The World of Process Simulation with John Hill, Silico

Welcome to the Mining Your Business Podcast, a show all about process mining data science and advanced business analytics, turning me as always is my co-host and my colleague, Jakob. How you doing, man? Patrick, as usual, I'm doing fantastic. Great to hear. Today, we're going to be talking about simulating your process through a digital twin. Have no clue what that means? Well, neither do I. But luckily, we have John Hill, CEO and co-founder of Silico on our show to talk about what process simulation is, what it's good for, and how it helps you drive digital transformation. Let's get simulating. Digital transformation, AI, machine learning, buzzwords, we keep hearing more and more often every single day, and another two buzzwords, simulation and digital twin. Now, the question is, are they really just buzzwords or is there more to them? To answer us all about it, we have here with us, John Hill, founder of a business process simulation company called Silico. John, welcome to our podcast. Thanks for having me, guys. Yeah, pleasure to be here. You know, John, Patrick always loves to say that he needs to explain the stuff so that his mother, who, by the way, is a huge fan of our podcast, understands it. So to warm you up a little bit, I have a question just for you. How do you explain what business process simulation is to your friends or relatives? Yeah, when I'm talking about it with my friends and relatives, which I try not to, and they still don't invite the conversation, you know, I think most people kind of have a concept in their head of what simulation is, right? We understand it's this idea of imitating something in the real world. So it's easy normally to think of this in terms of physical things, so, you know, the classic example in the digital twin world is you might have a wind turbine somewhere offshore, and you have a simulation of that. You have a digital twin of that running on a computer somewhere on land, constantly giving you information about how the wind turbines performing and perhaps helping you think about things like when I need to go and do maintenance on it. You can take that exact same concept, that digital twin abstraction, and you can apply it to non-physical systems. So there are lots of non-physical systems in the world, we're on a process podcast, of course, so this is going to be very centered around business processes, but there are a whole bunch of non-physical processes in the world, and we can apply the exact same abstraction. We can create a mathematical model, a simulation of a business process, and we can use that to poke and prod it and experiment with it and do all sorts of interesting things which are important for businesses. John, what I found quite interesting is that you got into this world of business process simulation through working as an economist in a bank of England. What led you on a path of not only business process simulation, but also basically a business owner and an entrepreneur? Yeah, you've outted me as a complete outsider to the world of business. We are all outsiders, don't worry. So yeah, my first life, as you rightly say, I was an economist, I was working on studying and designing policy in the banking system, I studied my career in 2012, so sort of still in the wake of the global financial crisis, and as I started to try and study and think about systems and particularly the banking system, I got drawn into a research field and a interdisciplinary world of complex systems, a generic class of mathematical systems that exist in the world that we see time and time again, which are complex and really hard to study. And so there are a bunch of really useful techniques for studying complex systems, and I toured across the mall and tried to try to get my head around what the different tools and techniques were out there. And simulation was the one that just absolutely grabbed me as a, as originally a social scientist. You know, I was always envious of real scientists, so you can go and do lovely experiments in a laboratory and everything's clean and set up and you can control just about everything. As a social scientist, if you want to change, you know, the topical thing right now, you want to change interest rates, really what you have to do is sort of throw it into the world and hope for the best and make sure that something good, you know, hope that something good happens. And there are obviously massive consequences when you do that. And simulation to me was sort of the closest you could get as a social scientist to having a laboratory, having a pet tree dish, you could put things into it and put them and say, "Hey, before I launch this policy into the world, let me have a go in this mathematical model and let me check that what I'm doing and my reasons for doing it are sensible." So, is that an example of a complex system or what type of complex systems exist and why is it so difficult to study? Yeah, complex systems are everywhere. They can be really benign. They can be things that look on the surface really, really easy. And, you know, the, I suppose the ahah moment for me, we're getting to founding the company as well. The moment for me in business processes was, oh, wow, these are complex systems that everyone's here desperately trying to manage these things and they're running into the same difficulties with how these systems behave, that I see in all sorts of other fields you see in population models in biology, you see in epidemiology, which was obviously a very fashionable branch of complex systems a couple of years ago, and I saw them myself in economics and in the banking system. So, like, you know, just to give you a really, really simple example of where one might show up in your life, you know, there are systems which are not complex, which are nicely behaved and by that we mean their statistical properties are nicely behaved and there are things which behave horribly. So, like if you go to the supermarket and there are a few people on the checkout, when that system's not a capacity, you know, you can add another 10 customers and maybe the waiting time increases by, you know, 30 seconds or so. When it's at capacity, it suddenly tips into a completely different, what we call regime or phase. Suddenly, I had another 10 people on top of those first ten and it's another five minutes to wait. So, statistically, I go, well, I added 10 people and it was a 30 second increase in wait, but now I've added another 10, there's a five minute increase in a wait, what's going on here? And you see these everywhere. John, what I found interesting is that you basically started a whole company around doing this business process simulation and you know, from an entrepreneurial perspective, usually you start a business around where you see a need or somehow a niche in the market, you're trying to fill in this hole. Where you also think, wow, these business processes are super interesting. I better start the company around simulating them before anyone else does or how did this whole idea evolve in your head to go from an economist to start your own company in simulating them? Yeah, I did this the wrong way around in some level as an entrepreneur. So I started, I actually, after my life in economics, I worked for another