Fishbone Diagram

Purpose

The technique answers a single question: where can this effect come from, across everything that might produce it. It exists because most failures have several contributing causes, which BABOK states as it opens root cause analysis, and because a group left to itself falls in behind the first explanation that sounds right and stops looking. The categories impose a sweep: they force the room to look under bones it would have walked straight past.

BABOK sets that purpose inside four activities, which shape the session as much as the deliverable: defining the problem statement, collecting data on the nature of the effect, its magnitude, its place and its timing, identifying the cause and identifying the corrective action that will prevent or limit the recurrence. Use is reactive or proactive: the diagram starts from a problem that has occurred and looks for where to correct it, or it starts from an area judged fragile and looks for what would make it give way, before the incident.

The deliverable is a one-page diagram: the measured effect at the head, the category bones, the primary and secondary causes hanging from them and a small number of causes retained as candidates, each one carrying the data that would confirm or rule it out. That last column is the real deliverable. BABOK's step 6 is explicit: the group has identified only potential causes, and further analysis, ideally data-driven, is required to validate the actual cause. A diagram that leaves the room without that column is a brainstorm that took the shape of a fish.

Usage

When to use it

• Multi-causal or unexplained failure: several plausible factors, none of them obviously dominant.
• Recurring problem despite repeated fixes: the fixes address the effect, reopen the space of causes.
• Ahead of a costly investigation: one hour is enough to bound where the data collection should look.
• Knowledge spread across several heads: the categories make operations, IT and legal speak in turn.
• Fragile area to sweep before the incident: BABOK allows proactive use on the same footing as reactive.
• Process already modelled: its own steps serve as bones, each cause attaches to a point in the flow.

When not to use it

• Single causal chain already plausible: the categories would fill with empty bones, run the five whys (BABOK 10.40b).
• Unmeasured problem: step 6 has nothing to decide with, instrument the process and count first, then rank with a Pareto.
• Combinatorial or safety-critical failure: a bone carries no AND, no OR and no probability, use a fault tree.

Building the diagram

The session starts with the effect, written in a box at the head of the diagram. It carries a fact, a measure and a period: "between the October wave and the November one, the abandonment rate of the online sign-up funnel rose from 18% to 34%" is an effect; "online sign-ups are going badly" is a complaint. A vague head produces vague bones, and the group spends the session negotiating what it is talking about. The facilitator then draws the spine from the box, across the board, then the diagonal bones that carry the categories and finally the secondary bones that carry the deeper causes: BABOK distinguishes three levels, primary cause, secondary cause and tertiary cause, and two are almost always enough.

The choice of the category set is the first real decision, and it is taken before the session. BABOK offers people, processes, tools and policies, specifying that the categories may include those, which makes it a starting point. Other sets exist, and each is tuned to a domain.

The process-type variant is worth a business analyst's attention, because it is the most useful of the six in that trade. When the failing process is already modelled, its steps become the bones, and the diagram stops being a classification and becomes a reading of the flow. The group no longer loses twenty minutes inventing categories, and each candidate cause attaches to a precise place in the process, which makes the next question immediate: what data does this point in the flow already produce.

Then comes the brainstorm, and the way it is run decides the quality of everything else. A public round-table in a steep hierarchy produces the causes that are safe to voice, and the cause that implicates a decision taken at the head of the room will appear on no bone. The counterweight is procedural and it fits in one instruction: everyone writes their causes in silence before anyone speaks, then the proposals are grouped without being attributed. Silent collection costs ten minutes and it is what separates a diagram of the real causes from a diagram of the admissible ones.

Step 6 is the one that produces the value, and it is the one that gets skipped. The wall is covered, the room feels it has worked, everyone leaves. BABOK is blunt: at this stage, the group has identified only potential causes. The analysis consists in retaining a small number of candidates, naming against each the data that would confirm or rule it out and going to get it. Only data promotes a cause. Step 7, the search for solutions, opens only once the actual cause is established, and opening it earlier means correcting a hypothesis.

Where the diagram misleads

BABOK names two limitations: the technique returns its full value when the analyst has had formal training in causal analysis, failing which the group files symptoms where it believes it is filing causes, and it becomes difficult on complex problems, where it can lead to a false trail or a dead end. Practice documents four more, and a practitioner who ignores them leaves with a covered wall and a false analysis.

The first is category capture. A cause is filed under the nearest bone, and it drops out of sight: it is there all right, in the wrong place, where nobody will look for it. Imbalance between the bones is the sign of it: one bone carries eleven causes, another carries one, dragged in by the hair. That imbalance calls for a re-check, bone by bone, on a single question: is each cause where it belongs. If the bone stays thin after that check, it is a finding, and the error is then to push a cause onto it to balance it out.

The second is the bone that carries an effect. "Delays" under Processes is a restatement of the head. The question to ask is what produces that delay, and the answer is the cause. This defect propagates: once an effect is written down as a cause, the secondary causes hung from it are the causes of a symptom.

The third is terminating on a person. BABOK's list of cause types makes it easy to reach, since it names human error and lack of training. A bone that stops there names a culprit, proposes a cheap action and closes the session. The useful cause is the systemic condition that allowed that error to count: the field the form did not require, the control that did not exist, the accountability nobody carried.

