When CMMN is overkill (most of the time)

Motto — Reach for CMMN only after BPMN has actually failed you in the modeler — not in the meeting where someone said “our process is too dynamic”.

Part of Phase 06 — CMMN: case management. Concept lesson — no code required. The phase’s honest closing.

The Problem

CMMN is elegant and standardised, and most organisations that adopted it regret it. Camunda deprecated their implementation years ago for lack of use. Flowable keeps a first-class one — it’s genuinely good, and Flowable Work’s case UIs lean on it — but the industry pattern is clear. Teams choose CMMN in the architecture phase for flexibility they never exercise, then pay its costs for years: scarcer skills, weaker tooling ecosystems, harder testing, and models reviewers can’t read. This lesson is the checklist that keeps you off that path, and the short list of cases where CMMN earns its keep.

The Concept

Why “too dynamic for BPMN” is usually false — the three escape hatches you already own:

Most “dynamic” requirements decompose into interruptions (event subprocesses), exceptions (boundary events), rework (loops), and variants (DMN-driven routing). What remains — genuine practitioner authority over sequencing, lesson 01’s litmus test — is rarer than the meeting believes.

The real costs you accept with CMMN, spelled out:

Cost	Why it bites
Skills	every engineer/BA knows some BPMN; CMMN fluency is rare — reviews, hiring, and handovers all pay
Tooling & ecosystem	modelers, linters, courses, StackOverflow depth — an order of magnitude thinner
Testability	a process has paths you can enumerate and assert; a case has state spaces — “any order, maybe never, maybe twice” multiplies test scenarios
Legibility	sentries encode logic invisibly (an AND across ON-parts, an IF-part in XML) — the diagram shows less of the behaviour than BPMN’s does, the opposite of Principle 2’s promise
Ops intuition	“where is case X stuck?” has fuzzier answers when nothing was ever supposed to happen next (Phase 9’s oldest-instance signal needs rethinking per case model)

CMMN genuinely earns it only when all three hold at once:

Practitioner authority over sequencing is real and defended — people like investigators, adjusters, and clinicians, whose judgment is the product.
The discretionary surface is large: a dozen-plus optional activities. Below that, one BPMN stage with event subprocesses covers it.
Guardrails still matter — sentries, milestones, required items. Without them, you wanted a task list, not an engine.

Fraud investigation (lesson 02) passes all three. Dispute handling passed by containing its discretion in one stage (lesson 03). The capstone fails condition 1 — and that’s the normal case, which is the point.

Ship It

This lesson ships outputs/cmmn-adoption-checklist.md — the escape hatches, the three-conditions test, and the containment strategy — closing Phase 6 and, with it, the course’s last engine.

Check Yourself

Q1. “Our onboarding is too dynamic for BPMN — documents arrive in any order.” The first answer is…

A) CMMN
B) BPMN with event subprocesses / message catches per document — arrival order is an event problem (Phase 7), not a discretion problem
C) a task list
D) Temporal

<details><summary>Answer</summary>B — unordered inputs are not unordered work. The litmus test asks who sequences the work, and onboarding’s work is prescribed.</details>

Q2. CMMN’s testability cost comes from…

A) slow engines
B) state-space explosion — optional, repeatable, order-free items multiply the scenarios a test suite must cover versus a process’s enumerable paths
C) missing assertions
D) XML

<details><summary>Answer</summary>B — discretion for the worker becomes combinatorics for the tester. Budget for it, or constrain the model.</details>

Q3. The containment strategy for a mostly-prescribed flow with one discretionary pocket is…

A) model everything in CMMN for consistency
B) BPMN end to end, with the pocket as a case (via case service task) or the case wrapping processes (lesson 03) — discretion quarantined where it’s real
C) two separate products
D) gateway spaghetti

<details><summary>Answer</summary>B — lesson 03’s seam exists precisely so CMMN’s costs are paid only on the stage that needs its powers.</details>

Challenge. Take lesson 02’s fraud investigation and, as devil’s advocate, model it in plain BPMN using only the escape hatches. Write down exactly where it breaks. Hint: “interview may repeat, statements may be skipped, order is judgment” survives, and “escalation stage unlocked by evidence” does too — what finally breaks is completion semantics. Knowing precisely where BPMN fails is what earns you the right to use CMMN there.

Phase README: CMMN: case management
The escape hatches: 7.05 · 4.04 · Phase 5

When CMMN is overkill (most of the time)

The Problem

The Concept

Ship It

Check Yourself

Related