Analysing simple and complex processes

Hi, and welcome to this lecture on analyzing simple and complex processes. So there’s several types of processes. And we usually classify them as lasagna or spaghetti type of processes. So spaghetti type of processes are called spaghetti because usually their connecting arcs between places and transitions are crossing each other. And it’s hard to see where they start and where they end– hence, spaghetti. It’s a big blur of arcs. On the other end of the spectrum, we have lasagna type of processes– nicely layered with a clear structure. And usually, you can just read them from left to right and understand almost all details of the process. Now, of course, these are two extremes. And there are several gradations in between.

But it’s important to note and know which type of processes there are and to quickly see what type of processes the data describe that you’re currently investigating. So a lasagna process is usually a process where, with limited effort, you can get a process model you can agree upon with others and that has a replay fitness of around 80%.

So this also means that the process execution is already structured because it’s generating a structured process in the data. So the execution is also rather structured. Unfortunately, the process is already structured and fairly well known, so any insight that you will get using process mining techniques or another technique will have relatively low impact. And typical example processes are the more administrative processes such as in finance or insurance or more bureaucratic processes where there are clear rules and flows. So usually, these are also the more high volume processes.

Then on the other end of the spectrum, we have spaghetti processes. And for spaghetti processes, it’s without extensive filtering, you cannot get a sensible process model to be discovered. And this indicates that the process execution is also less structured or formal. Analysis, however, can provide interesting insights. Because the process model is so unstructured, analyzing the data can really gain insights that the process owner can use to improve the process. And typical examples are the more service oriented processes where the end product or goal is known, but how to get there is less fixed– for instance, in health care or in high-tech machines and systems. So you can filter spaghetti to become lasagna.

So you can focus on a subset of activities, for instance. And usually, there’s a top x, for instance, top 25 or top 20, that already covers 80% of your events. So 80% of your events is usually only executed by 20 or 25 activities. You can also focus on a particular part of the process. So if you know that the process is consisting of particular phases, you can focus on a particular phase. You can also focus on a subset of the cases. So, for instance, you can filter out cases that are not of a particular type. For instance, silver customers, you can focus on them.

You can also consider the start and or completion time of a case– so, for instance, only consider cases that started in a particular year. And finally, of course, you can also cluster similar behavior together. Since traces in a particular cluster have similar behavior, discovering a process model for that cluster is easier. And finally, you can focus on a subset of the paths. So if you only consider the most frequent paths, these are usually more simple. And there’s always a few deviations that make them rather complex. For instance, in the Inductive Visual Miner, you have the sliders to help you with this filtering on the fly.

So as I said, the spaghetti and lasagna type of processes are on two ends of a spectrum. And there are many more gradations in between. And in a previous lecture, we also discussed several process mining activities that one can execute. So let’s see if we can position these activities in the spaghetti-lasagna continuum. So which activities can you execute only on lasagna processes? And which can you also do on spaghetti type of processes? So the first activity is exploration. Well, exploring using, for instance, a dotted chart, you can do on any type of data and process. So this is on the far left end.

You can do this even on the most spaghetti type of data that you see, you can still apply the dotted chart. The next activity of prediction requires that you know where and which date a particular active case is for you to predict where it will end. So this is on the more lasagna end of the spectrum. You need to know where a current case is in order to predict where it will go. And if you want to do recommendations, then you need an even better process model. You need to very carefully know where you are and what is going to happen with what likelihood in order to give good recommendations.

Then auditing, detection of breaking a rule can already be done in the more spaghetti like process as well as checking. Since these are more on the rule end of the auditing spectrum, you can do this on spaghetti type of processes. However, when you want to compare models, you need a bit more lasagna type of processes. And if you want to promote something that you found, then of course– this is close to detection and checking– you can do this on the spaghetti (!) end of the spectrum.

Finally, cartography, discovery, well of course, it depends on the algorithm that you use, but usually you need– for a good process model that makes sense– you need a more lasagna (!) type of process to be recorded. Then using the process model, of course, you can enhance the model. Then you need a slightly more lasagna type of process. And, of course, to diagnose it in a bit more detail, you need an even more lasagna type of process. So I’m not saying that this is the definitive positioning of the activities over the continuum, but it gives you an idea of what activities can be done on any data.

And these are usually activities that– for instance, exploration– you do in the beginning of the analysis. With certain activities, you really need a process model that has a high relation with the data. Hence, the model needs to be of the lasagna type.

So now, we’ve discussed many possible process mining activities and positioned them also in this lasagna and spaghetti continuum. And this concludes the last content related lecture. And the next lecture will be the closing of this third week. Thank you for watching.