Predictive Process Monitoring

For obvious reasons, seeing into the future has always been a very attractive idea for mankind. Prophets and seers are mentioned in the earliest written history as unique people who could see what tomorrow brings. If one knows what is going to happen, one will be able to plan their steps without the usual uncertainty that accompanies daily decision-making. Knowing that the next train is going to be late will allow one to make the decision of using their car for their upcoming commute, leading to a desired outcome of not being late for work.

In our context, forecasting business outcomes and performance measures is an attractive task for very similar reasons. If the business knows that a customer is going to be paying late, say as part of an order-to-cash process, it will be able to proactively contact the customer and urge them to pay on time. Knowing that a delivery is going to be late, as part of a procure-to-pay process, will allow a business to nudge the courier to be on time or else pay a steep penalty.

In recent years, AI and Machine Learning have significantly improved our capability to predict the future, using methods like deep learning, where neural networks are trained to predict outcomes and performance measures. It is therefore unsurprising that the area of process mining has made attempts to borrow methods from these advanced forecasting technologies and adopt them in the context of business process forecasting. Tasks, like predicting trends of Key Performance Indicators, foreseeing case outcomes, and predicting the timestamps of future activities, have been studied in the academic context for almost a decade yet did not receive much attention in industrial tools.

The area of predicting business decisions, outcomes, and measures received its own name within the process mining field: it became known as predictive process monitoring (or simply, PPM).

The idea behind PPM is that instead of using only generic features such as case type, or item type as input into deep learning algorithms, one can also enrich the feature space with process-related information, e.g., previously executed activities, their duration, and whether their execution was conforming or deviating from normal. This historical process information often available in event logs can help deep learning methods to better predict process outcomes such as delivery timeliness, customer payments, and other service level agreement-related measures. One can view this as applying AI technology in process context, which is what process mining is essentially all about.

Without a doubt, PPM will be a huge differentiator in the process mining market. The process mining race will be won by companies that will adopt PPM as an integral part of their toolbox and enable their clients to better plan the future, especially, given the amounts of uncertainty the current economies are facing.

Arik Senderovich

Advisor, mindzie
www.mindzie.com
Process Mining Simplified