startup, I worked for a very deep technology, large scale simulation company and we were used, you know, platform we built was used for training algorithms in equity trading, you know, working with banks and hedge funds and those types of customers. And myself and my co-founder Chris had the idea of just making a really easy to use simulation platform. I won't claim that we had any grand of vision than that. We were like, this could just be so much easier, it could be no code, it could be drag and drop, it could be really straightforward to build these really complex models. And then we didn't really know what the problem we were solving for was other than here's a really nice piece of technological innovation and let's find out what it's for in the real world. It was our customers naturally, it was our sort of customer development in those early days, you know, as two employees, as three, four, five employees showing it to as many people as we could and eventually we went through an innovation program at Vodafone and that was where our sort of our first contact with the process world happened. So a head of digital transformation looked at what we were building and we were really you know, four people, maybe five at that point in time, very young product, very, very rough around the edges. And he said, you know, I do loads of this thing called process mining, which I hadn't really heard of, which I know a lot more about. We do lots of, of BPMN, we have lots of maps of business processes, we do automation with, you know, UI path and, and RPA and my dream has always been to have an end, you know, a digital twin of an end to end process. So I can use it to think about prioritizing my transformation initiatives to rather than sort of like sitting around a table and saying, which of these ten ideas do we pick and which one do we do first and why, we can simulate them all out on the fly and, and make a really well informed choice that way. And then when I go to hand this over to, to operate, you know, to operationalize that process, how do I continuously monitor and ensure that that process is fit for person when things happen, when things change in the world like, we get a whole bunch more orders through, just the process fall over, do I need to hire more people? Do I need to go back and do another transformation cycle? So you went from the, the equities trading, yeah, sort of like platform to, then switching to the process simulation is same company or like other company. So originally it was, I was at a different company building and we did very, very, very, very complex simulation stuff, big, big computers. And then we, we set out with the vision of doing something completely different myself from Chris when we start, what has become silico? Let's just make this really, really easy to use. And then we tried all sorts of different things, you know, we had early, we did some early work around simulating COVID and patient flows through hospitals. We did some early stuff with, with the marketing team of an insurer about how best to kind of allocate marketing budget across different channels. And it was eventually, you know, this moment really I think, you know, vote of foam where they went and we went, aha, this is almost a perfect fit for this whole class of problem I didn't know exists, which every organization on the planet has. Yeah. John, could you tell us then, how does this conceptually even work? You were mentioning process mining, you were mentioning business process maps, all sorts of things as an input into simulation. Could you sort of walk us through what is then, let's say, the product of simulation and how does this really work? Yeah. So I think about, you know, I need two things to be able to simulate any system back to the sort of first principles of this. I need to know the structure of the system I'm trying to simulate. And this is actually the really hard bit as a simulation guy. This is actually the really hard bit knowing what affects what and what the, what the causal structure is mathematically, what happens first and then what happens last all the way across a complex system. And then I need to know some of the key numbers, right? I need to know what affects how that system and its numbers are going to evolve through time. And the really interesting thing when we started working in business processes was, wow, these guys all know the structures. There are these tools and there are these technologies out there. They've got the process maps. And when you look at, so with BPMN, the process map is the perfect starting point for a simulation. I now just need to do the numbers. And then when we started working with customers who had already done process mining, it was, wow, this thing actually extracts all the key numbers for me as well. It's almost, you know, it's 24 hours to go from having mind a business process to being able to do some really advanced simulation on it and show some really interesting, onto some really interesting business questions that people have been kind of scratching their head around for years. So I mean, and in process mining, we talk a lot about the event log, like this history of events for specific objects, is that kind of the thing that you need or to put into a simulation and it will kind of figure things out or I'm sure there's lots of extra things that you need to put on. But in process mining, we talk about the event log, is there something like that we talk about in simulation that that exists actually nicer than that because the, what we're trying to simulate and there's a key point about the level of abstraction we're trying to simulate a process out. So, so most of what we're doing with customers at Silico is about the aggregate process. So I'm not thinking about each individual order or each individual in a purchase requisition moving through a process. I'm thinking about, I have 500 of these a day. So what I need is I don't need the individual event logs. I need the aggregate process metrics. I need the average processing times. I need to know the branching factors across the process. I need to know what the automation rates are. I need to know the case volumes at each time step. Once I have those things, with a little bit more maths, we had a little bit more maths around this just to sort of join it all together and ensure we're understanding how things move from one end to another. You can take what is essentially the outputs of process mining, which is done all that heavy lifting, extracting the structure from the event logs, extracting those key parameters from the event log, and then move you into a world of, now we can look forward, now we can experiment, and now we can answer those really interesting questions about, right, how would we change this process or something in this process, and what is the impact of doing so? One of the key features of doing process mining is actually the ability to drill down on exactly pinpointing this count trees that are struggling vendors or specific, maybe even events that are influencing some very small part of the process, but in a radical or radical or drastic way. How do you account for these in simulation? You mentioned that you are interested in these averages, or the general process how it looks like, but I think one of the