An Introduction to MPM - MEHRWERK ProcessMining Janna Meyer Josua Reimold Constantin Wehmschulte Mehrwerk AG Mehrwerk AG Mehrwerk AG Karlsruhe Karlsruhe Karlsruhe Germany Germany Germany janna.meyer@mehrwerk-ag.de josua.reimold@mehrwerk-ag.de constantin.wehmschulte@mehrwerk-ag.de R Abstract—Qlik Sense platform based MEHRWERK Process- 2. Discussion of Innovation Mining is the first tool to offer self-service process mining in- cluding the data governance necessary for enterprise-wide use. MPM is characterized by the combination of interactive Fast implementation through various data extraction options self-service BI with lean process mining algorithms. As and integrability with machine learning functions make MPM flexibility and usability regarding analysis capabilities are a customizable tool for every use case. Basic process discovery key to successful generation of process insights, employing and conformance checking algorithms in combination with a Qlik Sense for process mining purposes is convincing. R powerful business intelligence platform allow in-depth analysis Even non-experts find answers rapidly in the appealing visu- as flexible as the investigation requirements themselves. alizations of MPM due to easily understandable associative analysis. keywords: self-service process mining, process discovery, conformance checking, variants analysis, log extraction 2.1. Qlik Sense R Platform The Qlik Sense platform provides user-friendly analy- R 1. Introduction sis based on governed data discovery. With the Qlik Asso- ciative Engine it is possible to discover relations in even vast amounts of data by technically performing a many-to-many As process mining is a highly explorative analysis full outer join in the back-end [1]. This approach enables technique, process mining software should provide powerful and interactive search, selection and filter functions interactive and visual analysis to support insights in due to a patented in-memory technology including com- complex processes. MEHRWERK ProcessMining (MPM), pressed binary indexing, logical deduction and dynamic KPI deployed on the Qlik Sense platform, is designed to R calculation and aggregation [1]. Recognizing the importance offer comprehensive analytics to the process analyst. of such capabilities for process analysis, the enhancement of MPM combines market leading self-service business Qlik Sense with process mining capabilities to create an R intelligence (self-service BI) with the insights achieved by innovative tool is the logical conclusion. As a result, analysts process mining algorithms and visualizations for process can select any dimension and every visualization, diagram, discovery and conformance checking and, thus, becomes table or metric is calculated at run-time with respect to a powerful tool for professional process analysis. Through the choosen analysis perspective and data set. A colouring Qlik Sense ’s APIs it is possible to enhance the solution R supports the comprehension of relations within the data according to the given use case’s requirements, for example, as the selected data is coloured green, the excluded dark by integrating data mining scripts from R or Python. To and the associated data light grey. The benefit for process introduce MPM properly to the interested reader we will mining is obvious: for example, to analyze the context of firstly discuss the innovation our software offers to the a given case or the influence of a resource’s involvement process mining universe by shortly explaining the platform on process delay. Another advantage provided by Qlik is and the implemented functions. Then a description of the creating ad-hoc analysis in a governed environment with tool’s maturity is presented to outline its usability and its validated KPIs and dimensions without the need of coding. adaptability to future developments. Future developments With modern drag and drop functionality process analysts are shortly mentioned to elaborate our vision and in the are able to generate new visualizations within minutes. concluding part of the paper is a link to a 25 minute Therefore, MPM is more than a simple reporting tool on introduction video. processes. By defining and sharing bookmarks and ad-hoc reports analysts communicate their insights to colleagues. mds Furthermore, they can export the visualizations and reports April 05, 2019 to present them for example to the management. To support mobile analytics, Qlik Sense ’s apps offer a responsive R the resulting directly-follows graph. Hence, the determin- design that can be used seamlessly on iOS or Android. istic MPM Process Discovery Algorithm is rather simple Offline analysis is provided likewise. but efficient. The MPM ProcessAnalyzer shows the real As business analytics is dependent on up-to-date and processes as-is and enhances them by standard metrics like high-quality data, Qlik Sense offers the import of various R process step duration, lead