Seven Process Modeling Guidelines (7PMG)

Mendling, Reijers and van der Aalst's synthesis of empirical process-modelling research into seven actionable guidelines for casual modellers. Distinguishes itself from prior frameworks (SEQUAL, Guidelines of Modeling) by being operational (each guideline is concrete and immediately applicable) and empirically grounded (each builds on quantitative studies).

Motivating gap

Existing model-quality work splits into four streams the paper finds insufficient for practitioners:

Top-down quality frameworks (SEQUAL, GoM) — theoretically clean, distinguish quality categories, but too abstract to be applied by novices.
Bottom-up metrics (Cardoso CFC, Canfora maintainability, Mendling et al. error metrics) — fragmented, partial empirical validation.
Empirical surveys of modelling techniques — language-level conclusions (e.g., EPC vs Petri net understandability), not directly actionable on a single model.
Pragmatic guidelines from practice (e.g., "make your models hierarchical") — actionable but anecdotal, no empirical foundation.

7PMG sits in the middle: actionable like (4), empirically grounded like (2). Real-world modelling projects show error rates of 10–20% in industry corpora, motivating the need for simple guidance.

Empirical foundations

The seven guidelines build on three measurement axes from prior Mendling-Reijers-Aalst studies:

Process model understanding — questionnaire study (n=73 students across TU/e, Madeira, Vienna). Found OR-joins and average connector degree negatively correlated with comprehension.
Error probability — prediction functions for relaxed soundness on the SAP Reference Model (600 EPCs) and EPC Soundness on 2000 industrial EPCs. Trace error probability back to structural metrics; size and complexity are the dominant drivers.
Ambiguity of activity labels — experiment with 29 TU/e postgraduates comparing verb-object ("send letter") vs. action-noun ("letter sending") vs. other styles. Verb-object significantly less ambiguous and more useful.

The seven guidelines (Table 1)

#	Guideline	Empirical basis
G1	Use as few elements in the model as possible	Larger models harder to understand and have higher error probability
G2	Minimise the routing paths per element	High connector degree (in-arcs + out-arcs) correlates with errors and reduced understanding
G3	Use one start and one end event	Multiple starts/ends increase error probability; workflow engines often require single start/end; enables soundness checks
G4	Model as structured as possible	Every split matched by join of same type — "balanced brackets". Unstructured models more error-prone, less understandable
G5	Avoid OR routing elements	Semantic paradoxes in OR-join, implementation difficulties; AND/XOR-only models less error-prone
G6	Use verb-object activity labels	Verb-object significantly less ambiguous and more useful than action-noun or rest-category labels
G7	Decompose a model with more than 50 elements	Error probability exceeds 50% above this threshold; split single-entry-single-exit subcomponents into sub-processes

Each guideline preserves model behaviour modulo branching bisimulation when applied as a transformation — the resulting model is more understandable but behaviour-equivalent.

Worked example (Section 3.4)

The paper uses a complaint-handling EPC from a Dutch governmental agency (37 elements, multiple start/end, OR-join, mixed labelling styles). Sequential application of the guidelines produces a transformed model with 31 elements, single start/single end, structured nesting, no OR-join, verb-object labels — same behaviour, demonstrably easier to read.

Prioritisation (Section 4)

Guidelines can conflict: applying G2 (reduce connector degree) may add elements, violating G1. The paper takes a pragmatic approach — workshops with 21 professional modellers (7 in Germany — Berliner BPM-Offensive; 14 in the Netherlands — major consultancy) — to elicit priority ordering. Theoretically motivated, empirically validated prioritisation is acknowledged as out of scope for this paper.

Limitations & merits (Section 5)

Limited to control-flow perspective — does not cover data, organisation, or compliance dimensions.
Some guidelines (e.g., G1, G7) are size-based heuristics, not formal correctness criteria — pair with verification (cf. soundness, sources/1998-vanderaalst-verification-of-workflow-nets) for full quality coverage.
Guidelines are language-agnostic in spirit (illustrated on EPCs, applicable to BPMN, YAWL, UML AD) but the empirical studies were predominantly EPC-based.
Builds on branching-bisimulation equivalence — guidelines describe transformations within a behaviour-equivalence class, not redesigns.

Connections

Concepts: concepts/process-model-quality · concepts/7pmg · concepts/process-model-complexity-metrics · concepts/soundness (G3 enables soundness analysis) Frameworks: frameworks/bpmn (canonical target notation today) · EPCs (paper's primary illustration) Related sources: sources/2018-dumas-fundamentals-of-bpm (§5.4.4 cites 7PMG) · sources/1998-vanderaalst-verification-of-workflow-nets (formal soundness — complementary verification layer) · sources/2006-krogstie-sindre-jorgensen-revised-sequal-framework (the SEQUAL framework 7PMG positions itself against) Authors / hub entities: entities/jan-mendling · entities/hajo-reijers · entities/wil-van-der-aalst

Open follow-ups

Theoretically-motivated prioritisation scheme (acknowledged gap).
Extension beyond control-flow: data-quality, organisational-quality guidelines.
Empirical re-validation on BPMN corpora (original studies were EPC-heavy).