powers of this process mining is this segregation, seeing like one part of organization against the other, and when you aggregate everything into one, at least from what I would expect that this might not exactly pinpoint these core issues. Yeah, really interesting. It's a really interesting kind of fundamental question here about what are we trying to do, and what are we trying to do with process mining, and I've got no interest in building a process mining platform, I think I speak to enough customers who understand why they're doing process mining, and I think it's really about that, it's what you talk about, it's having my head of process I want to be executing against, and I can go and look in microscopic detail where I'm not executing, and therefore uncover problems in the organization. I think there's sort of broad class of reasons that our customers are starting to use simulation issues. It's probably a little bit further on in the journey, I get to a point where I've got a pretty well standardised process, and I'm not running around fixing sort of conformance issues, and I now have a second order problem, which is, is that actually the optimal process design to deliver ROI to the business? I think we're moving you from a world of interrogating in my new detail and fixing a broken process or a process which isn't executing how you want it to to a world of. Let's look at the bigger picture of this process, is it delivering value to the business? When I connect five processes together and I look at the value chain end to end, is this delivering value for the business? Is there something I could do, changing staff, changing the structure of the process, fundamentally, bringing in more automation, and what's the ROI, and should I do it? When we look at processes, specifically, when we look at end to end, or we try to figure out the end to end process from an RFQ all the way to accounts receivable or something all the way to the end, the more activities and things that we talk about, the more varied the variance become, and it just keeps rising in number exponentially. If you talk about we're looking at five processes on a higher level, do you see the same type of explosion of different types of events that can occur, or simulations that fall out, or am I looking at this in a different way? Right. If what you're interested in is the variance in the process is in the, I'm a big believer in an 80/20 rule here, and I'm a big believer in that in life. I think with processes, what we see, and I'm only speaking from my personal experience and the customers that we're working with, we see customers who are already at a point where they're getting, and at least 80% of the cases through a process, in a way that's pretty consistent and that's pretty well understood, and they're broadly happy with the execution of that process. The next set of questions they have is, how do I make this deliver more value to the business? How do I improve the customer experience? How do I future prove this against some order growth that I know is coming down the line or some structural change in the business that I know is coming down the line? How do I, now I've discovered this process and I'm trying to automate it. How do I make sure I'm putting the automation in places which are actually going to deliver value, end-to-end, not just push a problem from one part of a process into another part of a process? So, I think the use cases for simulation are quite different. I think it's very complement, I very much see it as for a process mining customer, I see very much as a next step, most of our customers who have done process mining, and we have some who haven't, and we can talk about that as well, but typically where we start is a process which is pretty mature, they started their mining journey a few years ago, they're pretty happy with the conformance, and now they're into this next set of questions that, how do I actually improve it? How do I deliver strategic value back to the business from that process? Hearing all of this actually makes me think that it could be a pretty complementary way of working on improving your processes, because what we typically do in process mining, we do this root cause analysis, we are looking at the issues with, let's say, the biggest impact on the process in terms of, let's say, money or time wasted, and then we go on fixing them, but fixing them for us really means more of mitigating them, or making sure that they don't occur, whether it's some light payments on unbuilt orders and so on, so on so forth, and what I'm hearing from you is that you could actually go a little step beyond, so not only look at what is it causing historically, but if you are fixing that what could be the ultimate, let's say, a domino effect on the whole thing? Exactly right, so a) in the future, so what's the future impact of changing this, right? Let's say we've got all to grow three years in the future, how's this process going to be performing in three years time when we put the numbers through it there? And exactly right, and again, back to my complex systems background, those domino effects aren't linear, this isn't dominoes, this isn't one thing knocks over, another knocks over, another one small thing might knock over, and cause a huge problem somewhere else in the process, you might see on your dashboard when you do your root cause analysis, this is flashing red, this is the most problematic thing, and you fix it, and guess what, the problem just goes somewhere downstream, maybe it's another team, but it's still in the organization, it's just gone somewhere else, and unless you are able to do this, have a full picture of consequences of your actions, it's really hard to say with a straight face, this is going to deliver this return to the business, and that's archiving ultimately, we want to be doing transformation, if you want to be improving processes, you need to have a really good argument for why you're changing something. To move on a little bit, I was wondering you were also mentioning digital to win at the beginning. What I always thought that is one, you create this process in process mining, and you see the events and you see how the process unfolds almost in real time, or let's say on daily basis, which is pretty standard these days, I always thought this already is the digital to win, and I guess you're going to tell me, no, that's not what I mean by digital to win, is it? Spoton, you've preempted my question, yeah, look, I think there's a missing piece, that is what I would call, and one of our customers said this, they said I have a qualitative digital twin, I know what's going on, I can see it in real time, I have a qualitative digital to have a picture that's up to date, but what I can't do is I can't look at the behaviour of it forward in time, in an up to date way, in a way that is continuously updated, so if I take my wind turbine example, it's great to know what the temperature is and how quick the rotors are going around, I want to do something predictive on it, I want to say when should I do maintenance on this wind turbine, and it's the same in processes, right? It's great to know what's going on today, but my question is I want to be able to do simulations and make sure that if I need to take, do I need to