or idle times and further use case- data formats as well as connectors to a large number of dependent performance indicators such as automation rates. enterprise information systems. Official connectors can be Filter and selection options help reducing the spaghetti- found at [2]. Profiting from this connectivity, MPM allows diagram to relevant process variants. Through Qlik’s As- fast, tolerant and scheduled event log extraction from source sociative Engine the whole application content is then re- systems and data processing in a scalable manner. calculated for the selected data. The MPM QueryBuilder provides the useful capability to search for activity patterns 2.2. Log Extraction and Enrichment of concern thus, allowing to fastly check critical process variants and their context information. MPM provides two options of feeding event logs into the process mining algorithm. The first is to run the code 2.3.2. Conformance Checking. Another helpful function- on an existing event log that is read into Qlik Sense R ality is conformance checking. With the MPM Process- and transformed into a minimal log with information about Modeler the user is able to define a happy path via drag case ID, activity type and activity timestamps such as start and drop which is displayed in the process visualization or end of the event. The second option is to extract data to investigate process deviations. If a should-be process directly from sources, as databases, XML, xlsx or csv with model is defined, its list of distinct process variants can be the MPM RuleEngine and create the event log. Doing this, used to calculate the process conformance between reality there is no need to provide a process mining-ready event log and model. The MPM ConformanceChecking Algorithm beforehand and the import effort is significantly reduced. encompasses alignment based logic and, therefore, evalu- Advantageous in this data-directly from source scenario is ates activities and moves that are either synchronous or the opportunity to easily alter the scope. If time periods or in model or log. With these information a fitness metric activities of interest change, a few clicks lead to a new event for the most fitting happy path and each real-life process log that is automatically analyzed by our process discovery variant is calculated. Hence, the analyst can rapidly identify algorithm. Also, due to the MPM RuleEngine the event log highly deviating process variants due to deviating activities generation is implicitly and comprehensively documented. or changed activity ordering and compares process variants Through these distinct initiation possibilities, MPM has a that should, but, do not correspond to one common happy short implementation time since it adapts to the use case path. and the existing data architecture. To offer the best insights in process data, MPM con- 2.3.3. Process Analysis Perspective. Given the analytic nects the process log to context data by Qlik-typical data flexibility of Qlik Sense , analysts can use MPM to evaluate R modeling. Omitted KPIs are, if needed, fastly calculated processes from different perspectives. At first, the control and appended to the data model, often without the need to flow perspective can be taken to get an overview on the go back to the data extraction step. Thus, the enrichment general process by generating the process visualization and of event logs, process logs or data models for e.g. root calculating the process variants. The ordering of activities cause analysis purposes is rapidly accomplished given some and the deviation from a happy path get clear in the visual- knowledge in Qlik Sense formular syntax. R ization and can be further assessed by a variety of diagrams, KPIs and the alignment based conformance checking. With 2.3. Functionality the process instance as central investigation object, the case perspective is the easiest perspective to take with MPM. Through MPM, the complex task of integrating pro- Along with resource data, the use case dependent context, cess mining with modern business intelligence is achieved. e.g. supplier, material, machine or weather, can be included By defining reusable template applications the analyst is in the analysis. To cover the organizational or rather resource enabled to perform exactly the same analysis on top of perspective the analyst can employ a social network graph, different event logs – some refer to this as history mode. an extension provided by an open source extension on Qlik For example, the comparison of different time periods before Branch Garden [3]. To investigate bottlenecks, resource R and after a certain process intervention is simply achieved. utilization or service degree, the time perspective is also Since performance and quality analysis, compliance check- supported by MPM. Qlik’s standard analysis provides the ing as well as process monitoring are important areas of required functionalities when used on the