(upbeat music) - Welcome everybody to our Now and Now session of bringing AI and GenAI to every corner of ServiceNow. Before we get started, just a quick safe harbor notice. This presentation may contain forward-looking statements. Don't make any purchasing decisions based on the forward-looking statements. My name is Nick Borgwardt. I'm a senior director of our AI experience organization here internally at ServiceNow. With me I have Claudiu Branzan. He is a director of AI foundations as well at ServiceNow. Today, we're gonna talk through how we look at AI internally at ServiceNow around experience, speed, and decisions and the themes around that, how we've infused AI across ServiceNow. We'll touch on some lessons learned and then close out with some of the results that we've seen. So at ServiceNow, we've been investing in, you know, building AI/ML capabilities for many years, then infusing those capabilities into various aspects of our business. And when I say AI, I mean, we've been using machine learning, process mining, automation discovery, and many more to power our platform and workflows. And now we're starting to leverage generative AI. As we develop more and more use cases for AI and now GenAI across all of our departments, we found that AI profoundly improves three core aspects of our business. One being experience. This is how AI is changing the experience of our employees and customers. The second being speed, how AI is improving the speed in which we get work done. What's happening behind the screen simply happens much faster, right, than traditional manual work. And then the third is decisions. Empowering decision makers to make more informed choices. AI is not a one-off project or a series of projects. We think of AI and GenAI as a critical part of our operating model. The second theme you'll see today is how we're using AI in some of our departments. We will tie these back to the experience, speed, and decisions, But we're gonna touch on digital technology, that's our IT organization at ServiceNow, HR, finance, legal, customer support, as well as sales. I'll focus on departments as a whole, but we generally get started by identifying the personas that we infuse with AI. As you can see some of our personas across various departments. To reiterate, you'll see these two common themes. AI capabilities enabling better experiences, speed in which things get done, and decisions that get made. And second, departments showcasing use cases. These are the corners of ServiceNow. Let's take a look at our first department, IT. This is a department where we have invested heavily in AI. Well over 30 use cases live in production serving various different personas. Some key outcomes I feel, you know, relevant for our discussion here is I'm gonna pick on the speed theme here. A 15% productivity improvement of time spent on cases for agents. So now I'm gonna pass it to Claudiu to walk us through how we're achieving these outcomes based on the capabilities in some demos. - Thank you, so as we see the themes experience, speed, and decisions throughout the departments where we picked a few use cases to demonstrate some of these themes. In this one in particular, we are gonna look a lot on how we are improving self service for our employees and how we also improve the efficiency of our support agents. So let's take a look at example of this use case. If somebody has an issue with their computer, they can come on our employee portal and click on get help and fill in and describe the issues that they have, right? In this case, Yuching has a problem with their laptop battery not charging. The first thing that we do using analysis is providing an answer based on an existing knowledge article to try to help you Yuching troubleshoot the issue. In this case, we'll recommend resetting the SMC. It seems like for whatever reason, the user decides this is not a correct answer. After reviewing some of the other knowledge articles that we provided, he will end up reporting this issue by creating an incident. Even in this case, machine learning and AI is used to automatically route this ticket to the appropriate service that handles this type of issues, in this case, laptop endpoints Macs. We're doing that using predictive intelligence. So now that this issue is being created, this incident is being created, let's see how typically our support agents work and how we're helping them through generative AI. So this is the view that a support agent has on this incident. The first thing that they might wanna do is click on the summarize button and get a summary of what has happened so far on this incident. This is a brand new incident, so only an issue is being reported here. Not a lot of other activities have already been developed, but as they go back and forth with the employee in this case and they try to troubleshoot this problem, right, they will make work notes, they will add work notes to the ticket. And then something else that the agent might wanna do is see if there's an alternative solution already available. So they click on that AI suggested solution. And as you can see on the bottom left, we're able to provide a suggested solution based on existing incidents similar to this one that have been resolved before, and the information is available in some knowledge articles that are relevant. Now, it seems like in this particular case, the information is redundant. There's still recommending to reset the SMC, but the true