take an action today to prevent something happening in the future? That actually brings me to the first question from our audience, where Julian Lebhats was actually asking, how do different types of simulations step up against each other in terms of feasibility usefulness and precision, and he also has an example, for instance, simulating work-centric capacity changes, there's those simulating process sequence changes, and there are also many, many other ways of doing these things, and yeah, how would you actually react on that, because this already goes into this core of how you even create a digital twin? Yeah, it all goes to the core of what we've built and why we've made the trade-offs, and build the functionality that we have, and when we started working with our first customers, our first customers from the world of business processes, our platform was very well geared up around a particular mathematical approach to simulating systems, which deals very nicely as I can have alluded to earlier to this aggregate concepts. For the people who remember their high school maths, Silica treats the world as a system of first-order differential equations, all that means is I know the rates of change of things, and I know what we call the state variables, so the classic way of explaining this in school is, I have state variables, these are backlogs typically, which are like a bath tub, and I have some stuff that flows into them, out of the tap, I have inflows, and I have outflows, and the outflows would be the plug hole would be the water draining out of the bath tub, and so to understand how full the bath is at any one point in time, all I have to know is what's going in and what's going out, and that will tell me how that bath tub is changing. Under the hood, that's what Silica does mathematically, that's how we treat the world mathematically, and you can represent any business process in that way, and then that's really great because it means I can change any of the numbers anywhere across the system in any way, so dynamically I can change any number that influences those rates of change. And let's say it's a manual process, so I've got some staff, and I know the average processing time, I can work out what the rates have changed if that backlog is based on those numbers. So you can see what comes out tumbling out of this bathtub by the end of, right, so if I were to think about it, if I said, okay, I am going to increase the number of this one change that I have in my process, what is the outflow of my bathtub going to look like in terms of value, in terms of throughput or something like that? And when I stitched 100 bath tubs all together across a process, the water is flowing from one end to the other, right, from top to bottom, and it's branching off into other bath tubs, and it's a pretty crazy setup in terms of the bath tubs, but that's what we're doing, is we're putting water in at the top, and we're showing you how it's going to move through that system and eventually come out the other end, and we can then change, we can turn on or off any of those taps, we can stop up the holes in the bathtub, and we can show you what the impact of that's going to be on every other bath tub and the number at the end that you care about, the water flowing out the bottom. So that's like really important in terms of changing the numbers, so to bring this back to something more real than the kind of water and baths, that might be, you're experimenting with what if I could change the processing time, or what if I change the number of cases coming in at the top, if I change the top tap, what if I change the number of staff working on various stages across that process. So that's numbers, and then the other big thing that we built, when we came into the world of processes, is we realized in transformation, people were also trying to restructure the thing. So they were saying, well, what if I take these three steps out entirely, or what if I create a big bypass, you know, I automate something as a big bypass, which goes from one part of the process somewhere further down the stream. And so the other like sort of core functionality that we built in silico is the ability to create n arbitrary variants of parts of the process or of the entire process. And then what we effectively do is just simulate them all in parallel. So I can say, this is with structure A, this is with structure B, this is with structure C. And if I'm trying to think about 10 different transformations that I've got in my head and I want to evaluate them all, some of them, each of them will be a combination of changing some of the numbers in the system and changing the structure of the system. So what we can do is, obviously with a computer, as we can say, we can run those all inside a second each of those and show you all the numbers on a graph. And this is the impact it's going to have on this metric, that metric, and any other metric you care to compute at the end of this thing as well. Right. So you've mentioned transformation twice now. And I'm wondering in terms of the simulation, the clients that you speak to, what are some of the more common things that clients come to you with to try it for you to solve their problem. Because I know for it in process, mining it's, I want to reduce the amount of invoices that are paid late. I want transparency of my business or something like that. That's very common things that you hear. Yeah. Yeah. All sorts of different things. You know, I think we're seeing some patterns. You know, we've got, we've got one really common one that's coming up sort of repeatedly at the moment, which is about automation design. So I think lots of customers have got a process. They've discovered, they kind of know what's going on in there. They've got excited about maybe doing some RPA. And the first question they ask is, if I could it here, what's the impact on the business metric? I'm actually trying to drive. And I'm not trying to drive some intermediate process metric. I'm always trying to drive, you know, deliver faster, happier customers. I'm trying to, I've got a process and there's revenue stuck in it. I want to get hold of that revenue. If I automate this part in this way, does it actually, does it actually deliver me the ROI? So there's, there's a lot of automation design and automation optimization is a big use case that we're seeing a lot of. A really nice one that we're seeing quite a bit at the moment as well as customers who've got some sort of threshold that they're trying to avoid breaching. And that might be a regulatory compliance threshold, you know, a time to deliver something to a customer, which brings penalties or triggers or just internal SLAs that are breached. And what a customer wants to do is it's a bit more, a bit more likely digital twin example is I always want to be looking ahead and making sure I'm never going to be breaching that. And if I can see, I'm on a trajectory to breach that in six months time, what corrective action can I take today? Do I need to hire some more people? Do I need to change some rules internally in processes to ensure I can just like, you know, maybe stages below whatever that that threshold is? And then, you know, I think the sort of the classic one is also any process, which has been mined with