Full Length Conversation
Venki Raman (VR): One of the most common questions we get asked by potential candidates is about the kind of challenges we face. Another one that comes up a lot, mostly indirectly, is the kind of problem solving culture we have. So we decided to pursue this theme for our fireside chat, which will ultimately create our blog post. Where do you think we should start?
Anton Kurz (AK): We should describe a little bit the journey we have been on as a company, and also in Product Analytics and Data Science.
VR: How deep do we want to get into this topic of scaling the function within a startup? Many tasks that appear simple now - like counting customers uniformly across teams - were not so obvious at one time. Something like standardizing custom metadata categories within an organization can be a sophisticated exercise.
AK: Yeah, we can slightly paint a picture, right? So in the early days, there was a master spreadsheet where all the processes run. But, ultimately, this did not work out anymore because it doesn’t scale, and there is a lot of manual work. So you have to go to the next level and bring in applications and databases, which can handle these things. Also, you can’t have one or two people who oversee everything. You need to outsource the logic into different teams and structures to bring some rigor. Ultimately, we want to understand who our customers are, what are they doing with the product, and are they using some features more than others? And base our decisions on facts, and not on one partial glimpse of the elephant.
VR: That is a great high level overview, but it goes into the details fast. The journey you described is already the "how". Maybe we should start with why we exist? What is our belief, our mission?
AK: That is becoming more clear in the end when we share all the details. We exist because B2B is not B2C. And it's very heterogeneous. And you don't have insight into all the accounts. And still, you want to base decisions on something more objective. The first thing is that we support and influence decision making. And the other piece is educating.
VR: Educating is a good one. It happens as a consequence of all the work we do. Data-driven decision making is a topic of great interest for us. We can talk a bit more about this. What does supporting decision making and/or educating mean to you?
AK: Decisions are made anyway, if you have data or not.
VR: That’s an important point, and it is central to the whole topic of decision making. People have ‘default actions’ they are going to take anyway, in the absence of any other information.
AK: When you can make educated guesses in your decisions, you are already one step ahead of the game. Then you can back-test your gut feeling with data. Having a stronger understanding of the customer base outside of your personal experiences, you are able to reduce personal bias. On the decision-making maturity cycle, this is a good step to be. But while maturing ultimately, for some of these decisions, you really want to test hypotheses. Which, of course, is a lot of effort and also has its own biases, but gives you a completely new perspective into the problem.
VR: Is the conclusion that decision making is better with data because the noise goes down? Broadly, people who are convinced of the scientific method, you don't have to convince them that this is important. For example, all of the B2C companies that do A/B testing, they do it because they can point to causality of specific features, right, and this makes decision making much simpler. Because they get really a yes or no signal on whether things are working as intended. And that they can do at scale.
AK: Yeah.
VR: But, in our case - it goes into the B2B vs. B2C aspect a bit. The motion that I see, like bringing data into the conversation and educating people, It's one step ahead like you said, right? This rigorous approach towards decision making or building a bit more rigor, is the reason it's one step ahead is that opinions have much more noise than data? Depending on how you frame it, data can also have a lot of noise. There is a driving principle here that we should try and articulate better.
AK: I would not say that data or insights bring down the noise, necessarily. Because in it itself they can create noise. They can also contradict each other.
VR: Right, exactly. It's a more sophisticated approach, but it doesn't mean it's always reducing the noise.
AK: Yeah, exactly. It’s an art and needs to be handled in the right way.
VR: Yeah, it’s an art.
AK: Ultimately, what it can do in a conversation is that it can narrow down the problem space or the solution space. For example, with reference to where the current market is or where the current customer base is, a lot of assumptions made and other arguments can be rejected very quickly. And therefore you can focus on things which connect to the real world.
VR: I like that. It’s a good perspective, that we are really in search of things that are closer to the actual reality. When you do it well, the narrowing down of the problem and the solution space, that is really creating signal from noise. If you do it well. That brings us to a good segue into the problem space. Maybe we should talk about the unique challenges of B2B. A couple of ones are obvious to me, and they have already come up. B2C, you can A/B test at scale, so proving causality is easy. The second thing going for B2C is that the consumer is also the buyer for the most part, so you can get a closed feedback loop of the content - if it is valuable for the consumer or not. B2B, you’ve already mentioned heterogeneity. Why is it a problem, especially when it comes to building analytics software?
AK: On the one side, what you mentioned, sample size is not millions, but thousands or tens of thousands. So, you have smaller subgroups. Also you have typical dimensions in the game like countries. Every country is a different playground with its own laws and regulations which impact the behavior.
