Skip to content

Toolkit — Ishikawa (fishbone) diagram

Gate: G2 Decompose. Category: structure substrate.

What problem it solves

Some decomposition problems are not about the scope of a question but about the causes of an observed effect. The Ishikawa diagram forces the group to enumerate candidate causes across a fixed set of categories (typically six: methods, machines, materials, measurements, environment, people) so that likely causes are considered exhaustively before any single one is investigated. Without the discipline, the engagement pursues the first plausible cause and confirms it; with it, the space of causes is made visible first.

How it is used

A 60-minute G2 workshop, usually called when the problem is naming an effect and the team needs to organise causal hypotheses before routing. The chair draws the spine (the effect at the head), six bone-categories branching off, and the group populates each category with candidate causes. The output is a fishbone diagram, which is then ranked for testing priority and handed to the issue tree or pyramid for decomposition into questions.

Inputs

  • A clearly named effect — usually from the framed problem paragraph's slot 2 (observation).
  • Practitioners covering the six categories (or an equivalent set).
  • A willingness to include causes that are uncomfortable (people, measurements) — the most common omission.

Outputs

  • A fishbone diagram with the effect at the head and candidate causes along branches.
  • A ranked list of causes-to-test, prioritised by likelihood × investigability.
  • Category-coverage notes: which of the six categories returned empty, and why.

Visualisation

Ishikawa fishbone diagram — effect at the head with six diagonal category bones each branching into causes Effect unloading time +16 min Methods → radio-handoff gaps → dispatcher off-map Machines → yard-map out of sync → radio dead-zones Materials → carrier feed stale → paperwork late Measurements → ETA feed accuracy → override not logged Environment → weather rare → yard layout People → training gaps → staffing ratios

The spine points at the effect. Six diagonal bones carry cause categories; each bone carries the specific causes nominated in that category.

Anatomy

Effect. The single effect the diagram explains. Named precisely, with the observation's numeric form. "Unloading time has grown by 16 minutes" — not "operations are slow."

Spine. The horizontal arrow. It is a visual device, not a claim; the spine does not imply the cause is somewhere along it.

Six categories. Methods, machines, materials, measurements, environment, people. The categories are a prompt — they prevent the team from converging too quickly on one category and missing the others. Some domains use four categories (the 4M version) or eight; the discipline is the same.

Causes per bone. Specific, testable causes. A cause like "staff motivation" is too vague; "training turnover — 6 of 12 dispatchers have less than 6 months' tenure" is testable.

Ranking. After enumeration, the team ranks causes by likelihood (how much they would contribute to the effect if true) and investigability (how quickly the team can determine truth). The top candidates become G2's issue-tree leaves; low-ranked causes are logged for later if the top candidates don't explain the effect.

Category-coverage notes. If a category returned no causes, record why. "Environment — weather not a driver in the trend window." Empty categories are evidence of scope, not laziness, only if the emptiness is explicit.

Example

Paper trail — the unloading-time fishbone

G2 Ishikawa workshop, 70 minutes. Chair: Ada. Team: Raj, Priya, Jin, Amira, two dispatchers (Chen, Martinez), one dock operator (Lee).

T+0 — effect. Ada writes the head: unloading time has grown by 16 minutes, from 28 to 44 min.

T+10 — methods. Chen: "dispatcher radio-handoff gaps — we have 30-second lags between steps." Martinez: "dispatchers go off-map — the yard map is the plan; the radio is the reality." Two causes.

T+20 — machines. Priya: "the digital yard-map system is out of sync with the paper board." Jin: "the dock-door radios have dead-zones in bays 4–6." Two causes.

T+30 — materials. Raj: "carrier-feed staleness — the feed updates every 15 minutes but trucks move faster." Chen: "paperwork arrives late on about 12% of trucks." Two causes.

T+40 — measurements. Ada: "how do we measure ETA accuracy today?" Priya: "we don't — the feed sends a number; we never compare it to arrival." Martinez: "overrides aren't logged structurally." Two causes; both named as measurements the engagement will have to build to investigate the others.

T+50 — environment. Lee: "weather — but rare, maybe 4 days a year that actually affect unloading." Raj: "yard layout — bay 7 is on the far side, cold and poorly lit at night." Two causes; weather noted as a known-uncovered scope decision.

T+55 — people. Chen: "six of twelve dispatchers have less than six months' tenure." Lee: "dock operator staffing ratios — we're down two from last year." Two causes, both named without defensiveness.

T+65 — ranking. The team ranks. Top: dispatcher radio-handoff gaps, yard-map out of sync, carrier-feed staleness, training turnover. Bottom: weather (low likelihood), yard layout (low investigability in the engagement horizon).

T+70 — close. Twelve causes, four top-ranked, two coverage notes. The top four become issue-tree leaves. The training-turnover cause maps to a G3 observation — it is not an AI-shaped leaf, but it is a leaf the engagement's recommendation has to address.

Pitfalls

Vague causes. "Bad communication", "low morale". Push for one observable, testable form. The fishbone is not a place for abstractions.

Missing categories. The team fills three of six and moves on. Run all six, record empties. The most common omission is measurements — because teams often have not thought about what they are and aren't measuring.

Categories as buckets for solutions. "Training" as a cause can slip into "run more training" as an item in the same branch. The diagram is of causes, not remedies.

Branch-length bias. The team assumes whichever branch has the most causes is the main one. Branch length is a function of fluency, not truth. The ranking step is what separates them.

Stopping at the first confirmed cause. Ishikawa is resistance to this. A cause that fits the effect is a hypothesis, not a conclusion. The engagement investigates, and the fishbone records the others so they are not lost.

When not to use

  • Effects that are definitional ("users are dissatisfied because their job-to-be-done is not served" — the mapping is the analysis, not the causes).
  • Engagements where the cause is already known and the engagement's question is about the remedy. The fishbone is diagnostic, not prescriptive.

Provenance

Kaoru Ishikawa developed the diagram at Kawasaki Steel in the 1960s; the canonical treatment is his Guide to Quality Control [1]. It is one of the seven basic tools of quality (the Q7), established by the Japanese Union of Scientists and Engineers. The 6M category set is the most widely used; variations (4M, 8M, the service-oriented 4P) serve different domains.

  • Issue tree. Top-ranked fishbone causes become issue-tree leaves.
  • Pre-mortem (G3). Ishikawa is diagnostic-retrospective; pre-mortem is diagnostic-prospective.
  • Systems-Theoretic Process Analysis (STPA). When the effect is a safety-relevant accident, STPA replaces the fishbone as the decomposition of cause.

Verification

[1] Ishikawa K. Guide to Quality Control. Asian Productivity Organization; 1976 (orig. 1968, Japanese). [verified] The canonical treatment of the fishbone diagram and the 6M categories.