someone's got a really mature business process. People are looking at, hey, if I can simulate this, I can answer just a whole bunch of interesting questions, transformational questions, managing it operationally. And you will hear time and time again is people talking about just trying to drive more value out of their investment in process mining. They've done all this work, done a ton of work to get a whole of the event logs and clean them and do process mining and operationalize it and making it work. And what they're seeing is, wow, it's like a few days additional work to go from there to being able to do all of this nice forward looking, predictive simulation analysis and really show ultimately like my business, my business sponsors, I can deliver strategic value with business processes, not just come to you and say, hey, these are all broken and I need two years to fix them, but also, and hey, when I fix them, this is what I'm going to be able to do. But this is how I'm going to have happier customers. This is how I'm going to affect the, you know, the bottom line. This is how I'm going to make us more compliant and improve our reputation. Jonah, apart from the fact that you almost gave me a heart attack when you were talking about those best steps because it reminded me of the times when I was studying robotics and these PID controllers and everything. And the simulation part was always horrific and I, yeah, no good memories of that whatsoever. I was wondering if you ever had predicted things that were maybe not really going in the way that you would think they would, you gave us example at the beginning with the cashier where you have first 10 people and it takes 30 seconds. And then next 10 people and the process is impacted 20 fold by raising it to five minutes. So that would be the first question and as a follow up is how do you even gain a trust in these decisions because what we are talking here is prediction and prediction is nothing more than some statistical, statistical probability of something happening based on the inputs that you have. So how do you gain the trust of these insights by the business owners? Yeah, the trust, trust question is really good. I'll answer that one first and then remind me of the second part because the, you know, there's clearly a, there's clearly a philosophical answer to this question and there's a what goes on in reality. So philosophically, we are all making predictions. We are all simulating already, you know, we are already saying I'm going to do this improvement and it's going to deliver this benefit to the business. And at the moment, what goes on is people do that quite implicitly, they may be doing the head, they may be arguing over a PowerPoint deck, maybe they do some math on the back of a spreadsheet, you know, and this is kind of typically what our customers are doing today. That's, that's the bar that we really need to, to raise, I think. You know, we've done, you know, one of our customers wanted to do a fairly sophisticated back testing exercise, a statistical exercise where we looked at some historic data and we simulated as if we hadn't gotten the historical data and then we looked at it and we showed that there was a very good statistical fit. It followed what the, what we call stylized facts, all of the sort of trends in, in, in the, in the data and was a very faithful representation of their business. The reason as a simulation modeler, you know, you, you can have real trust in what's going on in particularly business processes is we've got no uncertainty over the structure and normally when you do modeling, there is a ton of uncertainty over the structure. This is a non philosophical kind of ones I won't open here. We've got a ton of, we've got a ton of trust in the structure already. Everyone can look at the simulation and they can see it and it looks just like the graph you've extracted from process mining, it looks like a BPMM map. We can all agree that this is the same process and the numbers, the key ones which are driving the system through time are either extracted from process mining. So, so these are data driven numbers, we know they're real or we've added them into the system but their numbers like FTE and how, how many hours a person works in a day, which are numbers we've, we've got a pretty good hand long. So, even though it looks quite complex and there's a lot of maths going on, there's very little uncertainty introduced into the model because structure is certain and we've all agreed on it and we can all talk about it and when you go into any individual step, the numbers driving the change through time are either a number that we've mined and therefore we, we have trusted it because it's come from the data or a number we've added and that's typically something like an FTE number which is, yeah, pretty straightforward, there's not a lot of arguing about those. If we say this is the average processing time in this step, this is how many people I have working on that step and this is roughly how many hours we think they work a day. Despite default is the number that comes out of that bathtub on a given day, that is what gets processed. And my, the question that led to this was, what do you ever saw an outcome of the situation that would be, you would be scratching your head, I would be thinking, why is it doing this if I would actually expect the other outcome? An example would be, you fix a part in your process or you simulate fixing a part in the process and it causes, you know, even further issue down the line. I mean, I guess, I guess the thing I should, should be potentially, maybe alarming for some of the listeners as we see these all the time, when you, when you, when you, when you map these things out and you, you look at a change that looks like a very intuitive change to, to make. And I think often people are looking at kind of heat maps of, kind of this is what's screaming out as the problem. This is where people are complaining about having too much work and this is where processing times are really, you know, are really long, often when you simulate out what, improving that does, it has no impact at the end of the process whatsoever because of this thing, because the domino tips over and it pushes the problems somewhere else. So I've got countless examples of where this is true, of where there's, there's a bottleneck which you've identified, maybe from process mining. And in isolation, when you don't think of the system as a whole, in isolation, it makes a ton of sense to remove that bottleneck. But when you consider the system as a whole, it probably doesn't. Or what it will show you and a very common thing that we see is, you should fix something else that isn't screaming first and then when you unlock that bottleneck, it will be able to flow all the way through to the end of the process. So there's actually a, not necessarily, you're not necessarily doing the wrong thing, you're going to fix A and then you're going to fix B, but actually the right sequencing to unlock the business value, fix B first and then A. And that will actually have a bigger impact on the business. When you extend that, obviously, across a more complex process, it, it, it, it just blows up and, and the other one, we see all the time is, you know, I, my view from