Process Mining at Rabobank with Frank van Geffen

Welcome back to the Mining Your Business Podcast. I show all about process mining data science and advanced business analytics. Jakob, how are you doing today? I'm doing fantastic, Patrick. That's great to hear, because today, we are joined by Frank von Geffen, process mining lead at Robobunk. We are going to be talking about elephant paths, goat paths, object-centric process mining, and how post-traumatic stress syndrome are related to process mining. All of that and more coming up right now. (upbeat music) First of all, since we are the media partners of ICPM, don't forget that ICPM conference will take place in a few weeks between 23rd and 27th of October in Rome. We will be there. The question is, will you? If you're curious to hear what this conference will be about and who will be attending, then go back a few episodes and listen to our interview with Max Rugglinger and Stefan Nier in Neralama. Anyhow, son and his world tour has been helpful. After all, if not for the world tour, I would have likely never got in touch with Frank von Geffen, our today's podcast. Frank, welcome to our podcast. - Thanks for having me, Jakob. Pleasure to be here. - It's our pleasure as well, Frank. And I went back all our episodes and I actually counted that you are our 10 Dutch guests. - Yes, Patrick, 10 already. And my question is, is a process mining architect a more popular job for kids in Netherlands than I don't know, maybe becoming an astronaut? - Wow, I'm not sure if it's anything of a process mining or the space agency of the Netherlands, I'm not sure. (laughing) - It's a good question, yeah. Yeah, it is added to education more and more often. So also the academia and the schools are aware that this is a field that has a lot of potential and it needs also a lot of people to be able to do this. So yes, I'm not sure if it's on the level of astronaut, but hopefully we can make it as attractive as that. I personally wanted to become an astronaut. So in that case, I've missed my colleague. - So you're settled for the second best job, which is process mining and possess. - Exactly. - That's how it goes, astronaut process mining expert and everything else, but. - Frank, as I was mentioning, I actually got not really to meet you because we haven't really talked in the Selonis world tour, but you were a speaker there. And my first question would really be, how are you enjoying this process mining events where you can actually meet all the other enthusiasts, such as yourself? - Yeah, well, usually I already went to a lot of these events, of course, so I find them very open in the sense that everyone is very eager to share knowledge, to connect, have a good understanding of where you are at at this point, can you learn from others? So there's always a very good vibe, you always learn new people, meet new people. And yeah, it's the feeling that I have, over the past few years, is that it has become far more commercial than it was for of course, but still people are down to earth, open-minded, and eager to learn. So that's what I like most about all these gatherings. Now, considering that you are a process mining enthusiast, you Frank, how do you get to be a process mining enthusiast? Can you tell us a little bit about the journey that you have undertaken to become where you are, who you are right now in the space? - Yeah, so in essence, it already started when I graduated, because I looked at the mortgage process of Rabobang, and I looked at handover moments in that process, and being an information management master, which was my study, it's all about information exchange in processes, and that's actually where everything comes together. So there, I already started with a view on, okay, how are processes unfolding? That was the question. And I did not have process mining at that time. It was around the time period, 2002, that we'll already discovered process mining, so we'll find out. And at that point, I heard about this discovery through my promoter, but it was still in the early ages, so I could not actually use it. But after that, about 2008, I struck on an article, IT audit their article that said, like, I have found this technique, and I just can give me an objective picture of what's going on in processes based on log data. And I thought, yeah, this is actually what I needed back then. So not making sequence diagrams by hand, but just feeding the data, and it will draw the sequence automatically. Well, in that sense, my whole career has been focused on process improvements for the one hand, like just from a lean perspective, reducing waste, rework, changing mindsets of people, on the other hand, the IT part. Like, okay, how can we redesign processes such that you can get the right parts automated? So how do you then become an enthusiast? If you know that for every improvement, you need to start an improvement project or program, you need to start with a good idea of the current way of working. Then process mining at that point in time was the answer, because I did not have to go to all the local banks and ask them, like, how are you working now? Look at that pile of paper at that point in time on the desks, go to the eight floor, where the account managers were that needed to sign off on the mortgages, for example. No, I just could see how they were working from a distance and could then get a good idea of what was actually going on and then start a discussion like, okay, what needs to be improved? How does this just need to be redesigned? So yeah, from that point on, I already was sold, of course, but I think what also helped was that Raubank also saw this potential. So I had a contact with one of the innovation managers and he actually facilitated me with budget to start exploring this. And that's actually the point in time I also sat down with Wilvan Raubank and just had long talks about how could this be made more practical? How could we as business actually do something with it? Yeah, and from that point on, it just became sort of a hobby next to my job. And yeah, you all know that now it's a professional process mining consultant or process mining lead. So in short, you're never really work when you do your work as your hobby, right? And what I really liked Frank, which