Add Data into Business Process Verification: Bridging the Gap between Theory and Practice

Riccardo De Masellis, Chiara Di Francescomarino, Chiara Ghidini, Marco Montali, Sergio TessarisAAAI 2017 · DOI

Introduction

In the field of Business Process Management (BPM), a significant divide exists between industrial practice and formal verification theory. While industry-standard tools like Bizagi or Camunda excel at modeling control-flow—the sequence of tasks—they often treat data as an opaque "procedural attachment." This paper, presented at AAAI-17, introduces RAW-SYS (Relational-AWare SYStem) to bridge this gap by integrating relational data directly into formal process models.

The Data-Flow Challenge

Traditional BPM tools focus on the sequence of tasks. However, complex data conditions—such as a loop that triggers only if a reimbursement exceeds a specific threshold—can lead to "unsound" processes that deadlock or fail to terminate. Because data is often handled via external scripts or black-box code, these errors remain invisible to standard modeling tools. Theoretical researchers have developed formalisms to solve this, but their models are often too abstract for engineers to use in practical software environments.

The RAW-SYS Framework

The authors propose a multi-layered framework that treats data as a first-class citizen. RAW-SYS combines Petri Nets for control-flow with Relational Models for data management.

The RAW-SYS architecture distinguishing between local and global data across multiple process cases.

A key feature of RAW-SYS is the distinction between global data (shared across the system) and local data (specific to a process instance). By adapting the semantics of Data-Centric Dynamic Systems (DCDS), the framework can handle bulk updates where a single task modifies multiple database rows simultaneously.

Decidability and AI Planning

Verification with unbounded data is mathematically undecidable. The authors demonstrate that by bounding the size of data stores and the number of concurrent cases, the verification problem becomes decidable in PSPACE. To perform this verification, they encode the RAW-SYS model into the C action language, allowing them to leverage off-the-shelf AI planners (specifically the $DLV^K$ system) to detect reachability errors—essentially asking the planner if an error state is reachable.

Evaluation: The Reimbursement Scenario

To test the framework, the authors utilized a Reimbursement (RB) scenario benchmark. This model simulates the lifecycle of an employee expense request, involving multiple review and filling stages.

A Petri net representation of the Reimbursement (RB) scenario used for benchmarking.

The evaluation focused on how verification time scales with complexity. They varied the Active Domain (ADOM) size (the number of employees/destinations) from 5 to 20 and the number of Pending Requests (PR) from 2 to 10. This allowed them to test how the verification time scales as the complexity of the data environment grows.

Key Results and Insights

  • Performance: Simple reachability queries (e.g., "Can this request be processed?") were resolved in under a second. More complex queries (e.g., "Can a request be accepted?") took approximately two minutes for a domain of 20 employees and 10 requests.
  • Sensitivity: Verification time was found to be more sensitive to the size of the data domain (ADOM) than the number of concurrent process instances.
  • Theoretical Bridge: The paper establishes a formal link between Petri Nets and DCDS, proving that size-bounded verification is feasible and decidable.

Conclusion and Future Work

RAW-SYS demonstrates that treating data updates as formal actions rather than hidden code allows for the detection of semantic errors that traditional BPM suites miss. While the current system requires manual tuning for state-boundedness and focuses primarily on reachability, future work aims to support complex temporal logic (LTL/CTL) and optimize performance for larger enterprise processes. By framing process reachability as an AI planning problem, the authors have opened a new pathway for robust, data-aware business process verification.

Summary updated

BibTeX
@inproceedings{masellis2017add,
  author = {Riccardo De Masellis and Chiara Di Francescomarino and Chiara Ghidini and Marco Montali and Sergio Tessaris},
  title = {{Add Data into Business Process Verification: Bridging the Gap between Theory and Practice}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2017},
  doi = {10.1609/aaai.v31i1.10688},
}
Figure