what I've seen in the industry is a lot of automation designers is based on a very similar principle is, well, this seems to be where all the manual work is. So this is, this is what we should automate. And again, almost always, what it does is, you've just, you know, you've just added a six lane highway onto a tiny little, you know, country road and it can't, it can't deal with the extra traffic. And you've got to think about what are the, what are the future changes I need to make and therefore what's the right sequencing of events? That's actually really interesting because that, let me to the question like, how do you do validation when you work with a client, right? You do internal validation to make sure that the numbers on your end make sense. And then when giving it to the client, like, how do they validate that what the simulation is and what is telling them is accurate because for us, the validation is fairly easy. We can just open the documents and look at the history and look, there you go. That's, that's what we see, but for validating simulation work, that's got to be a little bit more complex. Yeah. And again, it comes back, I think it's a really interesting philosophical question of, of what are you, what are you trying to validate? Because there's an interesting point now, you have to be mentioned prediction and prediction is a, is a, is an abused word, right? There's two types of prediction, there's a point estimate prediction, right? In, in economics and finance, I might be really interested in doing that point estimate prediction. What will inflation be in two years time? Whereas roughly where is inflation going to be in two years time and what are the sorts of corrective actions I need to take would be two totally different questions. And in business processes, I think what we're looking at is more, what is the direction of travel? What is the broad change that needs to be made in this process, rather than in two years time, did I, did I get the right number of orders coming out of the process in a given day? It's like, it's not interesting. Is that backlog 100,000 or 90,000, isn't as interesting as it's too big and something needs to be done to fix it? So, so what we, what we don't find and what we, we don't, we don't encourage our customers to do is think about this as a, as a crystal ball doing mathematical point prediction. This is about a decision, an aid to decision making about where best to target your efforts in a process, which is backed up by, you know, numbers are just a rigorous way of thinking. The maths is there to provide rigor to your thinking. I've considered the structure of the process. I've considered the dynamics of the process, how it changes through time. And this is why I'm going to take the, the action I'm going to take. And this is what I expect the business outcome to be, rather than, you know, hey, I've nailed the three year ahead forecast for this number, which I don't think any of us are interested in, frankly. And, you know, like I said, we have had a couple of customers who have wanted to do a statistical validation exercise. It can be done. You can validate these, these models very well statistically, again, having built models in, of other flavors in the past, I've built a lot of statistical models in, in, in my previous life. These things are much, much easier to validate. For the point that I made earlier about structure, structure is known, that is the right part to validate. It's really lovely. You can get everyone in a room, you can get the people working on a process in the room, and they can all agree on the mathematical structure. And then you might have an argument about a couple of numbers, but that's the easy part. Yeah. Well, I have to say, I'm starting to doubt a little bit on how I conduct the, the value creation with process mining and focusing on these biggest items in the process and fixing them. And then, oh, well, that actually goes as other problems elsewhere. However, I wanted to ask also about the, the big topic and this, the generative AI. And we just had a guest recently, actually, last episode, where we were discussing how, if you basically feed this generative AI with event logs, you will probably design much better processes. So it would be sort of, I don't want to call it simulation. It would be more like creating perfect processes out of your historical data. So that's the, that's this prediction on what you should actually do. Where do you see the difference between taking this and generating the processes from scratch with just inputting your historical data and the actual simulation that you guys are doing? So I, I'm very, I'm glad you bought this topic up because we, we've done a ton of thinking into, and I've done a lot of thinking about, like the implications of generative AI in business processes and obviously for simulation as well. And, and I've got to a point, I've got to a point, and I've not really heard anyone talking about this in the ecosystem, essentially that's because I'm, I'm mad, but I, I think I have an interesting take on this, which is what generative AI is, is fantastic for, is generating. Like, we can all agree on that. And I saw this first, you know, I, I was like, everyone else, I got super excited when, you know, GPT 3.5 and then GPT 4 came out and was playing with it and playing with text. And what you realize is it, it generates fantastic looking text until you really understand the topic and then you start to see the cracks in it and you realize it's not quite as good as I thought, which led me to the conclusion that the hard part is not generating now that solved. The hard part is evaluating. Is it good or not? And so what I think is so interesting about what we're doing and the, and the, and the promise of simulation here is, if I can now generate a million different processes arbitrarily, the question is now, what are any of them any good? And the beauty of the simulation is it tells me if something is good or not. So, so what we're doing today, what transformation teams are doing today is generating solutions. How am I going to, how am I going to change this process and am I going to fix a problem. And we might generate 5 or 10, right? And that might take a long time and we're drawing them on whiteboards and we're discussing things and there's lots of work going on to generate potential solutions. I think what's really interesting and we've prostitied this and it doesn't, it can be done. You can generate process structures and you process structures with parameters using generative AI, you can feed those through a simulation platform like Silico. And it will tell you, for all of those million ideas you might want to generate, which one is going to have the most impact on the business. So, the generation parts solved, the simulation solves the evaluation part. The only hard part is to tell, is to tell the computer what is possible in the world. And the bit that I think is really going to be the really interesting bit and we've, got some, I think, very interesting ideas here is, how do you ensure that what the AI is going to generate as a solution is actually a plausible thing that could be implemented in the real world. And the way you do this, I think, is you feed it enough examples of things that have, our