you also told me when we had our introduction cause that you have a scientific mindset, but you are a business focused and that really spoke to me because I like this thought. And you also mentioned that over the years you are working with different focus groups. You mentioned Wilvan Raubank as well. What I would like to ask is you had quite some support within your, with your employer in Raubank that pushed you towards this cooperation with universities. And one of the very loud voices in academia is that businesses don't want to cooperate with academia. They don't provide them with the information they need. They don't provide them with the data as they need. And what I would like you to answer for as here, what recommendation would you give to people who might be as enthusiastic as you are about let's say working with these universities, but feel like providing their data or going a bit more into the depth with universities is just beyond their reach. - Yeah, so the advice is actually a very simple in the sense that if you actually want to learn new stuff from your data, give them to science, data science universities because they can apply all kinds of new algorithms on that which gives you new perspectives as business on what's actually going on. So what we actually did at Raubank was provide a full data set of our service IT service management processes. And we did that as part of the BPM challenge that was also facilitated by the University of Eindhoven. And we anonymized, of course, the sensitive data so it was not traceable to any employees or agents or something, but that was also not actually needed because we had this overarching question like, can you actually predict what will happen in terms of incidents when we implement a new application based on all the previous implementations that we had, all the patterns, et cetera. And then also look at like, okay, how can you see the effect on the processes that are evolving behind all those incidents. So they could actually focus on multiple aspects, not only using process mind techniques, but also other data science related techniques to help us. So in that sense, it was valuable for us because we got a lot of new insights, things that were possible, things that weren't possible at that time. But also the academia then had a training set, a huge large training set, which they could use to verify new algorithms and test their developments. And in that way, you can actually help each other learn. And I think if you really want to, one of my managers in the past set a nice quote, he said, if you actually want to win, you need to learn faster than your competitor. So in this sense, we need to learn faster. So how can we actually get new knowledge? Well, usually that is also working along with academia. - So, I mean, yeah. - It's a really interesting point because I mean, we've talked to professors and people in this field, and they say getting data sets is close to impossible. There's just no availability. And if anything, the data sets are super small. So getting a large data set is really critical. But coming from a company perspective, even getting process mining implemented in a company, it throws all sorts of security and data protection questions around. And I can't even imagine how much of a hurdle it could be to then say, okay, you know how we got all this data. Now let's also just give it a way to somebody else. And did you encounter a lot of resistance when wanting to do this, or like, how is that process? - Yeah, so in a mom way, we always have this resistance. In the other way, we have now so many procedures in place. We know very well which data is privacy sensitive, which data can be used for ethical profiling, or which actual data is competitive sensitive, competitive, so we have a good sense of what data we have, and what we don't, and we have learned the jurisdiction around it, sort of the new GDPR, et cetera. So if you have some experience in the data analytics space, which, well, a lot of us now have, we have this large data analytics tribe within the Rao Bank, for example, I think each bank now has more than 40 data scientists on the list, et cetera. So they now know where that fine line is, and they also have set up laboratories where they can freely experiment with production data without anonymizing things. But it's a big lock on the door. It's like 25th floor in the basement. So you can't get anything in and you can't get anything out. But you can do all kinds of experiments where you also learn what kind of data do I have, and what might be sensitive and what not. So when doing this experiment with the university, we actually said like, okay, but we see all the benefits that we get in return. I saw, I said to Jacob in our previous meeting, one of the graduate students that now is active at our company actually saw this data set experimented with it already and said like, this is such a good company that wants to share the data with us. I'd like to graduate there because I, yeah, I'm really liking this mindset, like you want to learn, you want to share. And in any business that's difficult because there is this competition like you do not want your sensitive data or your competitive advantage be known to others, right? But in the other hand, if you don't share, you don't learn. So there are ways and ways in which you can easily share without actually, well, taking too much risk. But it's actually something you need to believe in that it will help you. And yeah, in that sense, I now have a lot of experience so I know how to go about this. If you're just starting, you need to find out where this line, where this fine line is. Now you don't want to actually get into profiling customers and then sending them all kinds of marketing signals like come by from us because then you're in, the wrong part of the of the data picture. So yeah, and also in the past, I had this elaborate conversation with our part, our jurisdiction around all the data. And you also need to think about the goal for which you want to use the data. It's gathered with a certain goal. And you're analyzing it with sometimes another goal. If you're analyzing it for business process improvement, so your internal business processes, for example, you have a lot of playing