Hello everyone, welcome to another Targeting AI podcast at AI Business. My name is Esa Shitu and I am here with my editor and co-host, Sean Sutner. Today we are pleased to welcome Alex Rinke, co-CEO and co-founder of Celonis. Alex started Celonis as a process automation platform right from his university days at Technical University of Munich with co-founders Bastian Naminacher and Martin Klenk.
We would like to kick off from that point. Why did you and your co-founders found Celonis? And where with AI were you at that point? Why did you see a need for a process automation platform? And now as a SaaS company, how are you pivoting toward AI?
SF, great to meet you and good to meet you, Sean. Good to see you. And yeah, no, thanks for having me, first of all. And you're right, we started 15 years ago out of academia. And back then, we actually didn't start as a process automation company. I said it was really about process mining, which is something slightly different. Let me explain. So the idea was we wanted to build simulation models for businesses. So think about the same way you simulate a factory or you simulate landing a rover on Mars. We wanted to simulate a business operation to make it more efficient and simulate decisions. And in order to do that, you need data and context about how that particular operation works. Let's say you want to simulate a call center or a warehouse, you need to understand how many packages are coming and how they are flowing and where they are getting stuck and how many trucks are showing up. So you need a lot of data and information about the company's processes. So we had actually built a simulation model and an AI engine to simulate business processes, and then we went to the first customers and said, hey, do you want to use this? And they were like, yeah, we're super excited about it. So we got people really excited, and then we wanted to implement it, and we said, we just need a little bit of information from you. We just need to understand your processes, what you have documented, how they are flowing, where they're getting stuck, how many exceptions you have. And then they were looking at us, and they didn't have all this context. They didn't have all of this information. And if you ask three people, you would get four answers, right?
So the problem with AI back then was that the organization didn't have the process context to actually inform intelligent models, right? You could build models — not as good as today, you can build way better models today — but even back then, 15 years ago, you could build models, but the problem was that the input was missing. And that's why we've built the leading process intelligence company. We also just got recognized in Gartner's Magic Quadrant as a leader in the digital twin of an organization category, which is sort of a related concept. And the idea is that we can bring together all the context about how a company operates its processes, and that turns out to be a super, super important input for all forms of AI.
So now as you want to deploy agents, and you asked about how we are sort of evolving Celonis in the AI era, we think Celonis is a really, really important player in that space, and so do our customers, because we have the best platform to collect all the context about how your processes operate, monitor them, govern them, observe them, and so on. As you want to automate and identify your processes, AI agents need to understand all that, right? They need to really become specialists at your business. One of my customers asked me, why is AI so great at planning my next trip to London, but it's not that good at automating my core business processes? It's not that good at planning my next factory, it's not as good at running my supply chain yet, right? Still need a lot of people to do that. And what will be the difference is that AI knows a lot about a lot — a lot about tourist attractions — but it doesn't know how this particular customer operates the supply chain, doesn't have that 360-degree context about everything that's going on in that organization. And that's the gap we close, and that's also why we introduced our Celonis Context Model a couple of months ago.
Okay, so I guess, correct me if I'm wrong, but you're calling yourself a process intelligence platform now. Okay, so that's a long way from the old process automation, process mining, and even farther from RPA. Now, can you talk a little bit more about the future role of process intelligence with the advancements in generative and agentic AI?
Yes.
Yeah, great question. So you think about work, right? You want to automate work. How does work happen today in a company? You have a mix of different things, right? You have systems like SAP, Oracle, Salesforce, etc. They do part of the work and people work within them, and there's automations that work within them too. Then you have work that happens completely offline — Excel sheets, documents, Teams chats, meetings captured in transcripts, stuff like that. Then you have work that happens in traditional automation workflow systems. Then you have work that happens in AI agents, right? So work happens across all these different forms in a company. And then you also have the work that happens today, but then also what your intention is, right? Sometimes you say, hey, I want to actually do this differently. I want to redesign this process. I want this to run differently. So you also have all of your process designs, your standard operating playbooks, procedures, strategies, etc. So processes are this complex form factor that really navigates across all of these different things — how you've designed it, how it really is executed, and it's executed across all these different formats.
