Developed by Kaoru Ishikawa (Guide to Quality Control, 1968).
Lays out the possible causes of a defect or failure on branches by category, such as method, machine, material, and people, so no family of causes is skipped.
Why invoices go out eleven days late, with the causes laid out by category
The effect sits at the head; each category branch holds the causes the team named, and the causes behind them. Select a cause to read what the team knew about it, then mark the two you would test with data first. Illustrative.
0 of 2 marked1 of 2 marked2 of 2 marked
An example
A facilities maintenance company was billing its commercial customers an average of eleven days after each job closed, against a target of two, and about $1.4 million of cash sat waiting at any time. The first explanation in the room was the billing clerk, who covers three regions alone.
The operations manager drew the effect at the head of a whiteboard and six branches behind it: method, machine, material, people, measurement, and environment. The group had to name at least one cause on every branch before anyone argued. The people branch filled fastest. The method and machine branches turned out to hold the bigger causes: billing waited for paper job sheets that rode back with the technicians, the field app captured the customer's signature but never passed it to billing, and invoices were cut only on Fridays.
The team marked two causes to test. A sample of sixty late invoices showed that forty-one had waited on a paper sheet. Sending the app's sign-off straight to billing and running invoices daily cut the average to three days within a quarter.
Why it works
The diagram forces breadth before depth. A group investigating a failure usually seizes the first plausible cause, often the one that names a person, and stops looking. Fixed categories make the group fill every branch, so causes in tools, inputs, measurement, and the setting get a hearing even when nobody in the room thinks of them first.
Putting every candidate in one picture does two more things. It shows when causes on different branches are the same thing seen twice, and it turns an argument about the answer into a shared list of possibilities. The last move matters most. The diagram proves nothing on its own; it is a map of hypotheses, and the likeliest two or three are tested with data before anything is fixed.
The thinking behind it
Kaoru Ishikawa Kaoru Ishikawa, an engineering professor at the University of Tokyo and a leader of the Japanese quality movement, popularized the cause-and-effect diagram from the 1960s, and it now carries his name as well as the fishbone label. He used it in quality work at Kawasaki and set it out in his Guide to Quality Control (1968). He is also known for quality circles, small groups of workers who study and fix problems in their own work. The basic idea of a cause-and-effect diagram is older than his books; he made it a standard tool.
How to run it
- Write the effect as a measured fact. Invoices go out eleven days after the job closes gives the group a number to explain; billing is slow invites opinions.
- Choose categories that fit the work. Method, machine, material, people, measurement, and environment suit operations; a service team may swap in policies or systems, so long as the set covers the work.
- Fill every branch before going deep. Ask for at least two causes per category, so the quiet branches get a hearing before the loud ones.
- Ask why of each cause. Add the causes behind it on smaller branches until the next answer would be outside the team's reach.
- Note how each cause could be tested. A sample, a log, or a timing study written beside each leaf turns the drawing into a test plan.
- Test two or three before fixing anything. Pick the likeliest and the cheapest to check, and let the data choose the fix.
- Strike what the data rules out. Keep the corrected diagram, since the same effect tends to come back.
Where it goes wrong
The drawing becomes the answer. A full whiteboard looks like analysis, and the team fixes the cause it liked best. Spot it when no cause has a test written beside it, and do not act until two causes have been checked against data.
The people branch fills first and fullest, often with names, and the session turns into blame. For each people cause, ask what in the method or the system lets it happen; that is usually where the fix is.
Branches get padded to look complete. A category with nothing real in it can stay short, and weak causes added for symmetry only dilute the test list.
When to use something else
- Five Whys. When the problem is small and one chain of causes is enough, so a full diagram would be ceremony.
- Pareto Principle. When the causes are already logged and counted, and the job is to rank them by how much of the problem each explains.
- Causal Loop Diagram. When the causes feed one another in loops, which a branching diagram cannot show.
Where it sits in SOLVE eX
Go to the source
- Guide to Quality Control, JUSE, 1968