event log times- process mining applications, the opportunities provided by tamps and the process indicators, which were calculated by the integration with BI that are offered by MPM are out- the process discovery algorithm. standing. Figure 1 demonstrates the seamless interaction of the Qlik selection (green, white, light and dark grey), the MPM 2.3.1. Process Discovery. Process Discovery in MPM is ProcessAnalyzer, and the MPM ConformanceChecking. The performed by fuzzy mining the event log and extracting left process visualization exhibits the most common process Figure 1. Comparing process deviations between purchasing organizations variant in company MEHRWERK UK’s purchasing group Data, MPM is prepared to confront the upcoming challenges M00. This variant is set as happy path for the right visualiza- with Qlik Sense ’s strategies. R tion to compare it with the process variants for purchasing group M03. Red moves in the right graph show where M03 3.1. Handling Big Data deviates from the process of M00 which is marked greenly. To overcome performance problems due to big data, Qlik follows different approaches like on demand app genera- 3. Description of Maturity tion and large scale architectures for horizontal or vertical scalability. On-demand apps allow the user to load and As Qlik Sense is the platform for MPM, the maturity R analyze big data sources, where aggregated, representative of the analytic capabilities is eminently high with more than visualizations are created for the whole data set to identify 50,000 customers of different sizes and industries trusting interesting data subsets on which detailed analysis is carried in Qlik Sense ’s functionality. Advantageous for any devel- R out afterwards by interactively generating an on-demand opment actions is the large community of Qlik users which app with the full Qlik in-memory capabilities [4]. With also provides open source code for extension development. respect to large scale architecture, Qlik Sense recently R The MPM functionality itself has been tested in various real- adopted Kubernetes cluster using containers [5]. Thereby, life projects receiving strong positive feedback by customers high-elastic deployments into Kubernetes clusters become with environments varying from small up to large and e.g. possible, running in either public or private clouds on cus- multiple SAP systems and heterogenous data sources. R tomer managed infrastructures [6]. Having access to this The deterministic MPM ProcessDiscovery Algorithm cre- architecture MPM can handle big data respectively. The ates a dependency graph as process visualization showing interested reader finds more information about scalability the spaghetti-like process variants. The filter and selection at Qlik’s Whitepaper [7]. functions of Qlik Sense help to reduce the process variants R to the relevant scope. The MPM ConformanceChecking 3.2. Case Studies Algorithm handles should-be process models by splitting them into the distinct process variants which are then com- One of the first projects realised with MPM was to pared to the real-life process variants. Planned but not yet rapidly identify compliance issues as well as to ensure and implemented is conformance checking that also includes the prevail corporate governance in a purchase-to-pay process should-be process step durations or process step idle times of a large German energy supplier. The audit, governance in the fitness calculation. In none of the projects where and industrial control system area of this enterprise initially MPM was implemented problems with respect to scalability comprised 1,828,864 cases and 3,036,204 events composed occurred. Nevertheless, to meet the future demands by Big by 27 activities. The data were obtained from SAP ERP R and Excel sources. With MPM the company achieved full as- Having recent developments of Qlik Sense regarding R is purchasing process transparency across multiple company chat bots and alerting functions [8] in mind, we are looking codes and purchasing organisations combined with auto- forward to real-life projects to embed these capabilities mated calculation and evaluation of risk indicators. within process mining scenarios. Hand in hand with the Another application was an order-to-cash project at a planned export of process models, MPM will not only be German raw material supplier where MPM was used to a tool to generate insights and appealing diagrams but to identify optimization potential for days of sales outstand- automatically communicate important process informations ing (DSO), order lead time, order processing cost and the directly to the responsible stakeholders. general process. 