Structural and behavioural commonalities of process variants

Matthias Weidlich, Mathias WeskeZEUS 2010 · OpenReview

Abstract

In global enterprises, business processes often diverge into numerous variants due to local regulations, differing IT infrastructures, or departmental needs. This "process sprawl" complicates organizational restructuring and software rollouts. This paper introduces the concept of a "core process"—the invariant nucleus shared by different versions of a business process. By defining this core through structural and behavioral lenses, we provide a mathematical framework to identify outlier processes and redundant variants that should be merged.

Introduction

A single business process, such as "Invoicing" or "Employee Onboarding," rarely remains uniform across a large organization. While some variation is necessary, the independent evolution of process variants makes centralized management nearly impossible. The challenge lies in exploration: given a heterogeneous collection of existing process variants, how can an organization identify common ground? We propose a spectrum of abstractions to define the "core process," allowing for quantitative assessment of consistency and redundancy.

Defining the Core Process

The core process is not a single fixed entity but a spectrum of definitions categorized into two primary dimensions: structural and behavioral.

The Structural Dimension

This dimension focuses on the static "map" of the process, including nodes (activities) and arcs (transitions):

  • Activity Occurrence: Identifying tasks that appear in a specific percentage of variants.
  • Path Existence: Evaluating the consistency of paths between shared activities.
  • Subgraph Isomorphism: The most rigorous structural measure, identifying identical sub-graphs across models. This is computationally expensive (NP-complete) but provides the highest structural precision.

The Behavioural Dimension

This dimension analyzes how the process executes, accounting for logic that structure alone might obscure:

  • Optionality and Causal Footprints: Determining if tasks are mandatory and analyzing "look-back" or "look-ahead" constraints on the order of operations.
  • Behavioural Profiles: Examining relations between activity pairs, such as strict sequencing or concurrency.
  • Behaviour Equivalence: The most rigorous level, checking for trace equivalence or inheritance patterns.
Figure 2: Analysis of variants with complex control flow, including parallel gateways (a) and behavioral comparisons (b).

As illustrated in Figure 2, processes can appear structurally similar while possessing distinct behavioral logic, such as parallel execution paths (indicated by the '+' gateway), which necessitates behavioral profiling to ensure true commonality.

Experimental Evaluation

To validate these metrics, we utilized a synthetic dataset of four process variants (V1–V4) modeled in Business Process Modeling Notation (BPMN). These variants were used to test the mathematical application of structural and behavioral consistency metrics.

Figure 1: Synthetic process variants V1–V4 (a) and their derived commonalities (b).

Key Results

The application of our framework yielded actionable business intelligence:

  • Identifying Outliers: Variant V3 was identified as a structural outlier. Despite sharing many activities with the core, its path consistency was only 50%, suggesting it may be implemented on a fundamentally different technical infrastructure compared to V1, V2, and V4.
  • Identifying Redundancy: Variants V2 and V4 exhibited extremely high behavioral similarity (approaching 1.0). This quantitative result suggests that the organization could merge these two variants to reduce complexity without losing functional nuances.

Discussion and Future Work

A key insight of this research is that the "core" is a spectrum; the appropriate definition depends on the organizational goal, such as database cleanup versus legal auditing. However, significant trade-offs exist between precision and computational complexity. While checking activity occurrence is trivial, behavioral equivalence and subgraph isomorphism are exponentially harder to calculate.

Limitations

Currently, the framework abstracts away the "labeling problem"—the challenge of identifying that "Verify Invoice" and "Check Bill" refer to the same activity. Furthermore, while the theory is robust, future work must test these algorithms on large-scale real-world repositories containing thousands of complex models. We aim to develop a flexible framework that allows users to mix and match variability aspects to automate variant identification in large-scale environments.

Summary updated

BibTeX
@inproceedings{weidlich2010structural,
  author = {Matthias Weidlich and Mathias Weske},
  title = {{Structural and behavioural commonalities of process variants}},
  booktitle = {ZEUS},
  year = {2010},
  url = {https://openreview.net/forum?id=qIyubXkggR},
}
Figure