And what Celonis is and the role of process intelligence is to really build a context model to bring this information into place and make it useful. So we can monitor work that happens on desktops outside of systems. We can bring in your Slack conversations, your Teams transcripts, all of your unstructured data. We can bring in how you designed your processes, any changes you want to make, and we can bring in all the logs from systems such as SAP, Oracle, Salesforce, etc. And we are building a digital twin of an operation. So we are building a digital twin of your finance departments, of your supply chain, how these things are connected, what the metrics are that you measure the work by, etc. And then we enable AI agents to access that information very easily. And it doesn't matter where you build your AI agent. You could build it in Microsoft, for example, it's very popular. You can use any type of model. You can use an open source model. You can build your own model. But what Celonis ensures is that this model always has the right context about how work is performed in your organization in real time, right? It knows everything that's going on. So it's this digital twin of the operation.
You are talking a little bit about your context, and I wonder how does that work exactly in terms of — is there certain permissions given to the model in order to know what context to know and what context not to know? And I ask because you mentioned the example of asking ChatGPT to plan something for you, for example, plan a vacation — they can do that because we know that models have been trained on the internet. But in the case of businesses and enterprises, they don't necessarily want the model to have access to everything. So how do you ensure security and privacy while also providing the right context and making sure that the model has the right context?
Yeah, I think it's a great question, and we have very sophisticated access policies. Our system is really an enterprise-grade system. It's used by around half of the top 200 companies in the world — companies with very high standards, and they really own their data and they own their context. And Celonis to them is a complete glass box. They can see exactly what happens. They can audit what the agents are doing. They can audit any actions that are being performed as part of the process. They can decide which data is exposed to what LLM, where that LLM lives. So we don't want to lock people in. We want to open things up for people and be a glass box, right? And this is really how people own their context and own their AI, because you don't just want to feed all this context into some provider and you don't know what they're going to do with it — because that tacit knowledge and that context about your workflows and processes is actually part of your competitive moat, right? How you serve customers, how you build products, how you manufacture, how you run your supply chain. Those are things that are really, really important parts of your moat as a company. And now with AI, you can make all of that better, more frictionless, less expensive to run, more scalable. Celonis helps you do that, but it helps you do that in a way where you own your future.
Sorry. Oh yeah. So you have said in the past, I believe, that there is no enterprise AI without process intelligence. So can you elaborate on that, and also go a little further on how AI agents and models struggle inside large organizations if they don't have visibility? So how do you see AI in general reshaping process intelligence or even process mining?
Yeah, I think a few things in that question, Sean. And first of all, tokens are expensive, but we spend a ton of money. We have almost 4,000 employees, we spend a ton of money on tokens, and so do all of our customers. And you want to make sure that you get an ROI, right? And the way you get an ROI is by changing existing processes. You make them faster, cheaper, better, automate them. And you really can't transform what you don't understand. And AI can't transform what it doesn't understand. So by connecting it with this process context and process intelligence that we provide, it knows every step that was executed. It knows all the context. It knows the why people made certain decisions. It knows what the guidelines, the rules are, how you want to run your business. And it has this context. It is more accurate. It is way cheaper. It's much more efficient. We can actually measure this. The token efficiency you gain in certain use cases is only possible because of this context. And you will be able to measure how much better are we getting every token that we spend, how much is it actually driving efficiency in terms of shipping to our customers, in terms of doing things with fewer resources or doing more things with the same amount of resources, in terms of optimizing inventory. So having this observability of your processes is a really, really critical part of knowing that you're on the right track with your AI journey. So without the right context, without process intelligence, AI is not going to deliver a strong ROI, and you're not going to be able to reinvent all of the work and the processes that happen in your company.
Thanks.
Many enterprises are rushing to deploy AI agents across their workflows. What are the biggest risks you would say that companies face when automating processes that they have not mapped or optimized with process intelligence first? And then you mentioned kind of like cost. I wonder how you guys deal with the cost question. And I ask because cost seems to be top of mind. Recently, when I'm covering models, there is this conversation of like tokonomics and not token maxing and stuff like that. So how do you guys help make sure that in the process of helping your clients with process intelligence that you're still helping them maximize the cost?
