- Good morning everyone. Thanks for taking the time to stop by. It's always hard to follow CJ and Jensen up on stage, so thanks for coming. With that, we had to have a catchy title, right? "One Small Step for SAP, one Giant Step for AI", big, overwhelming, but we'll try to show our journey. And to do that, let's first introduce who's on stage. For those that don't know, SAP is a large software company, global vendor of business application software. And my name is Hardy Kuhn. - Yes, thanks and glad to be here. Hello everyone. Steffen Meyer is my name working for Accenture, the Global, the largest global partner for ServiceNow and having more than 18,000 people skilled and certified in the ServiceNow space. And at SAP, we started the journey very much five years ago with a local team from Germany and meanwhile have a global delivery engine running 24/7 working for SAP. So how was the journey? - All right, let me start with giving you a little bit of context before we get into all the AI cases that we have to show. SAP started the ServiceNow journey about five years ago, implementing it. You can see the timeline over. We started with a lot of process workshops in the very beginning trying to define a golden standard harmonizing across 25 lines of business in our company that all have different ITSM processes, different case management processes across the company. It took us about a year to really go live with the first things. And then from there on, we had one line of business after another go live. And about a year ago we started our GenAI journey, which we will talk about a bit later as well. Just to give you a bit of context of the size of this whole thing, we have 400,000 customer plus, we have about 100,000 employees at SAP and roughly half of them are on the ServiceNow system active. So we have about 50,000 agents on the platform, and about 3 million cases incoming every year. Two to 4 million incidents depending on how well we do in our cloud operation space. Large CMDB 26 million active CIs currently, just to give you rough order of magnitude. And what do we do with the ServiceNow platform? It's all about support. So customer support as well as cloud operations processes. So classically case management, incident management, problem management, but also event management and the correlation engine with incidents and so forth, right? So it's all about outages, minimizing them or repairing them as fast or taking care of our customers in cases. All right, with that context, give us a little bit of a framework. - Yeah, sure. I mean we have GenAI in the session title and for sure GenAI is the leading scheme here at Knowledge this year, incredible, but there is so much more on the AI and automation agenda that we drive together at SAP. Starting from automation and orchestration of workflows, reducing manual efforts in that one to data analytics, track our key KPIs along the processes to machine learning capabilities, virtual agents for streamlined end user communication to process mining and of course Generative AI. And all of these pieces fit together, all the components work together and can even enrich each other by triggering each other. So it's more like a picture in the whole. At Accenture we call it intelligent automation because the main idea is to intelligently enrich the capabilities of humans with the strengths of machines and join these forces together so that the humans can finally, again, refocus on the most value adding activities each and every day, such as being here together at the Knowledge and talking about AI instead of all the daily routine tasks. And today we brought four specific specific examples that we wanted to show. One is the classical automation workflow. Then we have an interesting case that we build an SAP on machine learning for task assignments and process mining and our value framework underneath and of course the latest achievements in Generative AI. So Hardy, would you be so kind to share a bit more details on these examples? - All right, let's take them one by one. Automation and orchestration is I think the classical stuff that you've probably also done on the ServiceNow platform. We brought one example out of many where we're doing patch management on triggered by the ServiceNow platform, right? Patching servers and batch executing them. We've got a lot of integrations to some automation tools in the background. In this case it's Ansible, but we also have Terraform and other tools from SAP that basically take over automation procedures. But basically it gets triggered by a service request on the ServiceNow platform and then gets pushed down the road and fully automated until it's actually done. I think, like I said, that's probably the most classical example of automation that you can have and that's why I don't wanna bother you with too much details. Second case that we brought is a machine learning case that we've been working on for quite a while now and it's creating a swarm. Let me explain what swarming is about. Usually when we have incoming customer cases, about 10% of them are very complex ones with multiple products and multiple different layers, right? It could be a scenario where one application is broken or where the middleware is broken or where something is going wrong in the data center. And to really analyze the situation, you can't really assign and pinpoint. So you really need to bring a lot of experts together and that is called a swarm. So what we've done is, we've built a machine learning capability and we use that to bring in the right people that need to be in the call