Agentic AI Era: How ServiceNow Is Reinventing Enterprise Software | Utilizing AI Ep. 13

Hi, this is Amit Zavery. I'm president, chief product officer, and COO of ServiceNow. I've been working in the software space for 30 years and really excited about having a conversation about AI and software.

Yeah. No, I mean, there's definitely a lot of this conversation about what AI does to enterprise software, and I've spent 30-plus years in enterprise software, and I know how complicated and what it takes to run a business. And there's a huge amount of context, huge amount of understanding of a business's processes as well as kind of governance and security required. So there's a decent amount of understanding which you need to really bring in when you're running an enterprise business out there. AI is a very good technology enabler as well as an opportunity for us to redefine and rearchitect the products which we build today for enterprise software because it adds a lot of value. But having AI replace everything enterprise software does is really not possible. And we've seen this on a day-to-day basis. I mean, LLMs have come a long way today, but today not a single company runs just on standard LLM. They do require a huge amount of context and huge amount of capabilities which is built around those AI building blocks, but there's work to be done and that continues to happen, and enterprise software will continue evolving to really make sure we deliver what customers need without just saying that AI will take over everything out there.

Yeah. No, I think you're right. I mean, there's an opportunity now with agents to automate a lot of the business processes. There's always been this interest in enterprise software to automate and remove as much human interaction as possible, and agentic AI turbocharges that for sure, right? So the way to think about this is you think about a business process, what are the pieces of business process which you can now automate using AI. But there are a lot of deterministic capabilities also required because what AI does is gives you a probabilistic answer. Now you wrap that around with the data you might have for years in terms of running workflows like what we've been doing at ServiceNow for 20 years. What we do with these probabilistic kind of workflows or agentic workflows is add the data element and say, hey, is the outcome accurate? This is what you're supposed to do or not, and if it's not, you go back and redo things or bring in a human into the loop as needed. But that has to be done end to end, not doing pieces of it. So when you look at a business process, any of the standard ones, right, order to cash or procure to pay or things you might be doing for IT service management and other things like that, you require the combination of what a context is, what the data associated with a particular outcome is, and then AI enabling and kind of making some part of the decisions, but it cannot take over the whole thing without having some data associated with what the outcome should be, what we should have expected from that. So that's really what we're seeing now, and agentic processes are very powerful. It does reduce your cost. It makes you more agile as well as it gives you some ability to improve how companies run. But you need to wrap it around in many, many other things as well.

Yeah, so we've been working with all the frontier model company providers for many years now, and we have always been an open ecosystem. So the work we're going to do with OpenAI, which we've been using them as one of the large language model choices for our customers. In the first phase, we were doing a lot of summarization and giving ability for people to comprehend what happened as an action. The next phase, as we evolved some of those capabilities, is to really now use large language models for voice capabilities, for example, multimodal, multilingual, so that we can understand what the user is trying to say. But the intent in terms of knowing what the context is and then being able to action that, that's really the IP ServiceNow has been building for many, many years. So we still use these large language models to kind of take out some of the mundane pieces of our work and understanding a lot more of user data faster, and that's the relationship we've been building with OpenAI. We do the same thing with Anthropic in some cases, and we do that with Gemini. But our goal always has been to use whatever technologies are out there and the advancements in there, but then build a lot of IP around it to make those things much more efficient, much more context-driven, and much more valuable to a customer. So we're doing this for now, say, employees, right? Understanding that an employee is requesting help. What does that help mean? If they're requiring a fixing of a VPN issue, or they're requiring a new laptop, or they want to be onboarded. Once we know what the request is, we know what needs to happen because we've been doing that from the back end from an agent perspective with human agents before, now with AI agents, and OpenAI helps us now take that data and wrap it around to integrate with various different systems, but then our agentic processes and things we build in our AI platform is now doing the actioning part of it as well. So that's the relationship, and we keep on evolving that. But there's a lot of good technologies coming out, and we use that as a building block, and I would say 5 to 10% is IP from them, and 90-plus% IP has been built by ServiceNow to really get you the outcome customers want, because customers shouldn't care what you're using underneath the covers. It's really the outcome-driven kind of mindset.