nature of the problem is that the laptop is old and it needs to be replaced when the battery is not functional anymore. So after some troubleshooting back and forth with the employee, the agent decides to resolve this incident. So at this point in time, as they're going and do the state to resolve, analysis generates a resolution detail based on the information that is already documented in the work notes. And you can see that populates resolution details. The other option is for an agent to also create a knowledge article out of all this information that's documented on this incident. So that's what the agent shows us to do here, right? So clicks that box of create an AI knowledge article, and then does the necessary steps to essentially resolve this incident. So after reviewing the resolution details, he picks the necessary information to be filled in and then attempts to Now, as he saves the state of the incident, you can see the work notes are being populated with what was the issue, what were the key actions taken, and what was the resolution. And this is automatically done by analysis, it's adding all this information into the work notes. If you remember, we decided to generate a knowledge article. So once this incident is being resolved, the knowledge article is being created. If you take a look at the title, you'll see that the title that is there is nowhere typed by the agent, but rather inferred by analysis based on the work notes and the information available in the incident. If we choose to open this knowledge article, we see that it already has a predefined structure. Analysis is able to identify what was the issue, what was the relevant environment, what was the determined cause, and what was the resolution that was identified in this particular incident? All this is documented in the language that is meant for users to follow as opposed to the language that was used for the work notes that were captured in this incident. So what we typically do at this point in time, we ask our agents to review this knowledge article. There's actually a process where a manager would go and review a bunch of these new knowledge articles. And when they feel like the content is good enough for publishing, they click on this publishing button. Now, behind the scenes we're running again an index process for AI search. So now these newly created knowledge articles are getting indexed along the existing content in the knowledge base. So let's see what happens next time a user comes with a similar problem. They go through the same route, they go to our employee portal, get help report an issue, and now that they type an issue that is somewhat close to in the description with the one that was reported before. In this case, laptop battery is not charging no matter what I do. Then we will see that the information captured in the newly created knowledge article is taking in consideration providing the answer to the user. So as you can see here, right, there's a mention of the information in the previously existing article, but also the new one, right? If resetting the SMC does not solve the problem and you have to resolve there, you need to order a new one, and there's a link to order a new laptop in this case. So with this, we're showing how you can take advantage of the knowledge and expertise of your support agents to quickly respond to new problems by documenting them once and really enabling your employees, or as you'll soon see, customers to self-solve and self-serve. - That's awesome. So let's see what we've been doing in HR now. Think how important it is for employees to have a positive experience engaging with HR, and simply going through a process that HR is accountable for. For example, onboarding or inquiries. One highlight that sticks out here is in the experience theme, 92% employee satisfaction of our onboarding experience. So Claudiu, let's take a look at HR. - Some further examples I wanna provide here just to increase the variety. We're gonna talk a little bit about, you know, why or how we're improving the experience of our employees by making content recommendations. And we're also looking at how we're helping our HR support agents better understand the questions that our employees have. So this is Rajeev Khurana, he is a staff AI product manager in my team. And if we look at our employee portal, the Me tab, we'll see that we're making some groups and channels in Microsoft Teams recommendations for him that are relevant based on what other people like him have chosen to enroll in and follow in the past. So it's through this, we're really enabling our employees to create better connections and discover and increase their network within the company. And it is, again, you can do this using predictive intelligence similarity and make recommendations like this. Another good use case of content recommenders is in our frED learning platform. Rajeev, since he's a machine learning AI practitioner, we recommend content based on his employee profile. We see a lot of courses related to MLOps and machine learning engineering. But also based on his recent learning history, it seems like he was interested in process mining and learn a bit more about our people strategy. Now, as a user clicks in one of these courses, there's another recommender that we are providing content through, and here we're recommending content based on the existing course. In other words, similar courses to the one that people are looking up just