VR: Also, there is a large history of marketing, right? B2C is backed by a few decades of knowing how to sell to people. And in the end, we do see that selling can be simplified to “What segments are you building the product for?” And there the market behavior is relatively clear because they are human beings with some expectation of uniform behavior.
AK: Exactly, yeah.
VR: I think what you are trying to say with regions, content, countries, and so on. in B2C is that the market is well understood in a way. So you define heterogeneity as the opposite of this. The behavior of sub-regions, or even within companies, or even companies operating in the same industry have highly diverse approaches to similar problems.
AK: Every B2C consumer is a human, right? Of course, there are differences in age, education etc. There is some heterogeneity there as well. But, ultimately, everybody is using, say, the App Store, in a specific way, right? Because they are also educated to use it in that specific way. But with companies, within the B2B market, it is also artificial to some extent. Humans came up with regulations, laws, organizational forms, and how they want to run the company…
VR: Yeah, it is man made heterogeneity, right? And this is not stuff that is susceptible to behavioral modification.
AK: Yes, exactly. And therefore, it's maybe not always logical if something exists. It has a very good business model, right? This company will prevail and run non-standard processes for more than 20 years because their business model and product is simply so awesome.
VR: I think that’s a very good definition of heterogeneity in our context. You cannot have any a priori expectations of uniformity in behavior. Really, that changes the whole game of how one makes decisions in B2B.
AK: How to collect the data, how to interpret it. It can be way more nuanced, and it’s very hard to kind of bring that approach in reasoning.
VR: This is generally true for B2B. Do you see extra complexity or challenges with just Celonis? I think being Celonis adds to the complexity a bit more.
AK: Yeah, I agree. Because in Celonis, we are really tapping into the organization of the customers. Because the processes are the veins, the life blood of the customers - how they interact, how they operate, how they run the company. And this is not really some ensemble statistics. Of course, every company is earning revenue, every company is generating outputs, which you can use also for some aggregations. But ultimately, what we are tapping into is how they live together, they work together, how they organize themselves in their processes, and these kinds of things. And also, how they depict the data, how they bring it up again into some analytics. I think this connects us, it makes us go one step deeper into the problem.
VR: I think “one step deeper” is our greatest asset, but also a challenge maybe when it comes to decision making.
AK: Not even as far as decision making. Just narrowing down the problem space, it's hard to understand and segment the customers. Because of the small space. You can maybe say something about a few customers but then it's very close to anecdotes because you're talking about a few customers.
VR: Yeah, I think the “one step deeper” problem is that for us, the customer is no longer an aggregate entity. That's your point, right? We are voluntarily exposing ourselves to the variations inside the term customer. Different departments, subgroups, different functions. And how they interact. That's where the pitch is - that Celonis can help with all of these inefficiencies. But kind of doing that, exposes you to greater uncertainty when it comes to decision making.
AK: Yeah, I think you can say it like this.
VR: Alright, so I think that's actually a great segue. Which is that Celonis is a sophisticated product. You've seen a lot of it, from the beginning. How would you describe the evolution of Celonis? Maybe we can keep the company and the product separate and just talk about how you have seen the evolution of Celonis as a product. Specifically, how does that play into decision-making difficulties?
AK: Yeah, and over the course of the time, right? Celonis has expanded and has grown in all the different dimensions. Not only in maturity - how to act with the data, how to process the data, how to do process mining. Which is also the core of how we started. Process mining as a functional application which you can plug in on top of different databases. But also, now we really expanded it to a platform. And also verticalized it with Business Apps, right?
VR: In your eyes, those are the two major evolutions. One is the movement to a platform and the second one is the verticalization?
AK: The maturity is also a thing. Basically, I think in the early days, we were not able to handle big data with reasonably fast queries, right? Now we can handle billions of records with really fast response times. And I think this is really some strong progression of the tool stack.
VR: How do you see this? When it went from being a plugin application on top of a product into a platform. Celonis has so many capabilities, right? It can do ETL. It has sophisticated process mining querying. It allows you to build apps. It provides you with off-the-shelf apps. There are so many things it can do and that's what makes it a platform. So what challenges does that bring, in the context of analyzing a product like this?
