What is it and why do we need it?
When using Process mining to analyze common business processes like Procure-To-Pay (P2P) or Order-To-Cash (O2C), one makes two crucial assumptions:
- There is a single object being processed and thus there is one notion of case. This can be the Purchase Order in P2P or Sales Order in O2C.
- An event belongs to a single case. So, if Purchase Order is our case, and we have a received event for a case, the receipt event will appear only in that case (for the Purchase Order object) and will not be related to any other object (like a Packing Slip or an Invoice).
These two assumptions often fail in real-life. A single process may involve multiple objects: a P2P process involves Purchase Orders, Purchase Order Items, Packing Slips, and Invoices. Moreover, similar events, like the invoicing of a Purchase Order, can be related not only to a Purchase Order, but also to a specific list of items appearing on the invoice.
Object-Centric Process Mining, one of the hot new trends in process mining research and industry, relaxes the two assumptions [1]. Cases can comprise several objects with various hierarchical relationships between them: a Purchase Order relates to a set of items, a subset of these items may appear on an invoice, and another subset of the items can be packed together and shipped in a single delivery. But this is easier said than done.
Let’s consider the very basic task of process discovery, which involves creating a graphical representation of the “true” process as it appears in the data. In traditional process mining, with a single object notion of cases, one clearly understands process maps as graphical creatures that describe how cases flow between different stages. Relaxing the first assumption means that cases comprise several objects and it is no longer clear what a transition between two stages means: is it the item flowing between the stages, or alternatively, is it the invoice that is being processed?
One main challenge is how to integrate the understanding that process mining is currently making unrealistic assumptions into industrial tools. First, to represent the different objects, one must extend the classical notion of an event log beyond the current assumptions, as well as the basic notion of process map. Event logs will be required to contain object descriptions (and not case descriptions). Process maps must be extended with multi-object representations of activities or arcs. For example, one way to extend process maps would be to create different arc types (e.g., arc colors) for different types of objects. Another way would be to denote activities differently per different object type. In any case, there is no definitive answer as to the best way of representing objects in the discovered models.
Second, even if we have a representation of multiple objects in our data and in our models, a big question remains: how does one make use of the fact that we have integrated objects into our data/models to provide better insights to users of process mining technologies?
The answer, in my opinion, can come in various shapes and forms. Compliance checking using a single case notion shows only part of the picture. The purchase order was processed on time, but was the package fully received? Did every package get to its destination on time? Were some of the packages routed inefficiently? These compliance questions will be enabled once we have adopted our data and models appropriately. Another example is performance and bottleneck analysis. A purchase order may be delayed due to its internal processing (e.g., approving the requisition may be taking a very long time). But it can also be delayed due to other objects, such as a delay in receiving one of the invoices that contains part of the items on the purchase order. To be able to answer compliance and performance questions, we must consider drifting away from the single-case (per log/model) perspective and emerge into the new, exciting, and uncertain world of object-centric process mining.
Postscript:
A great scientific read on object-centric process mining that I found inspiring when writing this piece is the following paper by Wil van der Aalst:
Arik Senderovich, PhD mindzie