to further enhance the experience. As I mentioned, we're using now assist also to help our HR agents. Here's an example of a live chat between an employee and an HR support agent. As you can see, there's a lot of conversation that has already happened. You know, our employee in this case typed a lot of question. There's a pretty lengthy conversation. So our HR agent at this point uses the summarization function analysis to get a very crisp information summary of what has been discussed so far. In this case, the employee's interest to learn more about their dental benefits and has a few questions related to coverage and some of the details of the benefits. So with a few clicks, it is very easy for the HR agent to really understand what has happened so far in the conversation and provide answers, as opposed to having to go through all that lengthy process of reviewing and formulating his own answers. - All right, and now next up is finance. Some key outcomes, I think, relevant here for this discussion. I'm gonna take one that's mapped to our speed theme. 10% improvement in our order booking cycle time, right? That's a key outcome. And so, Claudiu, let's take a look at finance. - For this one, we are gonna pick a few examples that we haven't seen before. So I'm gonna tell show you a use case of process mining, right? Process mining is powered by machine learning and AI, and it's a really powerful tool to really understand how well your workflows work. And we're also going to touch on how we are enabling our employees to self-serve on topics related to finance. So here's a particular example. We did a study on using process mining on our workflow for purchase request to purchase order. We looked at six months historical data, and using process mining we're able to not only get this really good visualization of the workflow itself, of the process itself, but we were able to identify some key bottlenecks or key problems that existed here. And by taking some measures, we're able to improve the overall time. We reduced the overall time by a few days in this case. So some of the improvement areas we identified is, for example, 12% of the cases were moving back to draft because of missing information. So that was an easy fix. We pretty much modified the catalog form to always validate the content before submitting and making sure that we're capturing the right information ahead of time. 57% of the cases, for example, require contract signature, and that was adding two days to the cycle time. So we went and recommended policy updates to reduce the need for PRs to require contracts review by 50% of the current volume. So that in itself pretty much sped up by two days, 50% of all these in all these runs of this process. And then in another portion of this workflow, we've identified 11% of the cases are waiting for more than two days for BO response. Something that we've done there is we took a proactive approach of notifying the owners once this request has been approved so that the system doesn't need to wait for them to come back and provide a response, right? It's a proactive approach where we notify them, and they can take action right away. You can use process mining on any existing process, any existing workflow in ServiceNow. And, you know, following this type of example, I'm sure you're gonna get a lot of valuable insights into your existing processes. Here's an example of how we're using virtual agent in this case to really help our employees. So this is My Bot, our internal virtual agent. And let's say I want to increase the limit on my credit card. So I go here and I type limit increase on credit card, and then My Bot is essentially using analysis conversation catalog to walk me through the process of making this request. So first it clarifies what card I want to make this increase on, it's a travel card, and then is it a temporary or a permanent change? And yeah, I select the date that I want this to make the increase by. And then, you know, I define what is the limit, right? I could go to town and pick a very large value or something more moderate, you know, I'm picking something in the middle I would say. And then following the process of the catalog requested it asked me if I have submitted all my expenses in the last 30 days, and then finally it creates that request for me pending my manager's approval. You can see how this is a very user-friendly interface that collects the information required to fill in this request in a very succinct and efficient way. - All right, and here's legal. You know, we get the question quite a bit. You use AI in legal? Absolutely we do. And so if I pick out the speed theme here, we're seeing 30% deflection of legal related inquiries in terms of how we're using AI. So, Claudiu, let's see. - Yeah, in this case, I'm gonna show you another example of self-service for our employees on legal matters. You know, this time, not necessarily through the virtual agent, but rather through analysis and search. So if I go to our employee portal, and maybe I have a question related to our trading window, you'll see that the first thing that I get is a link to our application that we're using to manage and trade our stocks. So that is Fidelity. And then I also get a true answer or a very succinct answer to my question, right? So based on the trading policy, blah, blah, blah, it provides me the answer to the question relating my trading window. Down on the bottom of the page, you'll see