room because your employees actually accept Germany. But most countries, your employees actually already signed off on their data being used for optimization of the processes. So you have a lot of playing room already there. If you know that, if you have the, if you know that you just need to find the right goal, you will also see that it fits within all the laws and regulations because also they have already thought of that. Because you need to be smart about it. So you need to know, okay, what is the legislation? Which goal can I use it for, which not? And if you just involve those people at some point, just let them know, okay, this is what we're about to do. Do you see any objections? Well, then they will say, I would anonymize this field or do this or take that measure. Also security professionals, for example, they are actually inclined and willing to cooperate. You will see that most people like to learn. So, and if you just peel to that side, then they are willing to cooperate because they also had their say. And sometimes, yeah, in a large organization, it takes some time, of course, because you need to involve multiple different disciplines. But you will get there. And as soon as you've done it once, the second time is more easy. I would say if Rabobank, you know, company working in probably one of the most regulated industries, which banking sector for sure is, and if Rabobank can do it, then everyone can. Frank, maybe a follow-up question to that. So, you mentioned that one of the direct implication of working with academia was that you managed to find yourself a colleague who currently works in Rabobank because he or she was excited about the data and the possibilities that they have in Rabobank. What I would also like to ask, if there were some interesting findings that academia brought to your company, which was a direct result of sharing these results with them. And if these findings eventually translated in some sort of changes in the company. - Well, I did a lot of presentations on the premises of the academia was invited multiple times. And what you're actually seeing is that from the point on that we started to work together, I was periodically invited to share how we go about in practice. And then they would share what their theories were, what they are developing, what is 10 years ahead. It's like what also will be discussed at the IPCM and in the scientific track. And what you're doing seeing is that I think the most thing that we learn here is that you're put back with both feet on the ground. It's just a Dutch expression, which I now raped in English. But you get a reality check. So if someone says like it's gold, this process mining and then you talk with academia and they say like, but did you realize that 100 man years went into the development of this algorithm and that this is one of the 1500 different algorithms that we designed. Then you get an idea like, okay, wait a minute, we're just looking at one little piece of this entire puzzle that they are developing. So this exchange not only leads to well, in the challenge concrete insights into, okay, how predictable is a new implementation of a application? How predictable can we make the incidents? It's very difficult. So that is a very situational, very context-sensitive. So what you actually are learning, like these event logs just give you one picture of a process evolving, like what is registered in the systems. People are just doing a lot more around just using those systems. So be aware, and that's my background as a process analyst, process designer. Be aware that there is a lot more happening in those processes than only those algorithms are showing you. And that they could actually determine for us. So like, okay, this is the maximum that we can do with these algorithms. This is the insight that you have, but you have still a lot of extra data needed just to make, for example, a prediction about incidents starting to occur. So back and forth, you learn from each other like the depth and the width of the application of the algorithms on the one hand. On the other hand, we get a reality check like, okay, this will take like 10 years of development still, and they get a reality check like, okay, but this is how we are actually using it now. Can you also optimize the performance, for example, of this algorithm and what's needed to actually do that instead of developing new stuff like OCPM, for example. And so that is a bit of the exchange and what we get out of it. Now, Frank, you are discussing here academia and cooperation with universities. What I'm also wondering, now hearing this, it almost seems a travel bank is light years ahead of competition when it comes to applying process mining just because of these activities that are already ongoing. Could you summarize for us a little bit the current state of things of process mining with the newer company? - Yeah, so maybe also good just to mention that a bit of a sidetrack, I also found it a group of people within the Information Professional Society somewhere around 2012, I think, 2013, which actually was set up to promote the practical use of process mining. So we invited all kinds of companies in the Netherlands just to share their pilot experiences. And also there you saw that because I already started very early, we were some steps ahead in knowing what the possibilities were. But you also saw you are learning from other companies in the sense that each process is unique. It has different characteristics. If you take a mortgage process at Abian Amro, for example, and compare it to Rabobunk, there are always subtle differences in what you are doing. In general, it might be the same, but you can't enforce the same improvement in both contexts. It does not work that way in practice. So yes, we might be ahead of in sense that we know the extent to which these techniques are applicable. And I think if you now look at the state of where Rabobunk is, is that we are enabling our business units to perform process mining themselves. So as a sort of a self-service context, then you're noticing that the whole role of a process mining analyst is still unknown. So it's like a sort of a hybrid role of an economic trist, or you call it, there's someone studying economic math. So operations