, sensitive and that way you aren't going to push that over into a hallucination of the employee receives and it's like, "Oh my gosh, this is now a crisis." So that's an example where that plus then when the employee does get pushed to the live agent, we can take all that chat context, the back and forth between the employee and the bot, and we can actually summarize the chat summary, like we can create a chat summary for the agent so that they know right away and we can flag and say, "Hey, there's an employee relations-related issue here, you need to take this right away," right? So just thinking through that human experience, the impact it has from an experience perspective on the employee, the legal ramifications if we get it wrong, and then even just building it into the agent and productivity side of it, so that the agent can come in, get ramped up quick and know what that discussion was about. It's flagged as, hey, this is ER related. All those things make a difference and those things come from a domain-specific LLM. So when you ask me what's the approach from ServiceNow, I think our approach is we believe enterprise software should be safe. We believe that enterprise software should be effective. And we think the way to do that is not to just go hand people large language model, hand customers large language model and say, "Good luck." We think it means we take responsibility for building some of those guardrails that I've talked about so many times today.

I think we have time for one more, and I love this question, so I'll tackle this one. Why should we choose an AI vendor, our employees already have solutions of choice? So yes, your employee is probably the -- especially the ones who are like early adopters and love being on the cutting edge of technology, they probably have a favorite GenAI that they use like a ChatGPT or something like that. There's a few different things. The first one is, let's just talk about the whole philosophy of choosing a GenAI vendor. If you don't -- if you let your employees bring their own tools then there are certain limitations to that, right? So there are -- you won't be able to control as much in terms of all the things we talked about today, all the safety factors, all of the -- like where is the information going? What's being done with it? Is it being cleansed? Is it being sanitized? All of those questions, you can't control it because they're doing -- they're making their own decision about where that happens and where that goes. And again, remembering that 6%, it goes ahead and puts in IP and puts it in the information that shouldn't be in these solutions. If they're bringing their own solution, you can do less on your end to make sure that there's already safeguards in place to make sure that those tools are kind of helping with when that human error happens. So if you put something in place and give them a tool that will work for them, and again, remembering that this has to be something that works for your organization. So the last chunk of the presentation where we talked about a custom fit solution that works for you, because you're vetting it, do you know what should work best for the majority of your employees. So when you get to pick it, you can also know how security is going to work out, how it protects your organization and feel a lot more comfortable. And you should feel a lot more comfortable with the way that information is flowing and what's being done to help mitigate risk, right? And then the other is that tools like ChatGPT are not going to help you with some HR-related things, right? So if an employee is looking for information from their last pay stub, they're not going to find that in ChatGPT, right? These are some things that are going to go through your own systems, tools that are going to -- or they're not going to find information from your knowledge-based articles from a ChatGPT, they're going to be finding them off the Internet. So you want to make sure that you're also vetting solutions that are not just GenAI in general, but for HR, you're trying to enable them to do better work and be able to get things more quickly. And that's going to be internal information. So the general tools are not necessarily going to be able to help them out there. Is there anything else you want to add to that, Allan?

Yes. Actually, personalization. I just want to talk about personalization a little bit more because I do think that's quite important here as well. When we think about -- I use the example, you go to a consumer type of thing like you go ChatGPT through the consumer interface, you wouldn't expect an answer to a question like how many days of PTO do I have left? Of course, not, right? But what's interesting is even with an enterprise large language model that hasn't been kind of built for the purpose, it may lack the contextual awareness and the personalization that's necessary to actually give the employee what they were looking for. And if you really want to drive deflection, you need to answer the employee's question for who they are, not just generally. So for example, a very common use case that we see customers using GenAI for would be when an employee asks a question like how many days of paternity leave do I qualify for, right? GenAI is smart enough to understand the intent. The intent is, okay, this employee wants to know what they qualify for, for something called paternity leave, which is actually parental leave in business speak. But it's able to make the connection and connect the dots. It's able to go out and retrieve the relevant policy documents, knowledge articles, et cetera, to find where that information lives. It's also able to understand contextually who is this employee, are they full time or part time? Are they in the U.S., the U.K., India, where are they physically located? It might need to know, have you been here for a full year. It really depends on the policy, right? And this is what's really unique about generative AI. You can look at that policy document and recognize, okay, this is a parental leave policy. I understand that concept. And I know that for different countries, there's different pieces of information you'll have to look at to be able to answer the employee's question, give them the exact answer they need. So by doing that and looking at the fact that say, I am a U.K.-based employee, I have one year of service and I'm a full-time employee. For this particular company, the policy is that I get 29 days of parental leave or whatever. So that's what's really, really important here also is that we have the ability for the for the first time ever to really give the employee the actual answer they need that's contextual to them, and personalized for them and that's how you drive deflection.

Great. Awesome. I think that's all the time we have for today. So thank you, everybody, for joining us, and we hope that you've gotten a lot out of this presentation. Thank you.