plausible changes in the world. I think there's a very interesting thing that we want to build that sits on top of Silico, which will eventually, instead of the human doing the generation of the solutions and the simulation doing the evaluation, we can put that all in a loop and we can have that done by the computer. I've seen some very interesting staff as well around using it to sort of help with the explainability of process models and processes as well, but I do think this thing is generating new processes, is the interesting angle and simulate, if you haven't got a simulation frame what you're stuffed, because you've got, you can generate as much text as you like, you've got no idea if it's good or not. So I mean, I've encountered this when I'm asking a chat GPT for like code examples as I was learning a new language or something and it was like, yeah, I just, you know, use these functions and then your problems are solved and then I tried using them, they don't exist, of course, they just thought that they existed and it's, you know, okay, cool, gives me a good idea of what it should be, but I guess these things don't exist, so I'm not going to do this. And same in the generating process models, I can imagine that, well, if you want to increase customer happiness, have you considered the activity increased customer happiness? Yeah, maybe, maybe, sure, easy for that. So, but there are some things that it can generate that, you know, might not exist yet, but are technically feasible. Activities that haven't been considered or automations and places for activities that don't exist and we're feeding it things that only exist in the space that we're giving it. So is that like a bit of a problem like we have very, like, that's a structural problem. So I think the thing that's hard that's yet to be solved, but I think I certainly have some ideas of how this gets solved and, you know, maybe, maybe there's a future, future of podcasts. I'll come back on and tell you, but I think there is a, there is an outstanding challenge to be solved technically around ensuring that it generates valid processes. Not just processes. The thing that the idea it has for an automation is a legitimate thing that could be automated or the change in a parameter value is something that you could, is a lever you could plausibly pull them, get to that setting in the organization, doesn't say puts, you know, put three million people on process step two, obviously, that's not the plausible setting. So I think there's a, there's a, there's a, there's a piece around, can it generate processes absolutely, can generate as many processes you want, it's fantastic at that. Can we can, can, can silica evaluate if those processes are good or not? Absolutely. It can do that. A bit in the middle is are those things that suggested are those like your code example, are those actually valid functions or those actually valid things? And I think what it needs to be fed with is a, is a set of a big rich database of plausible things that it can change, lots of examples and you'd start this, you know, what, what we'd like to do, I think it's to start this in one common wide spread process that you see in every organization, you would build up, if you were to build up a set of examples of the possible ways it could be changed. The AI would be able to generate those for you that were, and there would be plausible. But unless you've got the simulation, you're not going to be able to work out if they're good or not. Right. So it's, I think it's very, very exciting. I think there'll be something very interesting to show, you know, in months, certainly next year. Patrick, I'm not sure about you, but I love how basically every new angle of thinking about processes and this simulation being one of those is just, you know, and widening the horizons on how you want to think about these problems because there are some things I haven't thought about just it. So this is, this is very, very interesting. And John, question back to you again, you, when we first talk to you, you talked about evangelizing this process simulation as a category, and I know that some other vendors, if it's software AG or even so long as they have some capabilities on simulations, where do you see this category going? And is it just going to be a subsidiary of what process planning is doing? Or you see, you can see evolving or I expect that you do see this as a business owner, this will evolve in a whole new world, a whole new thing. So where do you see process simulation in one to three years? Yeah, I mean, it's super young. I mean, we've really been evangelizing this in 2023. You know, I think we really understood that there was a category here that was to be shaped this year. So this is a very, very young thing. It's very, very new. I process mining now is eight to ten years old, I'd imagine, in terms of how long it's been evangelized for. So we're right at the start of the cycle. I very much see it as a standalone category. And I think the bit of this that's been so interesting for me to observe is that it needs to be built on top of the data that's extracted by other platforms. And so what you really want, I think what the market needs, what customers are going to want is a process simulation platform, which is interoperable with the different process mining vendors that are out there. So rather having a world where every process mining vendor goes and builds all of those rich and sophisticated simulation functionality that we've built at Silico and finds the people in the world who have got enough expertise to actually build it in the first place. There is going to be a category in its own right. And those tools are going to play nicely with process mining. We've talked about a lot of process mining. You guys are a process mining podcast. But we also are interoperable. We have customers who haven't done process mining, who are doing this just as a next step on from BPM and we are working with partner with a couple of the task mining vendors as well. I think that's something really interesting in doing simulation at that slightly lower level at the task mining level as well. So I think this is going to be a category which, to me, sits higher in the stack than process mining and really process mining is in that layer of discovering what the process, reaching into the ERP, reaching into the enterprise systems, figuring out what the processes are doing the discovery. And then there's going to be a layer that sits on top of that, which is going to be about simulating those processes. And then eventually there's going to be a layer on top of that, which is this world of AI reaching in and using those simulations to help augment human decision making and maybe in some places start to actually make decisions autonomously as well. So I think you're absolutely right. There is some kind of quite limited simulation capability that some of the vendors have today very tightly coupled around a process mining view of the world and I think it's something bigger and wider reaching in that. Right. John, to wrap up the episode, we have, let's say, a list of six questions from John Markery. And we have, by scanning them through, I think we almost touched every single one of them. But let's imagine this being some sort of an