research, that kind of education, versus a process design, process expert kind of role, combined in one. So and you need to also have analysis skills. So problem solving and deep dive analysis skills. So having learning about like, okay, I'm now trying to educate you to do process mining yourself and then seeing what we still need to develop in terms of skills and knowledge. And we need to divide all those things across different current roles. We are at the point in time that we have enabled several business units to perform process mining themselves. I just looked at the list of regular requests and we are already at 58. So and that's 58 different processes that have been analyzed by the businesses themselves since we started by about one and a half years ago. So I think the community is growing. We're also building a community around this called a bit of a process intelligence community. And what you're actually noticing is that it is leading to a very high operational enthusiasm so people at the operations side are actually buzzing around the whole topic and more and more people are trying to start this up. But the executive buy-in lacks legs behind. So you see a gap in between you have the middle management layer which actually holds everything or slows down every adoption of every change. And the executives actually could speed this up by actually sponsoring it from an executive perspective. But that's actually what we're looking at now to actually get that executive buy-in and linking those operational enthusiasts to their own executives. And then at that point in time, we will probably be able to grow a lot more and also overcome challenges that any company has in the sense that we're not always organized in the process or chain way. So we're suboptimizing every expert is optimizing his own part of the customer journey. But in a whole nobody has the actual view. And we're now able to create that view but someone has to act on that. So we, I guess, are at the point that we are actually addressing those organizational challenges. We know what is possible, the technique has proven itself. We have a lot of people now actually being able to conduct it themselves. But who is taking action? You can actually compare it with a hospital if you're introducing a MRI scanner or something like that. And you have surgeons that can only operate on your leg, for example, then you have a picture of a brain but you can't do anything with it. Or it's just optimizing your brain, but not your leg. And it's all interconnected. And we now see with all process mining initiatives how interconnected things really are and how complex the whole landscape is. But then you're actually seeing that the culture, so data driven working, had deciding based on data facts. That's something that needs to be improved but also the way we organize around a customer experience. Now has this fostered conversation between departments, say one, one, people in the head department and the leg department, they start communicating, you need to move the leg this way because blah, blah, blah, does that, have you seen this happen? - Yeah, yeah. So at the mortgage side and the business lending side, they are like years ahead of the rest of the organization. They are actually now connecting all those people. So they have this entire view of the entire customer journey and who is responsible for what? And now they are actually connecting. There were improvement cycles of courses. So there is a, we need to run the bank on the one hand and we need to change it on the other hand. And there are already ways of working of doing that. So they are now integrating these process mining insights into that current way of working, where you see that different roles need to adjust, adapt to the new insights that they're getting. So they have other or more enlightened discussions in the sense that they now can see, okay, if I do this, this will have this impact on you. So they now need to elevate their discussion to another level and see, okay, how can we actually cooperate? And by having that insight, they are actually confronted with the fact that now I really need to talk to those people. It's not only just keeping tabs like, okay, how are you doing? No, it's actually now they are seeing this dependency, they're experiencing it. And we can also measure if things, when things have been improved, we are measuring or we are generating these scans again. And actually, it's been pointing to the fact that they are still not talking to each other. And so this is actually the most confronting thing that's happening now, is that the few people are actually having this overview now and they are actually saying like, this can't be happening, right? It can't be happening that we're just seeing, we're looking at the same picture, we're also all recognizing that this is a reality, that's actually happening like this, and you're not talking to each other, what's happening? So yeah, the most interesting thing, and I think the largest hurdle to take, it's also recognized with our executives. I just got an email from my sponsor saying, okay, I discussed your initiative now in the management team and I get these hugely positive signals. Like they are all flabbergasted and all so enthusiastic about the value that it can create from the operational risk side, the head of operational risk said, like just exposing things that are happening not as designed, already lowers the risk profile, just by exposing the fact that this is happening, that they are not adhering to what we are designing. So it's not only that we are trying to improve those processes and optimizing, but just exposing it, is already lowering the risk profile because of the insight it generates. We are now aware, objectively aware, that we're deviating from some designed plan. This is also flabbergasted, right? How are you not aware of the fact that you're deviating from your own designs? It's actually, it shouldn't be possible, right? But it's all in the heads of people at the moment and we are now actually making that knowledge explicit. And on the other hand, they said like the main impediment is that if we want to improve, we cannot actually find a process owner or a process owner that actually owns the whole journey. And