The fourth is taking a complete diagram for an answer. It is the most expensive, because it bears on what the technique delivers. A dense, well-ordered figure, exhaustive in appearance, looks like a conclusion. Every one of its bones is a hypothesis. Peerally and co-authors (BMJ Quality & Safety, 2017) bring three charges against root cause analysis as it is practised: undue focus on a single cause, which is what parallel categories exist to prevent; weak analysis, which is step 6 skipped; and poor follow-through, which is step 7 assigned to nobody.

The boundary with the five whys

The two elements of root cause analysis answer two different questions, and BABOK supplies the articulation itself: the five whys can be used alone or inside the fishbone technique, and once all the ideas are captured in the diagram, the why-chain serves to drill down to the root causes. The fishbone diagram goes wide: it opens up the families of causes, inventories them in parallel and commits to none. The five whys (BABOK 10.40b) go deep: they follow a single line down to a fixable condition, each link tested against evidence.

The junction happens at step 6, on data. A candidate cause the data confirms becomes the starting point of the five whys: the diagram says where to dig, the five whys dig. Committing to a chain is an act of judgement taken in the room, and the five whys show what that commitment costs and how to contain it. The diagram delivers the whole set of candidates and hands selection over to the data. That is its strength and it is the price it charges, since it leaves the validation work entirely undone and that work is precisely the one a room pleased with itself does not do.

AI considerations

The soundest use bears on what the diagram exists to produce and on what a room does least well: populating the space of candidates. A model that has read the incident history, the support tickets and the operations logs proposes causes category by category, including under the bone nobody was going to open. This is a direct attack on the room's blind spot, the one silent collection only half addresses, since it frees speech without extending knowledge. The second use is merging duplicates: forty proposals that say the same cause six times in six wordings are a facilitation cost, and a model groups them and proposes the category set that fits what the room said, rather than the one chosen out of habit.

Width is the deliverable, and a cause proposed by a model is worth what a cause proposed by a participant is worth, since neither is accepted without data (the calculation is different on the five whys).

Step 6 is a strict limit. Ranking the candidates and deciding which one warrants going after the data is an act of judgement anchored in the organisation's context. A model asked which cause is most likely answers with confidence from the text it has read, without having seen this health insurer's data. And a generated diagram, dense and complete, is this technique's most dangerous artifact: apparent completeness is what a room takes for an answer.

Examples

The case is a Swiss health insurer's online sign-up funnel, swept with BABOK's category set.

The real deliverable is what comes after the diagram. Three causes were retained as candidates, and each one carries the data that will confirm or rule it out. The People bone, as well stocked as the other three, produces none: the advice line was already saturated in October and the funnel had no more of an owner than before. The effect is a change, and a cause that did not change does not explain a change.

Cost

Tooling

The technique was designed for a wall and it asks for nothing more: a whiteboard, sticky notes, a marker. Ishikawa meant it to be runnable by a quality circle at the site of the problem: no data prerequisite, no tool to procure, a room and an hour. Sticky notes beat drawing straight onto the board with a marker, because a cause moves from one bone to another during the session and category capture is corrected by moving things.

Remotely, a digital whiteboard (Miro, Mural, Lucidspark) reproduces the method, and it brings two things the physical wall does not have: silent collection is native, everyone posting their causes without seeing them attributed, and grouping happens without copying anything out. The loss is the one common to every remote workshop, the instant overview a wall gives. A spreadsheet is enough as soon as it carries one row per cause, its category, its level and the column for the data to go after, and it is often the right support for what follows the session, once the diagram has become a list of candidates to validate.

Specialised tooling sits downstream, and it bears on validation. Quality management platforms (TapRooT, Intelex, Cority) include root cause analysis modules that impose a template, keep an auditable trail and allow one analysis to be compared with the next, which counts when the diagrams pile up and someone has to see which cause keeps coming back. Data visualisation tools serve step 6 more usefully than a drawing module: it is abandonment by funnel step that settles the candidates.

Sources

• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.40 Root Cause Analysis: the definition, the four activities, reactive and proactive analysis, the seven steps of the diagram, the categories given as a starting point, step 6 and its merely potential causes and the two limitations the guide names.
• Kaoru Ishikawa, Guide to Quality Control, Asian Productivity Organization (original Japanese edition, 1968): the origin of the diagram and its place among the seven basic tools of quality, designed to be run by a quality circle.
• American Society for Quality, What is a Fishbone Diagram?: asq.org/quality-resources/fishbone. The alternative category sets, 6M, 8P and 4S, which BABOK does not carry.
• Institute for Healthcare Improvement, Cause and Effect Diagram: ihi.org/library/tools/cause-and-effect-diagram. The classic set and the process-type variant, the most useful of the six in a business analysis context.
• M. F. Peerally, S. Carr, J. Waring, M. Dixon-Woods, The problem with root cause analysis, BMJ Quality & Safety 2017;26:417-422: qualitysafety.bmj.com/content/26/5/417. The peer-reviewed critique: undue focus on a single cause, weak analysis and poor follow-through on actions.