AK: So, I think here for us a challenge is also - it's a combination of platform and vertical business apps with a lot of possibilities. Every customer uses it in a different way and for a different problem to tackle. And it's also very hard to understand how they actually want to use it, for which problem do they want to use it? Because, ultimately, what we're seeing is the footprints of their behavior, right? So, we see traces in data integration, in the engine, in the studio, right? But reverse engineering and painting a picture of what was the customer up to? This is one of the challenges.
VR: Yeah, and finding common threads, right? Because that's what we do.
AK: Yeah, common threads are important. Is it working for the customers as expected, right? We have some ways and we have developed some techniques to identify this. Of course, there's the users coming back and complaining if something is not working. But then, of course, it's difficult to identify these nuances. Is it okay for them and they struggle to get the job done, or is it really a good experience?
VR: I think one deeper way to say this is that we can help identify where or for which sub-problem it is working for the customer, and where it is not. Because they're also in a constant motion of re-engagement with our product and expanding its use.
AK: Yeah. This is also the challenge of B2B eventually, right? There's a commercial and a conversation thread which is ongoing, but typically we have only the product usage to analyze.
VR: Yeah, and so should we talk about how extraneous factors affect our analytics?
This is a difficult and deep thing to discuss, but is there something we can say quickly that's meaningful about it? For example, customer success might be coming also from the creativity of individual customers, or from services contributions.
AK: Yeah, basically I think it's the difference between B2B and B2C. To some extent, the personas are different. Kind of because the buyer persona is outside of the product. And it really is that extraneous factors are way more important and you can't spot them. And I think here for a sophisticated product, we should also mention the scale and the amount of things we have to deal with. The number of services and features. Also the progression of it, new ones, old versions, deprecations, and these kinds of things. Which is a scaling challenge with a small team, basically, to deal with all this complexity and progression.
VR: Yeah. And should we make it explicit that given all this, unlike B2C, it is a big challenge to identify the value moments. And some of the work that we've been doing is kind of along those lines, finding those parts of the product that are related to customer happiness relative to other parts. I don't know if we can go too much into detail on this, but I think finding the value moment or understanding which parts of the product are actually delivering value is maybe the problem that is unique to scale with the wide feature landscape.
AK: I think it's a little bit deeper also. I think we are a platform, and I think in a platform in general, it is difficult to say where the customer is creating the value moment.
VR: So the point is, should we say anything about other problems like attribution?
AK: No, I think we can mention it with the platform. I think it is generally true. So, for an Azure platform, where can you tell that this customer is happy? For every customer it is a slightly different scenario and setting. So, it's very hard basically, to point out one moment.
VR: We can refer to the cloud platforms as a good analogy.
AK: Yes, especially for the platform. So I think for business apps, it’s clearer. It's more targeted and for a business app if they are coming back, this is a great reward. And if they are part of operational processes, if there are action flows, these kinds of things. It's different angles basically, but our focus is more on the platform as a team right now.
VR: We've spoken a lot about how the product has evolved, and that there's a wide landscape of features, and many versions, depreciation, and so on. Many challenging things in terms of making data useful in order to explain if the product is working or not. How does the growth of the company play into this? Because clearly the product is evolving because of some beliefs about the nature of business and what we are achieving. Do you think being a category leader and doing all of the constant innovation that we've been doing over the last few years has changed the nature of the company and that actually impacts our team?
AK: I think it goes hand in hand. So we would not be category leaders if we hadn’t evolved the company. We would not be able to serve the largest customers at scale. We would not have moved to the cloud, we would not have moved onto the latest engine stacks, and so on. But basically, at the same time, we were able to do it because we had these great partners in the industry who helped us and challenged us to do these things. So I think product and the market, they are basically pushing each other to the next limit and that's very good and great.
VR: Yes.
AK: A challenge for us, or one of the areas where we can and do have an impact, which is a reason why we exist, is because now with these amounts of customers, these different business models, it is separated into different functions also within our organization. We are basically bringing together these details and information to a common and global picture for Celonis, which is: helping decision making because we educate people, we show what is actually important to Celonis, we help to frame decisions and also to model scenarios on top of it.
VR: Yeah. The bringing together is an important aspect of what we do.
AK: We keep learning, So I think also, as a growing company, you have to learn how to operate and how to organize yourself in the next stage of your life cycle or in the next stage of size. It is a scaling problem.
VR: But I think the value of that is usually underestimated, I suppose. That we are not restricting ourselves to product, and we are saying that product and platform usage has some unique signals that are useful in other contexts. And that there's this cross functional reach of our data. Right?
