Modeling Business Process Variability: Are We Done Yet?

It is common for organizations to maintain multiple variants of a given business process, such as multiple sales processes for different products or multiple bookkeeping processes for different countries. Conventional business process modeling languages [2] do not explicitly support the representation of such families of process variants, requiring one of two paths to be chosen. Either each variant is modeled separately or multiple variants are modeled together. The former option results in duplication as the variants have much in common, potential inconsistencies as the variants are modified independently of each other, and missed opportunities for identifying shared IT services that can support these business processes. The latter option leads to highly complex consolidated models, which hamper the analysis and maintenance of individual process model variants. These shortcomings triggered significant research efforts over the last two decades, leading to an array of approaches to business process variability modeling [1]. A common trait of these approaches is that they extend a conventional process modeling language with constructs to represent customizable process models. A customizable process model captures a family of process model variants in a way that the individual variants can be derived by adding or deleting fragments according to customization parameters or to a domain model. Accordingly, a customizable process model encapsulates customization decisions between process variants that need to be made either at design-time or run-time. Designtime customization decisions lead to a customized process model that is intended to be executed in a particular organizational setting. Hence, these decisions affect all instances of the customized process executed in this setting. The timeframe associated with these decisions may be long (e.g. months or years). In contrast, run-time customization decisions are punctual and affect only one or a few process instances. Such decisions may be visualized on top of a process model, but they are not intended to modify the executed process model itself, beyond its effects on the process instance(s) where the decision is applied. This talk retraces the last two decades of research in modeling business process variability via customizable process models. The talk draws up a taxonomy and comparative analysis of approaches in this area, shedding light on strengths and weaknesses of each approach, with the ultimate goal of distilling relative trade-offs and practical criteria for selection. Are there still research gaps in modeling business process variability, or are we done with it?