your typical search results with a few more references to other knowledge article that are relevant. But also we get people also ask questions which sometimes provide very relevant answers to similar questions or questions that people had in this context to further enhance the experience of the user trying to find an answer for their own question. And yeah, this is, again, using analysis and search to provide accurate and succinct answers to people's questions, reducing the need for them to go and browse and further investigate, you know, to find their own answers. - All right, we're moving along. Let's go to customer support. Now, this is something that you all most likely have experience, if you think of Now Support or any help you received as a customer. Some examples here would be, you know, the search that we're running across our properties or how we provide some self-service experiences. Some key outcomes in the realm of customer service here is, going back to the speed theme, 30 minutes of average savings for our KB writers when we generate an article. So, Claudiu, let's take some examples. - Yeah, we've seen a few examples of how we're helping our agents, right, in supporting our employees. And you're gonna see another example on how we're helping our agents support you, our customers. But before we go and do that, we're gonna touch a little bit on one of the most recent dashboards that we've built. You might ask yourself, you know, "How are we tracking value generated by GenAI?" Well, we actually built a dashboard, and this is a snippet of it, right? So for customer experience, you'll see our leaders are looking into what is the value generated in dollar figures? What is the return on investment made? And then what are the hours saved? Because this is all about efficiency. And since we have all those, you can quickly equate for the number of virtual FTEs. You'll see that we're doing that across operation, adoption, and model performance, those being key indicators for us. So, for example, for summarize and deflect, right, we have a dollar figure in the total saved hours. We're also looking at what is the adoption, right? Seems like there was a bit of a downtrend there in the last month that we need to look at. And then also, model performance is critically important because the better the model, the better the result overall for our agents and for our customers. So using this type of dashboard, our leaders are able to track on a daily basis, on a weekly, monthly basis what is the true return on investment and what are the key areas that we need to focus on to better get advantage of the GenAI investments that we make through analysis? Let's take a look at an example on our customer support website. So if you go to Now Support and you have a problem, and in this case I go and type something like the scheduler is overloaded or stuck, the first thing that you will get, you'll get a summarized answer, you'll get a concrete succinct answer that will try to help you troubleshoot that. It is doing that by summarizing the information available in one of the most relevant knowledge article. Now, in this case, I find that this answer is incomplete. So what I do, I go and create a case. So this is a somewhat similar process to what you've seen for employees, but a bit more detailed. So I have to go in and provide further information. What is the problem that I'm describing? I have to select what instances impacted and whatnot. And then further provide, you know, is this a business critical or not? But then I have to further provide description of the issue. What were the steps that I took to reproduce the issue? And, you know, going through the process, creating a case. At this point in time, I'm intersected if you want, or provided with a further information that is attempting to help me troubleshoot this. It seems like in this case it wasn't a relevant piece of information. So I end up opening this case. Now, let's take a look and see what a support agent will see once this case has been submitted. If you look at this, right, so for the problem that for the issue that I have described, it seems like there's already been some activity. So this particular agent looks at the existing case and it sees there's a lot of scripts that have been run. There's a lot of logs that need to be analyzed. So the first that he wants to do is to create a summary, right? So he goes and clicks the summary button and analysis provides them with a very succinct description of what has happened so far, what was the issue, what were the key action taken, and if there's a resolution or not. So based on this information, it seems like at least this is a resolved case and our agent just needs to perform that and essentially close this case at this point. Now, alternatively, let's look at a scenario where the recommendations we make are actually good. So if somebody comes and creates a case, right? They provide some sort of a description of the problem they're having, in this case, general results show full description in the portal or better said not show full description in the portal. Again, by going through the process of creating the case, we will be interjecting at some point after we describe the issue further, define the steps to reproduce the issues and whatnot, we will be providing another succinct answer. In this case, it's a troubleshooting guide that is a summary of a knowledge article that's very relevant to the problem