so people start just suboptimizing their own part and they are not interconnected. So they're actually also recognizing this as a huge challenge. And I'm very curious what will happen in the coming months once we might get that executive buy-in. For example, how are they going to actually address this organizational/culture/corruptive? I'm not sure what it all entails, but I think the essence of how we work together and how we organize ourselves. And it's actually a very interesting time to see if we can like influence this enough that there will be a sort of a change in how they deal with these insights. - Frank, you mentioned an interesting thing which was enabling business units. And you also talked a little bit about different roles and also that you have to train your people to work with the MRI scanner which process mining technically is. And I think you're hitting Nail on the head here because also within different process mining vendors, there are maybe not everyone thinks or not everybody has the same goal in how the process mining technology should be using. So there are vendors that are targeting more the BPM side of things. So rather redesigning processes, there are vendors that are targeting automation or are introducing operational use cases where you really just take a little subpart of the process and train your people in basically making this small piece of process more faster, more efficient. When you say enabling business units, does it really mean for you that you teach them how to think about process mining and how to really navigate in a process explore themselves or what does it mean for you? - Yeah, so they are actually using, they are configuring the MRI scanner themselves and they are actually using it themselves. So as they start, we help them. So we do it for them, then we do it together and then they do it themselves gradually. So we help them mature. So in all the trainings that we do, we start, for example, what actually is process thinking or what is process management? How do you define a process and what is it like in reality because things happen in parallel? There are all kinds of automated actions that happen at the same time. It is just created and they have the same timestamp. How do you deal with that? Because a normal process mining tool without a sequence column cannot get the right sequence out of actions that have the exact same timestamp in milliseconds, for example. So there are small things that we're teaching them like how to look at a event log. What is an event log? Where is the event log suitable for a process mining tool? And then you're seeing me and my colleague have a lot of experience together in a lot of very different event logs that we've seen over the past years. I also wrote several articles about date quality together with Anaroji, not, for example, where you actually see all these different quality aspects that you can encounter when you get event logs. So we teach them that. We also teach them how to look at variation, how to zoom in, how to zoom out, which parts are elephant parts, for example. Also actually starting to think about, we actually say like we have two different approaches. We have the exploratory approach where you actually can just look at the process and well, let it let the tool let the MRI scan and tell you where things might be going wrong. It's like without too much assumption or the confirmatory approach where you actually have, you are feeling the pain, you have a sort of hypothesis like it might be due to bottlenecks there or there or it might be due to long waiting time. Can we actually spot that? Can we confirm that this hypothesis is true or not? And that latter approach gives you more focus so then you will not lose yourself and that we experienced it in the analysis. So this is what we experienced a lot already. People are overwhelmed with the amount of variation that they are seeing in those pictures and those diagrams. And yeah, we also teach them to simply, because as Rahul when we chose the tool solones in the end, because our purpose is to use it for all use cases that it's possible to use. So we use it at the moment as an MRI scanner, not per say to actually directly intervene in processes, but actually to give this visibility. And then to let the process managers in this case or the people from the business think about or reason in another way about what needs to be improved. Or can we actually measure the improvements that we have or how effective are they? But that's the way we're implementing it now and you're seeing a spillover too. Like could we also simulate scenarios or could we then not actually refresh that data more often and just monitor what's going on? But I said first learn what event looks are and that not every system is actually logging them as you would expect and how to make a translation to a functional business activity. That's then I think the most difficult and the most important step. If I have this system logging like cases updated and then I know okay, that might be some normal update action. I don't know what is actually updated. What is the contact details of the customer or something? Or how do I know which actual business value a system step has? So in all those things we are training them to look at it from a business perspective start with a business question. So do you actually have some pain? Is the market share going down? Are customers complaining? Are what is actually going on? Because you do not go to your doctor and say like okay, just scan me because yeah, magically you will find a tumor or something. And then he will say I don't have the equipment to scan you for those kinds of things. So maybe I will refer you to the hospital that they have this expensive MRI scanner and then we will scan you. But you might not find anything. Are you willing to pay for that? That's also what I always say. It's like this is an expensive tool. Are you actually willing to pay for such a scan? Because and are you actually then say we find something? Are you able to act on it? And that's where we are enabling them to integrate those insights in their improvement cycles. So that they already think about okay, if we get these insights, how can we follow up on that? Frank, one of the things you