Yeah. No, I think as you point out, I mean, we have expanded our use cases in quite a few areas, right. So what we did when I joined, I made sure that one, we are very clear about the customer outcomes, what we're trying to solve for, what are the areas where we can be differentiated as well as value-add to our customers. So we did get to prepackage agentic flows and say, you know what, customers, if you're doing this currently deterministic workflows, can you replace that with agentic while I guarantee you an outcome? And that resonated a lot with customers. Second thing we wanted to do was really think about controls, be it around governance, security, auditing, and giving customers visibility in terms of what they're doing with AI. Every customer I spoke to and every C-suite member I spoke to, they were worried about proliferation of AI technologies without any idea what happened, who did what, and who's running what, right? And that doesn't work in enterprise software. This is not like you can just click to a different website and start using something else. You have to make sure your business doesn't come down, or also anything which is an outcome which is wrong, and it really impacts your financial standing and everything else. So we had to make sure that we're building that controls, providing governance, providing all the scaffolding required, and guaranteeing customer confidence. As soon as we gave them a lot of this thing, and we launched this product called AI Control Tower, which allowed you to get visibility, lifecycle management of AI technologies, ours and third-party, as well as the ability to now audit and do risk management. Once we gave that to a customer, they started doing a lot more agentic use cases after that because they got that worry out of the way. So my experience at Oracle, my experience at Google, we really understood that AI at Oracle, of course, understood what enterprise software means, what it means to really run large companies' business without having any kind of wrong outcomes. At Google, gave me a very good grounding in terms of what AI is capable of, what you can do with it, and it's really an amazing amount of IP been built there. But combining the two is very tricky, and that's what I think ServiceNow has been able to do a good job of, is really taking AI building blocks, using the technology provider, and using it where it makes sense, but then really building the frameworks around it and making it easy for customers to get value out of it. So our AI Control Tower, the agentic workflows, the connectivity with third-party systems using AI agents, what we're doing from the perspective of understanding intent, and it could be a domain, right? We're doing it for HR, we're doing it for finance, supply chain, procurement, and now expanding it to CRM and then security and risk. Those things made customers much more comfortable that we're looking at it end to end. We call ourselves the enterprise operating system because of that, like going east to west and north to south. We are just not a verticalized stack, but really looking at business processes which cut across, and that has been the game changer for us in most of the customer conversations, and that's what we're bringing to the table to our customers, and they still feel very excited about what we can do for them.

Yeah. No, I think there's been, of course, a lot of promises out there, and as you said, there's a lot of announcements. What we've been very happy about and very excited about is the adoption we've been having with our Now Assist product. We have delivered that early last year, and as I was talking about earlier, this AI Control Tower, once we gave that to customers, they started feeling more comfortable going to agentic use cases. And give you an example of companies like Bell, where they've been able to now use all of the customer service, our AI automated agentic flows for doing deflection, but also understanding intent and resolving issues. So I'll tell you the common use cases we see in adoption-wise: incident management, resolution, solution of any kind of request, things like triaging, coordination of any kind of issue they might have run with disputes, like things we're doing with Visa when a customer requests an issue with their credit card bill. Having that been automated between a merchant, a credit card issuer, the consumer, and Visa in an automated fashion, and reducing the friction around that has been a game changer for them because they're reducing cost using AI agents, but it's an agentic flow, not just an AI agent talking to each other, right? What we have introduced recently is this idea of taking out L1 support for any IT-related requests, right? Autonomous IT, we call it, and that really changes the game for a lot of the companies because a lot of these questions and requests they get, we can resolve it as an agentic employee in a way who's doing all the work for the request and resolving it without having to really file a ticket. When you file a ticket, we resolve that without having human interaction. So those are all use cases we're seeing live being used, and we have customers across the board who are adopting it. Last quarter, we announced the adoption rate went up to 55x in terms of the amount of calls we're getting back and forth between our AI systems and the agentic use cases. So adoption is there. There's going to be a turbo shot this year, I think. But there's a lot more comfort in this area from our customers than probably previously out there.