that I described that gives me a step-by-step direction of what I need to do in order to troubleshoot the problem and fix this on my own. So with this, you know, I'm essentially self-serving. I don't need to further open a case, and that also means for us, ServiceNow, you know, more time for our agents to deal with other customers who need more complex problems. - And last up is sales. In fact, sales is a department where we've also invested very heavily in AI, whether that's predictive modeling or question and answering. We have high adoption within sales. If I pick our experience theme, we've had over 70,000 queries when we initially launched a new solution that we had rolled out for sales called Sales Assist. So sales is certainly on board with AI. And, Claudiu, let's take a look. - So Nick already alluded to Sales Assist as being something that we've built for our sales organization. So let's take a look at what it does and what are some of the capabilities enabling our sales. So this is a conversational search type of experience. So if anybody has a question related to our products or our marketing materials and whatnot, they come here. In this case, I come and ask, "What is the analysis for HR?" You'll see that I have an answer that is generated based on multiple articles that are relevant describing this out of our product documentation as it gives me a bulleted list of features of analysis for HR. You can also go and take a look at the underlying sources, right, and confirm the information, but most importantly, people now have the ability to ask a follow-up question. And this can go on and on, right? In this case, I go and ask, how do I enable Now Assist on my instance? You know, now I understand what Now Assist for HR is. You know, we are also capturing feedback from people at every single point, and we're using that to further improve the conversational experience. But this could continue forever, right? This could be a means for somebody to really research a topic or get better at understanding, you know, how to position ourselves or better sell in a sense our offering to our customers. - So a quick recap. We covered many different departments. We showcased use cases, we shared some of the outcomes that we've seen. Now, I'm gonna hone in on specifically just Now Assist and the outcomes that we've seen with Now Assist use cases. These are fascinating. First and foremost, we've realized a $10 million annualized cost takeout in productivity saving. That's without factoring in any of the benefits of software engineering, but we see that happening. So, you know, we expect that number to increase. The second is, you know, we've done all of this by changing search results, the search experience, enhancing our virtual agent, adding case summarization and resolution for our agents. The metrics that are driving the biggest reduction in cost is deflection and deflection for all types of cases. And that's also adding back to our increased agent productivity. We've done all of this with responsible use in mind and in process. We have four key components when we speak to responsible AI. One is being human-centered. The use of AI follows our human-centered AI guidelines. The second being inclusive. Customer supported AI research and design program to ensure diversity and inclusion. We also have our domain specific LLMs directly integrated into our Now Platform. Third being transparency. Customers can control over their data sharing, they can opt in or they can opt out. And the fourth being accountable, customers always have control over their data sharing. So the accountability is provided back to them. AI has enabled us to scale as we've grown, right? And that doesn't mean it's easy. We've changed the culture, we're shifting left opportunities and operations, and what's that really mean? That means we're transforming the skills of employees. They're doing more complex activities. You saw today a lot of what we're doing, you know, now through demos. That's just some of what we've done. If we had hours, we could show much more. More importantly, we're also constantly focused on what's next. And that's very important. That's how we keep, you know, our AI journey moving every day. And what's coming next? We see digital workers self-healing flows, focusing on automation and AI together. To double click a bit into our talent readiness and how we're, you know, transforming skills, I just want to, you know, call out an important statistic here. You know, last year we had over 1 million applicants for open roles. On average, that's about 600 applicants for any particular role that's been open in ServiceNow. Yeah, and we need new skills, right? We need to develop them, we need to train, and we're hiring for the right talent when it comes to AI. We have upskill and training programs available for existing employees. We have the right workforce strategy and the right locations at the right levels. And we're enabling people by embedding them in the right product organizations, getting them exposure to how products are built, how they're maintained and operated, and people with the product mindset and the best engineering talent. - Now, people are really important, but also we emphasized on the process. And when we started this journey, especially around generative AI, we put together a strategic program within digital technology. And we used a model of a hub spoke to really design this program. You can consider this as the