mentioned before the recording is making this X-ray is still highly underrated even in today. So vendors are throwing at us new and new possibilities which you can do outside of this simple process explorer. Why do you think that the X-ray itself is still highly underrated? - Okay, so I have multiple nice examples of that, of course. But I think the most striking example is that I'm working together with the process manager at the business landing site. And he said, yeah, we have a huge problem here. We are too slow and we are losing customers because we're just too slow and we need to keep them on board and just get them the finance that they need. So we need to improve on our lead times. And we have these ideas. We have a program that's already started, but we want to confirm that the improvements that we are already suggesting are the right ones. And that will give us the, well, reduction in throughput times. And the second problem that he had was like, okay, I also see that we're not compliant. We're not following, we're skipping steps. We already could show that in an exploratory process mining is like, okay, if you compare this with your design here, you have the elephant paths. They are quicker. So might you suggest that every bank follows those elephant paths? You are quicker. So you might be able to help more customers. Okay, that would be an idea, but that's not compliant because we skip necessary steps that we need for making sure we've done the right analysis. So at a certain point in time, we measured and we analyzed that process with the process explorer and used a bit like this conformance checking functionality. And then they started to improve. And after the fact, the same period of time for a half year of data, I just opened that process explorer and I said, I see three times the amount of cases that have been processed in this half year compared to the last half year. I also see just in the process explorer, they are more compliant that the elephant path have been reduced. But what I also see is the amount of employees only has risen with 1.3 instead of three. So they are doing twice as much work. Is there someone spurring them on with a whip? Yes, he said, there is some stress at the local banks. People are overworked and we have asked too much of them. And that kind of behavior is just apparent from the process explorer. If you just tweak it with the right KPIs, you can see actual human behavior. And that is overlooked often. I've seen a lot of simple behaviors, like when people have meetings at the local banks at Tuesday morning and at Thursday morning, why can I note that? Because the controller opens a report on Monday evening and the director at 5 to 10 as before the meeting. And it's happening all over the country. So we have a very strict or good cadence in the way we organize our meetings. Something we did not know before that we have actually implemented this meeting structure very structurally within the Netherlands. Okay, interesting. But another way, yeah. So I think that's one of the nicest examples that I saw in the past also at the ISCPM was a researcher at the Art University from the psychiatric department that actually said like, okay, we are researching War veterans, their post-traumatic stress symptoms. We are letting them play war games. We have these scans on their heads. So they scan their brain, the brain activity and we have some other sensors. And now we also activated event log in the game. So we can actually see each click that they do. And she just put a process visual that the animation part of the process mining and functionality in the process explorer. And she said, okay, here I have two players on the one hand player on the left player one, on the right player two. And you saw those bullets just go along the lines and then they started to circle very quickly. As she said, now they are shooting at each other. It's like, yeah, okay. So you can actually see game behavior in the sequence of patterns that actually are happening. And they could actually now make a certain behavior objective in the sense that if we actually put a matrix on there, we can see that evasive behavior is more effective for these people than attack behavior. - And you can tell all of that just through the process explorer. - Exactly. - It's the movement of things. So how quickly are they moving? Which way are they moving? Which direction? What is the next following action? Are they going back? Are they taking a long time? If you're just in a management team meeting and you're showing such an animation and you see one case just edging along the line and all those other cases are going through, they are just focused on that one case. What is happening there? So you can highlight behaviors very quickly. And if you have some of the, if you have these actors here, so you know that someone has performed the action, you can just count how many people are in this process, how many actions have they performed. If the volume is actually as risen, throughput time has been as going down and the conformance is rising, with the same amount of people, then they have become twice as productive. Yeah, that's something happened there, right? So it's very easy to show all these things from the basic functionalities that the process mining has. - So previously you mentioned elephant paths. I'm not sure if that's an expression but I'm not aware it's a goat bed or a. So it's like when you design streets in a city and you have a bit of grass or something and then the sidewalk just goes to the left and then to the right and then you go to the right again. So a lot of people just go across the grass and they create that direct path. That's what in Dutch is called elephant path and Will van der Raal makes a lot of reference to this site. It's called elephantpaths.nl. It's only from the parties for the Dutch people listening. You know, they know what I'm talking about. And usually people that use systems always find these elephant paths. They always find the back doors. They always know how to peep the system. So they know how to get around things that are obligatory. And that's actually what you're seeing in these event logs. And sometimes it's faster. And so sometimes they consciously do that because then they can