Yeah, let me maybe take the second question first. So how do we make sure that you don't just token max but you outcome max? And the way you do that is by measuring the cost of the process, not just the cost of the tokens. So you have to measure how much does it cost us to process an invoice or to sell to a new customer, and Celonis does that very, very well, and how is it getting better and cheaper and faster by implementing these AI agents. So for every agent you can now accurately determine the ROI. And that leads me to your first question, which is what is the biggest risk? Well, I think one of the biggest risks is to deploy agents but not actually become more productive, right? I think that tangible productivity, specifically once you go beyond knowledge work — everybody knows if it summarizes our emails, great, if it makes suggestions, if it helps us automate generating a presentation, that's tangible benefit, I don't want to deny that — but now as a big enterprise, you run a huge supply chain, like our customer Novo Nordisk, you conduct clinical trials, you have these very, very complex, high-stakes operations that you want to make more efficient, that's really critical to your business, and that's where process intelligence comes in because it helps make the AI work and it also helps you measure the result.
Okay. So in June, Celonis partnered with Deepset, an enterprise software vendor, to launch a sovereign AI platform. So can you describe how you see sovereign AI and how this partnership will help further sovereign AI in Europe, which Mistral seems to be pioneering — doing the whole thing, infrastructure and the models both. So how does Europe or sovereign AI fit with that, and where do data quality and governance come into that sovereign AI and process intelligence?
So, yeah, we are very excited about that partnership and other partnerships that we're doing to really allow our customers — whether they are public institutions, governments, government agencies or commercial businesses — to really own their destiny. So we see a lot of demand from Europe but also around the world on bringing in Celonis to own the context, own the data and own the intelligence that fuels the AI. And then you can make Celonis, through our partnerships, interoperable with open source models, with self-hosted models, etc. And we also, through our acquisition of EKG, allow you to build specific models on our platform, for example for decision intelligence. So we allow our customers to really own their AI, and we're also embracing open data formats, for example, so that you're not locked in, right? Our platform is transparent, it's a glass box, and you don't need to say, hey, I'm going to bet my future on this AI model, I'm going to give them all my data and then hope that my business is going to get better. And that's a theme that resonates strongly with our customers right now. There's even SaaS companies and enterprise software vendors these days that are trying to sell their customers' data back to them, right?
So they basically say, well, you have all this data in my platform. If you want to access it, that's going to cost you, right? Customers don't want that, right? And by the way, there's some very prominent examples of that. So customers don't want that. They don't want to buy a software product, put their data in, and then not have access to that data or do whatever they want with that data. And this is where Celonis has a very different approach: we are a glass box, you can zero-copy the data in if you have it somewhere else, and you can make your AI better with Celonis, but you always see what it's doing, and you can use it and you can access it from anywhere, and it's not going to cost you extra, right? So that's really, really important, and it's one of the reasons we are seeing so much momentum — enabling our customers, together with our partnerships, to own the AI journey.
I don't think you quite described what you guys think is your definition of sovereign AI.
Well, there's, I would say, different nuances, right? Some government customers, they want to own everything — the infrastructure, the model, the context layer, the data layer — and basically run everything in their own infrastructure and own data centers. We enable that architecture, right? We work with some very high-stakes government customers in security and intelligence so that they can really own the whole stack. Then there's customers that, for them, sovereign AI means, well, we are okay having sort of third-party infrastructure, but we want to control it, we want to manage it with our keys, with our encryption, with our accounts. And then to others, sovereign AI means, hey, we want to use an open source model and control the model layer. So I think there's different definitions depending on who you ask and depending on the customer requirements. But it's all really the same idea: as an organization, I want to own my destiny. I don't want to outsource my intelligence. I don't want to leak all of my secret sauce, my tacit knowledge, to anybody. And I want to have very, very clear transparency controls around that. And I also don't want vendors to work with me, help me, and then sell my own data back to me. I don't want that either.
You've mentioned — we've talked a lot in this conversation about the context layer, and you guys also have a partnership with Oracle Cloud Infrastructure.