central hub being composed of multiple functional leaders and technical leaders from across different departments in ServiceNow, you know, some of the core people like myself, providing services and building new capabilities that will enable this program. There were people focused on governance and true enablement pretty much showing others how to use these things. But we've also had what we call supporting hubs, right? Folks from our legal and data governance organization really coming in and leaning in with their expertise in providing guidance and governance on this program. We had our colleagues in security and our communications departments helping us making sure that we're mitigating all possible risks as we went through this process and as we turn this program into a successful endeavor. But also because ultimately this is requiring change of the organization communicating very well and getting everybody on board. Now, the real value generator, as much as I would like it to be in the core capabilities hubs hasn't been there. The real value generator was provided by what we call our solutions folks. These are leaders and key individuals across the key teams, across all the departments. Whether it's our customer experience, whether it's our go-to-market to name a few, finance and legal as we've seen some of the examples really leaned in into this organizations who know their OKRs and drive their KPIs on a daily basis to come up with use cases and implement generative AI solutions that are really driving those KPIs and those OKRs to better outcomes. And as I mentioned, the hub was there to support them on the way, right, by providing the capabilities and giving them the necessary enablement and support to do so. So this has been a very successful endeavor. Now, throughout this presentation, you might consider that, oh, those are very good outcomes. And I know some of you have already started your AI journey, but there's many of you who are maybe contemplating whether to do so or not. And some of you might even be intimidated by some of the things that you've seen here. And the message that I wanna leave you with is do not be intimidated by any means. You can literally start by looking at things like If you remember we used that, we gave you an example of routing and categorizing records like routing and ticket and whatnot. You know, process mining is a really powerful tool that will help you really understand better your existing processes and workflows, and then essentially provide you for means to create a roadmap of where you can gain benefits by automation and by using AI. You can also look into enabling AI search for workspaces and portals to really revamp, or improve the employee or customer experience by providing them meaningful results and meaningful answers out of the knowledge base that you have, Or you can enable Now Assist whether it's an ITSM, CSM, or other products that we offer depending on your needs to get things like summarized answers to provide that really conversational aspect as part of employee experience or, you know, help your agents resolve issues and cases faster. Or you can use Now Assist in search or for question answering. Now, all these things in one way or another, those are things that you can enable really quickly and, again, benefits within 30 days even. Before we part ways, here's some key takeaways that I want to highlight. First, you know, don't think of this as something that we've done overnight. You know, we've implemented AI into our operations, going deeper over time. Meaning, we started with some of the simplistic models back in the day, right? Automating the process of ticket routing, then going and making recommendations here and there, then improving on how search results are being surfaced and utilized and whatnot. But the way we've done this was through an approach of constantly looking at what is the lowest hanging fruit, how we can further develop a specific area and gain more benefits from using AI. The other aspects is, while we went deeper in certain areas, we also took a broader approach. Remember the hub and spoke model. We had spokes across the entire company. So what we've done is we took this holistic approach across all departments and challenged all of them to start leveraging GenAI. in order to gain traction and to gain, if you want, momentum was through anchoring on key results, right? We never made a decision on investing on building something using AI without having a key result in mind. Something that will really drive the needle for that organization. Now, we obviously ran everything on the ServiceNow platform. For us, it's not a choice, it is our platform, but we do believe it is a good strategic platform of choice for many of you, since most of your data, most of your workflows are already running using ServiceNow. And, you know, the last advice I would say here is even if you don't get to choose ServiceNow as the sole platform, don't have too many choices. And it's very hard to combine AI engines to work together across different platforms. So limit yourselves to one or two or very few of these platforms to enable AI as opposed to having a variety of them and having to figure out integrations. With that, thank you. Please scan the QR code on the screen to view other sessions presented at Knowledge and to learn more about how we're using ServiceNow technology across the organization to run our own business more effectively. Thank you.