help the customer quicker but they forget that certainly within the banking industry that this administration is needed for compliance. And so to show that we've done enough to get the risk profile of the customer that we gave the right finance that matches his or her profile, et cetera. That's something very important within the banking industry. So if you just skip those steps then or do not register them in the system, we cannot objectively show that those steps are actually taken. - Hearing all of this, it seems that most of the businesses are still, or at least from my experience, a lot of the businesses are still at the very beginning of exploring their processes and analyzing the process paths and really digging into the event logs of a single processes. And now at the same time, we are already producing a new technology in OCPM, object-centric process mining, which in essence is connecting those seemingly isolated processes into one holistic view of your organization. I find it fascinating but also very challenging to wrap your head around. Frank, you also mentioned that it was introduced to you by a graduate student, also to RebelBank. And it's RebelBank currently also exploring the possibilities that the OCPM has. And what is your take on this? - Well, recently we presented our use case of OCPM through a salon as webinar. Maybe you can search for it at the salon as site and you can review it. It was my colleague Mike and our graduate student Tim. So what Tim actually did in his graduation thesis is he had the question, can processes be predicted more easily or next actions more easily using an OCPM instead of a traditional event log? That was his larger question. We as RebelBank said, we would be happy if we just know how to build an OCPM model. So yeah, multiple goals, just showing us, okay, how does this work in a practical process and showing us the value of course of having that. So we saw with our forensic economic crime, it's the know your customer process is so a good use case for using an OCPM because they have multiple sorts of reviews that they're doing in that process, which you can then view more holistically. So that's the actual use case that they researched together and yeah, in that session of Salona's, they actually explain how this can work in practice from that forensic economic crime perspective. He also elaborates on the challenges that he had to get a OCPM. So what we're actually doing is researching this. So now we have actually a good idea of what is needed to build an OCPM model. It's not easy. So there where you have to use your suspects like order to cash, kinds of processes, these objects are very stable, they are all known. So Salona's could be able to provide those up front in the sense that you only need to connect activities and timestamps to them and the objects you already have. That's all fine, but in the finance industry, we do not have those models yet. But on the other hand, we have all those objects in our process management tool implementation. So we know, as we have this information architecture also where those objects that you should need are already defined. So it's more like, okay, in design, we might already have those things. Now we need to see if we can actually connect data sets in such a way that we can easily form an object-centric model. From a use case perspective, you could think about a customer journey, for example, that starts with contact moments. So a customer contacts us through online, through app, or a phone call, or he just still able to walk into a physical bank, where it is possible. And then starts interacting with us. And at the one point in time and employee actually starts doing something offline, registering things in the system then gets back to a customer. And it goes through different departments that our side customer is doing different things in different channels. You can already see the different objects or the different metro lines as salones calls them, flowing along each other and they sometimes interact and sometimes not. So I think the most useful use case is from a customer journey perspective is that you can actually show what the customer is not interested in, like what we are all doing just to keep in touch with this customer. And on the other hand, if we have these internal processes that actually go across different departments, it's really helpful to, I think, deal with the intersections. So it's the transfer, the handover moments that I graduated on. These become very visible in the OCPM models that you can actually get that parallelism and that they connect somewhere and then they deviate again. And at the connection points, there is handover, usually. So you can also just reason about, okay, who triggers, who, which step is triggered by which other step. So yeah, in that sense, we have looked at it intensively. We now know what is needed. And now, actually, I think the question becomes how easily good we scale this to other processes. So that's a big question point at the moment because he's now finishing up his graduation thesis at him. Yeah, and then the question becomes, okay, are there already processes suitable? Is it valuable enough? Because it takes time to create those OCPMs. Once they're in place, okay, are we then able to already use the accompanying functionalities? Are they mature enough already? So that's still a bit of a question. But in theory, yeah, so like in the other realms that we are working together with academia, we are already prepared for the scenario that we could start to get value out of it. Now, what I feel like the OCPM might do is that it will just throw everything at one pile all these processes. So you really see, you mentioned customer journey, but it can also be production of certain goods and then distribute it and cashing out the money for it and all sorts of things. What I feel like is then it's leading towards is a huge predictive capability on top of that. Because suddenly you see all different parts of specific process and it's not just purchased to pay. It's really the whole path from the moment you interact with your vendors to create contracts, you request some quotations by and so on, so forth. All the way to the end when you are getting your money or you are actually