477,345 process instances with more than 4.4 million events leading to over 45,000 process variants 4. Conclusion have been analyzed. Another project by a German automotive logistics sup- MEHRWERK ProcessMining based on the Qlik Sense R plier was the optimization of their logistic process. The platform is first class self-service process mining that pro- small logistics data set came from a SAP ERP including R vides the data governance necessary for enterprise-wide use. 6,159 cases, 49,711 events which were built up of 15 activi- MPM offers fast implementation through flexible data ex- ties and formed 85 process variants. MPM generated insights traction options for a wide range of use cases with or without in optimization potential, for example, by identifying the a previously prepared event log and is easily adaptable to refueling process as bottleneck. Major achievements were specific analysis requirements due to integratability with the discovery of “forgotten” cars on parking sites and a new machine learning functions (e.g. DataRobot/Python/R). benchmark system due to understanding which transport We see application possibilities of MPM in technically system - train or truck - was the better option in certain every context where IT systems track events. Experiences situations. have been collected in P2P and O2C processes, classic logis- One project that was not based on SAP input data R tic and production processes, banking portfolio management was the analysis of a production process which’ data were and the monitoring of user-interaction with machines. MPM recorded in two different information systems – a computer- can be generally helpful for improving processes due to aided quality system and a manufacturing execution system. algorithmic process discovery and analysis. Furthermore, it During production, the process instance changed from batch works as enabler for root cause analysis. Improving auditing process to single production process. Therefore, the require- and compliance by algorithmic process comparison is an ment was to achieve an overview on the overall process application field as well as supporting process automation or by investigating the subprocesses in one single app. Hence, digital transformation by discovering opportunities and link- two event logs were extracted by the MPM RuleEngine ing strategies to operations. An interesting use case would from the source systems, the first with 6,483 batches as be the monitoring and controlling of S/4 Hana migrations. process instance, 11 activities and 51,334 events and the A 25 minutes introduction video can be found at: https://mpm-processmining.com/demo-video-mpm-2019-en/, second with 49,396 single production numbers as process password: MPMDemo2019 instances, 17 activties and 322,201 events. The two event logs were processed and led to two process logs that were References associated to one another via the batch information. The processes were visualized seperatly but in the same app so [1] QlikTech International AB, The associative difference: the analysts could clearly see the overall process and track Freedom from the limitations of query-based tools. the single production number down to batch events. The https://www.qlik.com/us/resource-library/?search=associative+ difference\&Resource+Types=-1, 01.11.2018. aim of employing MPM was to generate a starting point for [2] QlikTech International AB, Qlik Connectors R . https://www.qlik.com/ continuous process improvement. Slow process variants as us/products/qlik-connectors, 01.04.2019. well as production bottlenecks were discovered and selected [3] Qlik Branch R , Network Vis Chart. https://developer.qlik.com/garden/ for improvement activities. 56cea95aaaacff050825f6f8, 01.04.2019. [4] QlikTech International AB, Managing big data with on-demand apps. https://help.qlik.com/en-US/sense/November2018/Subsystems/Hub/ 3.3. The Vision Content/Sense Hub/DataSource/Manage-big-data.htm, 01.04.2019. [5] QlikTech International AB, Qlik Sense R Enterprise for elastic A first version of decision tree forecasting and root deployments. https://help.qlik.com/en-US/sense/February2019/ cause analysis capabilities is in testing at the moment. Subsystems/PlanningQlikSenseDeployments/Content/Sense Deployment/Deploying-Qlik-Sense-multi-cloud-Efe.htm, 01.04.2019. Planned for the future is an intuitive root cause analysis modeling functionality and stronger prediction capabilities [6] QlikTech International AB, Qlik Sense R Enterprise architec- ture and scalability. https://www.qlik.com/us/resource-library/ with deep learning algorithms for decision support and qlik-sense-enterprise-architecture-and-scalability, 2018. automated activity triggering. The enhancement of MPM’s [7] QlikTech International AB, Qlik Sense R Performance process visualization with hierarchical functionalities that Benchmark. https://www.qlik.com/us/resource-library/ allow for the aggregation of activities to a super-node is in qlik-sense-performance-benchmark, 2017. development. Aggregating and deaggregating the view on [8] QlikTech International AB, Qlik Sense R Bot - Power of the Qlik Plat- as-is processes will thereby be supported. form. https://www.youtube.com/watch?v=aoZ-LaicXHs, 18.04.2017.