---
title: Process Model Complexity Metrics
type: concept
tags: [bpm, modelling, quality, metrics, complexity, understandability, error-prediction]
sources:
  - "[[sources/2010-mendling-reijers-vanderaalst-7pmg]]"
  - "[[sources/2018-dumas-fundamentals-of-bpm]]"
created: 2026-05-04
updated: 2026-05-04
---

# Process Model Complexity Metrics

Bottom-up structural metrics computed mechanically over a process model graph. Used to predict **error probability** and **comprehensibility**, and to operationalise size/complexity-related guidelines from [[concepts/7pmg]].

[[sources/2010-mendling-reijers-vanderaalst-7pmg]] cites these metrics as the **empirical foundation** for guidelines G1, G2, G4, G7. They were validated on the SAP Reference Model (600 EPCs) and on a 2000-EPC industrial corpus.

## Core metrics (control-flow models — BPMN, EPC, YAWL, Petri nets)

### Size

- **|N|** — number of nodes (activities + events + gateways).
- **|A|** — number of arcs (sequence flows).

Empirically: error probability rises sharply with size; ≥50 elements → error probability > 50% ([[concepts/7pmg]] G7).

### Density

$$\text{density} = \frac{|A|}{|N| \cdot (|N|-1)}$$

Ratio of actual arcs to theoretically possible arcs. Higher density → more interconnected → harder to understand.

### Connector degree

For each gateway/connector c:
- **in-degree(c)** = number of incoming arcs
- **out-degree(c)** = number of outgoing arcs
- **degree(c)** = in-degree + out-degree

Aggregates:
- **avg-degree** — average across all connectors
- **max-degree** — worst-case connector

7PMG G2 ("minimise routing paths per element") targets degree directly. High avg/max degree negatively correlates with understanding (n=73 student questionnaire study).

### Control-Flow Complexity (CFC) — Cardoso

Sum across split connectors of:
- **XOR-split**: degree contribution = number of outgoing arcs
- **OR-split**: degree contribution = 2^(out-degree) − 1 (combinatorial branch-set count)
- **AND-split**: degree contribution = 1

CFC is bounded above by the number of execution paths the model permits. Validated against perceived complexity (Cardoso) and predictive of errors (Mendling).

### Depth and nesting

- **Depth** — maximum nesting level of structured blocks.
- Deeper nesting → harder to comprehend for human readers.

### Mismatch

Count of split-connectors whose matching join is of a different type (e.g., AND-split matched by XOR-join).

7PMG G4 ("model as structured as possible") aims to minimise mismatch.

### Connector heterogeneity

Diversity of connector types (entropy across {AND, XOR, OR}).

### Cross-connectivity

A graph-theoretic metric (Vanderfeesten et al.) capturing how strongly nodes are causally linked. Negatively correlates with errors.

### Cyclomatic complexity (graph-theoretic)

$$V(G) = |A| - |N| + 2 \cdot p$$

where *p* is the number of connected components. Adopted from McCabe's software complexity. For most well-formed process models *p* = 1.

## Interpretation

| Metric ↑ | Effect on understanding | Effect on error probability |
|---|---|---|
| Size (|N|, |A|) | ↓ | ↑ |
| Density | ↓ | ↑ |
| Avg/Max connector degree | ↓ | ↑ |
| CFC | ↓ | ↑ |
| Depth | ↓ | ↑ |
| Mismatch | ↓ | ↑ |
| OR-connector count | ↓ | ↑ |
| Cross-connectivity | ↑ | ↓ |

Empirically validated on EPC corpora; replication on BPMN is partial.

## Use in 7PMG

| 7PMG Guideline | Targets metric |
|---|---|
| G1 (few elements) | size |
| G2 (minimise routing paths) | avg/max connector degree |
| G3 (one start, one end) | reduces start/end count → enables soundness analysis |
| G4 (structured) | mismatch → 0 |
| G5 (avoid OR) | OR-connector count → 0 |
| G6 (verb-object labels) | label-ambiguity (orthogonal axis) |
| G7 (decompose if >50) | size threshold |

## Use in evaluation rubrics

For an as-is BPMN model produced from interview discovery:

1. Compute the metrics above with a static analysis tool (Camunda Modeler shows size; ProM plug-ins compute CFC, density, mismatch).
2. Flag thresholds: size > 50 (G7); max degree > 4 (G2 violation); any OR connector (G5); density > 0.1 typically too dense.
3. Cross-reference with 7PMG guidelines to suggest fixes.

## Limitations

- **Predictive, not deterministic** — metrics correlate with errors but do not entail them. A high-CFC model may be perfectly correct; a low-CFC model may still hide a soundness violation.
- **Notation-specific weights** — metric weights calibrated on EPCs; BPMN re-calibration is partial.
- **Pragmatic-only signal** — these metrics target understandability/error-likelihood, not [[concepts/sequal-framework|semantic]] correctness (whether the model matches reality) or [[concepts/soundness|behavioural]] correctness.
- **Aggregation hides locality** — a model with one very-high-degree connector may have acceptable averages; report max as well as avg.

## Tool support

- **ProM** — Petri-net / EPC plug-ins compute CFC, mismatch, structuredness.
- **Camunda Modeler** — element count, basic structural validation.
- **Signavio / SAP Signavio** — bottom-up metrics + 7PMG-style heuristics.
- **bpmn.io** ecosystem — programmatic access to BPMN graph for custom metric computation.

## Related
[[concepts/7pmg]] (the actionable distillation of these metrics) · [[concepts/process-model-quality]] · [[concepts/soundness]] (orthogonal correctness layer) · [[concepts/sequal-framework]]

## Sources
[[sources/2010-mendling-reijers-vanderaalst-7pmg]] (primary — empirical validation) · [[sources/2018-dumas-fundamentals-of-bpm]]