for this case to analyze and then fix the problem. So that's called swarming and basically that's based on in machine learning from past cases, from resolutions, from who fixed which kind of problems. And if we see and detect similarities, that's basically when we pull these people into the case, they get automatically called and then we basically have a channel where they can start analyzing it. Once the problem is pinpointed, then we get to the normal resolution, right? So this is getting the right people together. Now the next case is pretty much about process mining and I wanna tell, talk a little bit about what this is. Usually when you look at your KPI dashboards, and we are a large company, so when we implement a feature, so we activate something new on the ServiceNow platform, you don't see our KPIs make big jumps, right? We all wish for that, but it realistically doesn't happen. So dashboarding and the operational KPIs is something not the full truth. Also, if we implement, for instance, a deflection feature where easy cases are auto-resolved or the customer doesn't even open up a case, the remaining cases might take longer to be resolved, right? So your KPI is going up with the resolution time. So sometimes it's basically a delicate balance and you need to do a lot deeper digging to find out if we created value. So what we do is we look at User Experience Analytics. So which feature gets used quite a lot, but we also look at process mining. How is the process actually running and which of the artifacts are the bottlenecks and how does it look like? And from that we create then a lube of a value tree and value case basically finding out how well we have done implementing this feature and how helpful it was moving the needle going forward. But now I think everyone talks about GenAI, so let's do that too. And Steffen, take us on what did we do in the beginning of this journey? And it was a real rocket start, thanks to the great folks at SAP that we have the pleasure to work with every day. So pretty much more than a year ago, meanwhile we brought up this new Generative AI thing in one of our regular innovation workshops and we discussed that there might be a huge potential pretty much across all the processes of the platform. And then based on that discussion, we pretty much had the same day decided, let's try it out. Let's start with an early proof of concept and experiment on all the possibilities and potential limitations that might be there. And then based on that, we started with an earlier adopter setup and a playground setup with limited data for limited users. Also to limit the risk behind it, but still to have a playground with nonsensitive data for usage within our own team. Based then on the very promising results early on, we started to further create idea ideation for use cases, we spawned a new idea of GenAI format, GenAI Day format specifically for SAP, where we are bringing together the various different business stakeholder groups, the different user groups, and also various different partners like the hyperscalers and vendors for AI solutions, ServiceNow of course as well. And also there we invited also the business counterparts in the prompt engineering labs to try it out by themselves. And by the way, you can do exactly the same here also here at Knowledge within our Accenture launch. So everybody is invited to go there and try it out yourself. And then we started with a pretty extensive testing phase because testing for GenAI is very different in comparison to usual, the classical test approach because even with the same inputs, you will always get different outputs so it's not reproducible. And also to define what a good enough output would look like, defers very much from individual to individual. So everybody has a different taste on wording, on level of detail, on the length of the output for example. And therefore we decided to go for a different approach based on feedbacks and based on voting in the end. So every user had a possibility and still has the possibility to upvote or downvote specific results. And then based on that feedback, we were able to continuously and iteratively refine the prompts to make the use cases as effective as possible for the users. And then by end of the year, we finally made it, we pushed everything to production. We managed to solve all the compliance and all the data security issues and brought everything to production ready to use for the largest user groups at SAP in the product support area as well as the IT support area. Then SAP became one of the first AI lighthouse customers globally and the rocket took off. And then on the use cases themselves, so what did we actually implement? And the possibilities are endless. So we had to prioritize in a bit. Prioritization based on value and at scale to make that actually happen. And at scale, I mean the use cases needed to work for pretty much most of the 50,000 agents of SAP working with the ServiceNow environment each and every day. And what we wanted to do is to achieve or to help in the understanding and also the communication during handling a case. We did that via a summarization of cases, of incidents, of alerts, bringing all of the information together and based on that information, propose the proper communication that the agents can reuse in this regard. Secondly, we wanted to boost the efficiency by collecting required information that might be helpful in the context of the specific ticket that would then also help on faster resolution. And by that we are able to save significantly time that would be otherwise