elevator pitch of the simulation because maybe the listeners of our show will want to go for some sort of simulation and will have to defend it in front of the management. So instead of them having to dig into every answer, we've had over 52 minutes of the recording session so far, let's actually try to condense it into these six questions. So once again, John Mark, thank you for posting those and let's get into it. So first one, why I should add all simulate processes. Yeah. I think, I think you know, again, we've got a lot of this. The reason to simulate a process is when I'm trying to make a change in a complex system, when I'm trying to change a business process, I need to have a really good rationale for why I'm changing it and what the expected consequences of that change are going to be. So anywhere you are trying to improve a process and there's high value, there's high cost and there's high value in the impact of that change, it is very likely that simulation is going to help you find a better outcome and justify the decision that needs to be taken to your management or others across the organization and get broad buy in for what your recommendation is. All right. Second question, how is simulation helping me to understand the impact on process changes? So the beauty of not just simulation but actually specifically about what we do at silico, the fact that we have the full visual picture of not just the structure of the simulation, but how every number is changing across it means that you've got that full visibility and that's absolutely like that cannot be expressed clearly enough that it's so important that this is not a black box, this is not some regression model operating behind the scenes or some statistical model running in the background, which is what's in quite a lot of the process mining vendor's capability. It's something why I can see if I pour water in at the top of a process, I can see exactly how it flows through that process and exactly why the changes happen. So it's going to give you like not only a full view of the process but a full view of how that process will change through time as things inside it change. Next question, does simulation help me to prioritize opportunities? Definitely. Cool. Let's move to next one. How is simulation helping me to better understand the financial impact by these changes? Yeah, really, and there's something I haven't covered here which is really important actually. So I'm glad the questions come up. One of the things, one of the beauties of the framework that we built is that you can extend beyond the world of just processes. So because originally, remember what I talked about when I started the company, this was also about just general maths, which means we can build very rich, complex, mathematical models as well as the process simulations themselves and join those two things together and have them talk to each other. To give you a customer example of this, one of our customers had a very complex order cash process. And then there was a very complex breakdown of the contracts once they came out the other side. So once they turned, we're trying to be turned into cash, there was some very complex financial modeling about how that actually happened in the real world. A really key part of the value proposition of silico, never mind the simulation was I could just join the process view up to this complex financial model. And now when I changed the order to cash process, I can actually see the P&L impact, not just I've changed the throughput rate and I've changed the backlogs, but I've actually had an impact on the P&L. Okay, well, moving on to next one, fifth, does simulations add additional insights or dimensions I would not see with other conventional tools? Yeah, and I think again, we've covered this one extensively. It's a different realm. We're not talking about the source of insights that you're getting in process mining. This isn't about conformance, this isn't about exceptions. This is about a slightly bigger picture view. This is what are the domino effects? What are the cascading effects of changes that I might want to make to this process? What is the view of this process, which I can think about taking decisions on and measuring what the business impact of that's going to be? I think it's a slightly different realm. If you're someone who's been doing a lot of process mining, you probably need to, you know, this is something you need to consider in its own right, not as, you know, it's not the simple bolt-on on top of process mining. It's difficult for us to, you know, to just let go of things we already know. I haven't let go complex systems yet. Last question, though, how is organizational behavior over time changing the impact, meaning like is a behavior of change and input in the system? Can you, what's the question again? Can you ask anyone again? Yeah, yeah, yeah. So how is organizational behavior changing the impact on the simulation? So if suddenly the organization starts changing, is this an input that you feedback into the system? Yeah, that's right. So the boundaries of the simulation, what the edge of the system is that you want to consider is something that you choose. So typically, we're still talking about processes. We're typically stalling about a process. But we are, you know, one of the sorts of questions that comes up quite often with a customer is really big strategic stuff. Like, what if I got rid of this product line entirely? What does it do to the process? What does it do to the financials, let's say, or what if I got rid of a warehouse, or what if I got rid of a factory? What if we acquired in another business and the case for, and we added their case volumes to the current case volumes that we have? So there is another lens to this, which is actually really big sort of macro changes coming in from the outside, either into the organization or into the process itself, which again, are just that just changing numbers and changing structures. If there's one thing to take away, it's all changing numbers and changing structures. Jon, all I can say is thank you very, very much for a very pleasant talk. Maybe in a few years, we will rename our podcast from process mining and business simulations, because the new category would arise. And I can already see that there would be dozens and dozens of episodes on adopting this technology in the first place, which I still think could be very, very challenging seeing how difficult it's to adopt even process mining. That's what I'm thinking. Yeah, so Jon, thank you very much. Good luck on the journey and on evangelizing this topic. I'm sure we will hear of it again. And thank you for coming to the show. Pleasure. Also, thank you, dear listeners, for standing by with my new business podcast and listening to another episode. If you have any questions or if you would like to hear about some further topics, reach us, reach out to us on LinkedIn or write us an email at miningyourbusinesspodcast at gmail.com. And we will be happy to consider these topics and people in our show. Thank you for listening. If you like us, leave us a rating or review on our NEPL podcast or Spotify and tune in again in two weeks of time with another episode of mining your business podcast. Thank you and talk to you soon. Bye-bye.