Take us to the PROM
Process Mining is very similar to Data Mining and is based around the same principles. By applying analytics to large sets of data, we try to find patterns, similarities or exceptions that help us predict, manage and inform with regards to data, processes, people and transactions. Since I travel very regularly, I get phone calls quite often from American Express. They will ask me: "Sir, is it correct that you have performed credit card payments in Amsterdam, Munich, Qatar, London and Minneapolis all within 48 hours?". Apparently it looks like fraud, and they get these fraud patterns trough data mining and use the same logic to call the credit card users to prevent it.
Now take the same principle and add to it that the primary focus is the process, its flow, the chronological order of individual steps that a certain transaction type walks through based on time. Many organizations use process modeling (of course in BWise and sometimes in Visio ;)) to describe their process flows that depicts the individual steps as transactions are handled. For instance the sales process from retrieving the first phone call to, sending the quotation, receiving the order, performing credit checks and such, sending the goods, billing and finally receiving the payments. In many large organizations, these steps are performed and recorded with the help of information systems. Could we use the data in the information system to automatically generate the process models since the bulk of the process steps are in there? YES you can!: through Process Mining.
PROM takes a large set of data from an information system and analyses that data based on timestamps and other time related indicators to determine all possible unique flows of a certain process from start to finish. It then draws a visual diagram pretty similar to a process model to show the possible flows of a transactions and the number that follow each fork. Making it visible gives the viewer of the diagram more information than a typical list of possible flows and makes deriving conclusions from the model quite simple. An example to illustrate: One of our customers uses an electronic invoice receipt system. All invoices that they receive are scanned and automatically processed. Through OCR text recognition they get the creditor, vendor, amount, date and invoice number and the invoice is automatically posted in the SAP system. When posted, the large value invoices are checked by a service center in India after which the invoices are matched to received goods or services and automatically released for payment. The PROM generated model showed (quite easily since it draws a process model of the possible transaction flows) that some invoices were paid before approval, posted before being scanned and some did not have a scanned version at all. It also showed the number of invoices that did not follow the designed process which helps to determine if it is an exception or common way of processing because of some unknown reasons.
Now you might ask yourself what does this has to do with BWise. Well, besides me having a personal interest in this subject, there are clear linkages to our process modeling capabilities, our continuous controls monitoring capabilities and our IC frameworks that describe the most common risks and controls that are involved in each process step.
Think about the practical use and visit us next week on this blog to get our vision in how this al is VERY relevant! Interesting reading material:
- https://www.ngi.nl/Regios/Den-Haag/Verslagen/Process-mining/Bijlagen/ProcessMiningNGI22-09-10.pdf (Dutch)