AK: Yes, absolutely. So it's cross-pollinating. We have different analytical teams for different areas of our company. That doesn't mean that we are working in isolation but it's rather that we are the experts in certain areas. But one of the values analytics brings for decision making is bringing your data into the perspective of the other teams. Yeah, I think that's also learning we made over the last few years, how to organize, and to leverage some standards, how to collaborate. I think we are starting to see the fruits of it.
VR: Yeah, that's a really good instance, where we have worked together on standards across the organization.
VR: So what makes product analytics and data science irreplaceable? Everybody and everything is replaceable, of course. But currently given the context in which we are chatting, what makes us irreplaceable in your opinion?
AK: It's a very good question. Creating context for data. I think creating and collecting context for data. Which is putting certain data points into perspective using different approaches and methods. To come up with higher level abstracted concepts - be it deterministic, be it heuristics, or algorithmic like modeling - so that you can tell a story about the data.
VR: Why is it not easy? What does it take to do this well?
AK: What you need is rigor. I think that's where we can contribute. Having the experts in one team and dealing with these problems constantly, you develop rigor and some methods, approaches, and skills. To back-test, to give it meaning, to correlate with other data. Keep a constant reminder that correlation is not causation, right? It is a common user’s bias, a human bias even.
VR: Yeah, it is a human bias.
AK: We also keep our power stakeholders and internal customers honest, to keep objectivity high in data. I think that's also a part of some argument why our team is so responsible because we act as a kind of neutral observer.
VR: I think we've segued really beautifully between the original points that we considered. Now we are actually at another natural point. You said, it's important to have people in one team to deal with these problems constantly and in the end, our role is that of a neutral observer or a good advisor. So what do you like about how the team works, and what should we always be striving to do better?
AK: I think a general Celonis principle is that “the best idea wins” or that the best idea should win. And I think something that most of the time we try to live up to and try to keep very high in our team.
VR: Yes.
AK: And we discuss the problems until we come to an agreement. Of course, sometimes it also means that we have to accept, because we can't endlessly discuss things.
VR: Yeah, disagree and commit.
AK: But typically also we come up with a compromise to test out things. I think this is one very good thing and also helps us build a culture where we can be the neutral observer. We also can argue about data points and discuss with our stakeholders about these things because we are also honest with ourselves within our team.
VR: I think this is a very deep point, right? I think your point is that you cannot be a neutral observer for the organization, if internally you are not running as how neutral observers do, which is to evaluate ideas scientifically, try to deal with facts rather than opinions. It's not easy, I suppose. But it's something that we have to strive for, to be honest to ourselves that we run the team in the way that we would be useful to the organization.
AK: In the end, it’s the only way. Every team member needs to act as a neutral observer, so we have to train them to be one. It's a cultural thing, ultimately. Especially when it's about data analytics, when we can convince others with data points, arguments, and reasoning, then everybody's listening to it. We try to have a very scientific mindset.
VR: The idea is that we tell people to bring evidence for your beliefs, right? That's the signal that the whole team gets, which is to bring evidence for your arguments.
AK: Yeah, exactly. And also frame your hypothesis so that it's meaningful. The general Celonis cultural things (like FISA, Ownership) are important to our team also. But that’s a little bit standard to Celonis. Where we excel is really the culture of scientific thinking.
VR: And what adaptations do you see coming up in the future? Because you have seen multiple evolutions and adaptations that the product has had to make, the company has to make, and the team has had to make. We seem to be doing more and more broader work, more and more sophisticated work. Our pipelines are growing in size. We have many more customer endpoints over time who are using our data for decision making.
AK: Basically we are still growing very fast as a company. So kind of one challenge will be to definitely keep up with the company, right? As our team, we want to be ahead of the curve, so basically we don't come into the firefighting traps.
VR: Because that's a poison for a team like this. If you're in a firefighting trap, then you're not producing useful output that guides others. You need the mind space to create useful output.
AK: This is maybe another aspect. We have to also find and renegotiate contracts and engagement models with all the different teams, because what's working with smaller teams is not necessarily working with larger teams and global teams who organize themselves differently. I think these are some organizational challenges. Also, especially in these fast growing environments and when there's a lot of chaos and questions floating around, it's also a challenge to keep calm and focus on these few things which really matter and constantly develop in this direction. Typically these big improvements are not made overnight, but through constant evolution incrementally, to build the quality and the rigor to open a new area and dimension.
VR: Yup, and we are hiring.
AK: If you're interested.
VR: That's nice. I think that went really well.
AK: Yeah, I think the podcast style really worked well talking about these things. It felt natural to talk about these things.