paying them, sorry. And now with the big boom of generative AI and creating the data sets, I feel like all of this leading towards one common goal and it's to say, what happens if and give me a predictions towards what's going to happen in the future based on all this data we have compiled in one big process. Would you agree with this or how do you see this evolve all the years? - The whole OCPM is one large language model. So just giving that to a algorithm that can deal with large language model, it's very easy just to ask the question, what if? So I'm not entirely sure what the conclusion of the graduate student is yet, because I did not read the end report yet, but I think this is really the possibility because you can connect more dependencies in your model, you can actually be far more precise. What influences what in those old change and connections? What I'm curious about is the way we operationalize this because what we're seeing at the event log stage now is that we're lag behind in understanding why it's important just to log events in the online environment. We know it's as a normal thing and we do behavioral targeting. We do, well, we generate the next page almost based on your profile. So you get a specific price just for you. So the idea here is that you could do the same in process as so you can automatically generate the right workflow. You can, well, correct things will prevent things before they even happen, but to get there, the data actually needs to be very good. And there I think still the challenge is is like go, how do we get traditional organizations at the point that they start logging data in such a way that is better or easily interpreted? Just by having the objects does not mean you have the activities that are influencing those objects. And that's something I'm noticing in general. We are very good at thinking about these things, designing those things. So designing the solution is one thing, but implementing these across multiple departments also means, for example, that those systems that are generating those activities have to be linked in some way. So it's not that you need to have one K-CIT, but the activities that are happening need to be linked to an object is that impossible? Is that activity in that system even related to the object that's actually a part of another system? So I think there will be a lot of efforts still needed to get the data suited to actually work. But if we can show this in production environments, where we can show it in order to cash environments, what the actual impact is there, then we can learn from that and say, okay, how is that build up? Because I'm also just free thinking about graph databases. Like they have this whole node network of all the objects in a system environment. If you just have that already, and those graph databases also include the verbs, like the influence those objects, you're almost there. - Yeah, that's like 75% of the recipe, really? - Yeah, yeah. But then you need to have all kinds of graph databases that we do not have yet. - Right. - Because we have those traditional relational databases, or we are not able to think in those terms in general like let's implement all our workflow systems in that way. I think that will be the most challenging part still. - Yeah, and at least from what I've saw so far, it's also going to be extremely challenging for a human being to even analyze these processes with all these activities in intersection, some as we really dissect 50 cases or something, and then check the possibilities, the shortcuts, the alterants and so on. It's just going to be incredible to go through. - I really like the simple example that Will from the house used to explain it to us at Salos Faire was the last year. He said, like, think of you having a agenda, me having an agenda, a Patrick have an agenda, and we are meeting in a meeting room. Like the meeting room itself also has an agenda. There are multiple meetings in this meeting room, but at a certain point in time, we three have a meeting in this meeting room, but your own agenda is fully booked with meeting meetings. But the meeting room also contains equipment. So in this case, it's in your homes, but if you have a physical meeting room, you have a screen and you have a teleconferencing equipment, that's part of the meeting room is also an object, and it is used at some point in time. So the whole objective of a OCPM is that you can follow either each individual throughout the day, but they are intersecting at that meeting. So you can also follow the day from that meeting room perspective. So who is entering, who is leaving. But you can also look at it from the equipment that is used in that meeting room. That is the whole concept is like the meeting itself, that verb having a meeting has a relationship with all the different objects surrounding that verb. - Yeah, and then imagine you have a company. So you have 1.5 million purchase orders in year 2022. Now connect it with your three million invoices. - Easy. - Yeah, so we're still learning. - Exciting future ahead, though. And Frank, to wrap up the episode, I first of all, I really enjoy the discussion here and a conversation and also to hear your insights on some of the newer topics that are currently coming up. Where can people eventually find you and get in touch with you with any question that they might have? - So usually just look on LinkedIn. Catch my LinkedIn profile if you're not connected yet. Send me a message like I listen to the podcast, I'd like to ask some follow-up questions. I think that would be the best way to link up. - Amazing, Frank, thank you very, very much for attending our podcast. It's been a pleasure and I wish you and your company, Rubblebunk, all the best with process mining and also getting your head rubbed around the OCPM. - Well, thank you for having me. It was a pleasure. - And also thank you, dear listeners, for listening to our show. So if you have any questions, you can always send us an email on mining your business podcast at gmail.com. We are also very active on LinkedIn so you can find us there as well. Thank you for listening and we will be talking to you in two weeks of time with another episode of mining your business podcast. Bye bye. (upbeat music) [BLANK_AUDIO]