wasted and going to different platforms, different environments to collect all the required information. So it's everything in one place. Last but not least, we wanted to streamline the decision making process and that especially for very complex cases that SAP usually they're running for quite a long time. So it's hard for an agent to get an overview on everything that's happened and then take the right decisions. This is also something, one of the use cases that we set up here, analyze everything in this regard and propose categorizations for example. And the use case and the results so far are super promising. I mentioned earlier during the testing, the feedback-based approach. This is still something that we do and more than 80% of the people that use exactly these capabilities are satisfied with the result. And at the same time with pieces like summarization or also supporting and enabling the searching, the collection, we were able to significantly improve also the speed of handling times. Now, one step back again. So we need to speed up, I got it. We need to, I mean, at the moment everybody is mainly talking about Generative AI and foundation models like GPT or mixed drill or others, but there's so much more of AI solutions available and depending on the questions to address other options might be more viable on that one, such as having a look on the past on Diagnostic AI features or identifying patterns that might repeat in the future with Predictive AI. So, Hardy, we did a lot at SAP, how does a picture look like? - Yeah, we're a software company too, so we invest heavily into AI solutions as well. And I think you can see the logos underneath. So for all of this spectrum, we have an own solution and own model, a trained model. Sometimes we choose to basically use out of the box capabilities from ServiceNow, and sometimes we bring our own model to the case. And I think that's exactly what CJ said in the keynote as well. You have to look at each and every use case, see how performance, how costs, how benefits really reap. So I think you can see that, and the race with large language models and specific large language models have, has not been ended. So I think we keep a lot of options open and we're experimenting with a lot of them. So you see, it's not just one thing, right? There's a whole spectrum from diagnostic all the way to generating texts and so forth. And I think that's very important that you're looking into each and every one of them, weigh the benefits, et cetera. But let's come to the learnings that we had over the year of this journey now. I think the most thing that everyone will look at is what benefit do I get? So how do I prioritize my investment? How do I get started? Where should I get started? Where is it even worth it investing? And I think that's one of the leading questions that you should have. In the very beginning, the answer is easy. It's get your hands dirty, start somewhere, somewhere, right? It doesn't really matter, the more you have learned, then you invest in the use cases that really matter to your business, right? But that requires a certain learning. So second point is my data integrity? Do you have the foundation for it? Is it digitized? Did you have a digital transformation that gives you the fundamental to reap all these data that you need? So basically you have to continue your digital transformation journey to be ready. And then I think how do I make the right ecosystem solutions? We're standing here with our partner Accenture together. So how do you do that, right? How do you choose the right partner? How do you choose the right ecosystem decisions, large language models that you select, et cetera? - Yeah, exactly. I mean we have just seen on the slide before that the options that are available are manifold and one of the key tasks is to identify the right ones for the right task. I mean, as Accenture obviously we can help on guiding that decision process on also integrating, implementing the various different solutions and integrate them into a seamless experience than for the users. Nevertheless, this really depends on the specific client needs and client applications. Then as important as ever, the people topic, the best technology is completely useless if people don't use it. And especially with the Generative AI capabilities, the fears are as high as the promises. So it's very early to take your people on their journey early on, educate them, enable them, and prepare them on what's possible and where the pitfalls are, which things should be avoided. So this whole topic of learning and education is of utmost importance. And last but not least, I mentioned that in the beginning how we started the journey at SAP with that limited user, limited data approach. The balancing of the value of the cost for the use cases with the risk is very important from an early on stage. So we did already pretty much for the first users cases that we built in the beginning, we did already, let's say mini business cases to really identify does it make sense to invest that or just grab that and maybe park that for a later point in time. And only by that, from our perspective, we can also make that tech successful. - Absolutely. - By that, we are at the end of the session today. I hope it was inspiring for you for only 20 minutes and spawned some good ideas for thought. Happy to pick up the discussions and throughout the Knowledge or right after, happy to do that. Hardy any thanks for being together with me on stage. - Thank you. - Right? And many thanks for listening everybody. (audience applauds)