Using Process Mining to Generate Accurate and Interactive Business Process Maps

The quality of today’s digital maps is very high. This allows for new functionality as illustrated by modern car navigation systems (e.g., TomTom, Garmin, etc.), Google maps, Google Street View, Mashups using geo-tagging (e.g., Panoramio, HousingMaps, etc.), etc. People can seamlessly zoom in and out using the interactive maps in such systems. Moreover, all kinds of information can be projected on these interactive maps (e.g., traffic jams, four-bedroom apartments for sale, etc.). Process models can be seen as the “maps” describing the operational processes of organizations. Unfortunately, accurate and interactive process maps are typically missing when it comes to business process management. Either there are no good maps or the maps are static or outdated. Therefore, we propose to automatically generate business process maps using process mining techniques. By doing this, there is a close connection between these maps and the actual behavior recorded in event logs. This will allow for high-quality process models showing what really happened. Moreover, this will also allow for the projection of dynamic information, e.g., the “traffic jams” in business processes. In fact, the combination of accurate maps, historic information, and information about current process instances, allows for prediction and recommendation. For example, just like TomTom can predict the arrival time at a particular location, process mining techniques can be used to predict when a process instance will finish. 1 The Need for Accurate and Interactive Business Process Maps Process models are vital for the design, analysis, and implementation of information systems. Their role is similar to the role of maps for navigation systems, mashups, etc. For example, people increasingly rely on the devices of TomTom and other vendors and find it useful to get directions to go from A to B, know the expected arrival time, learn about traffic jams on the planned route, and be able to view maps that can be customized in various ways (zoom-in/zoomout, show fuel stations, speed limits, etc.). Maps do not only play an important role in car navigation, but are also crucial for all kinds of innovative information services. Figure 1 shows two examples combining cartographic information Fig. 1. The role of maps in Funda (top left) and TomTom HD Traffic (bottom right). Funda dynamically shows houses for sale in a particular area (in this case town of Hapert) meeting specific criteria (cf. www.funda.nl). TomTom HD Traffic is calculating the best route based on cell phone information provided by Vodafone, i.e., the locations and directions of cell phones are used to predict traffic jams (cf. www.tomtom.com). Both examples use a combination of high-quality maps augmented with dynamic information allowing for seamlessly zooming in and out. This paper advocates the development of such functionality for business information systems. with dynamically changing data. However, when looking at business processes, such information is typically lacking. Good and accurate “maps” of business processes are often missing and, if they exist, they tend to be restrictive and provide little information. For example, very few information systems are able to predict when a case will complete. Therefore, we advocate more TomTom-like functionality for business process management, coined “TomTom4BPM” in [2]. Besides navigation systems, there are many applications based on Google maps. For example, real-estate agencies dynamically projecting information on maps, etc. A key element is the availability of high-quality maps. The early navigation systems were using very course maps that were often outdated, thus limiting their applicability. A similar situation can be seen when looking at information systems based on incorrect or outdated process models. In this paper, we advocate the use of accurate and interactive business process maps obtained through process mining. The goal is to provide a better breed of Business Process Management Systems (BPMSs) [1, 15, 29]. BPMSs are used to manage and execute operational processes involving people, applications, and/or information sources on the basis of process models. These systems can be seen as the next generation of workflow technology offering more support for analysis. Despite significant advances in the last decade, the functionality of today’s BPMSs leaves much to be desired. This becomes evident when comparing such systems with the latest car navigation systems of TomTom or innovative applications based on Google maps. Some examples of functionality provided by TomTom and/or Google maps that are generally missing in contemporary BPMSs are: – In today’s organizations often a good process map is missing. Process models are not present, incorrect, or outdated. Sometimes process models are used to directly configure the BPMS. However, in most situations there is not an explicit process model as the process is fragmented and hidden inside legacy code, the configuration of ERP systems, and in the minds of people. – If process models exist in an explicit form, their quality typically leaves much to be desired. Especially when a process model is not used for enactment and is only used for documentation and communication, it tends to present a “PowerPoint reality”. Road maps are typically of much higher quality and use intuitive colors and shapes of varying sizes, e.g., highways are emphasized by thick colorful lines and dirt roads are not shown or shown using thin dark lines. In process models, all activities tend to have the same size and color and it is difficult to distinguish the main process flow from the less traveled process paths. – Most process modeling languages have a static decomposition mechanism (e.g., nested subprocesses). However, what is needed are controls allowing users to zoom in or zoom out seamlessly like in a navigation system or Google maps. Note that, while zooming out, insignificant things are either left out or dynamically clustered into aggregate shapes (e.g., streets and suburbs amalgamate into cities). Process models should not be static but allow for various (context dependent) views. – Sometimes process models are used for enactment. However, such “process maps” are often trying to “control” the user. When using a car navigation system, the driver is always in control, i.e., the road map (or TomTom) is not trying to “control” the user. The goal of a BPMS should be to provide directions and guidance rather than enforcing a particular route. – A navigation system continuously shows a clear overview of the current situation (i.e., location and speed). Moreover, traffic information is given, showing potential problems and delays. This information is typically missing in a BPMS. Even if the BPMS provides a management dashboard, TomTom-like features such as traffic information and current location are typically not shown in an intuitive manner. – A TomTom system continuously recalculates the route, i.e., the recommended route is not fixed and changed based on the actions of the driver and contextual information (e.g. traffic jams). Moreover, at any point in time the navigation system is showing the estimated arrival time. Existing BPMSs are not showing this information and do not recalculate the optimal process based on new information. The above list of examples illustrates desirable functionality that is currently missing in commercial BPMSs. Fortunately, recent breakthroughs in process mining may assist in realizing highly innovative features that are based on high-quality business process maps tightly connected to historic information collected in the form of event logs. In the remainder of this paper, we first briefly introduce the concept process mining in Section 2. Section 3 introduces the ProM framework that aims at the generation of accurate and interactive business process maps obtained through process mining. Based on ProM and process mining it is possible to provide TomTom-like functionality as discussed in Section 4. One particular example of such innovative functionality is “case prediction” as described in Section 5. Pointers to related work on process mining are given in Section 6. Section 7 concludes the paper.