Problem solving

Five whys

In French: Les cinq pourquoi

The five whys consist of going back from an observed problem to a cause one can act on, by asking at each step why the previous step occurred.

The principle, and what the number five means

You start from a precise fact, ask why it occurred, then put the same question to the answer obtained, and so on. The point is not in the question — a four-year-old asks it better than we do — but in the obligation not to stop at the first satisfying explanation.

The number five is in no way a rule. It indicates an order of magnitude: below three levels, one almost always stays on symptoms; beyond six or seven, one generally leaves the scope where one can act to arrive at general considerations about the organisation or the market.

The right stopping criterion is not the count of levels, it is the nature of the answer: you stop when you hold a cause whose removal would prevent the problem from coming back, and which belongs to someone identifiable. Everything that comes after belongs to another problem.

What distinguishes a valid chain

Each level must read backwards without effort: if the cause put forward had been absent, the effect of the level above would not have occurred. That reverse-reading test eliminates most complacent chains, those where one has slipped from a causal link to a mere chronological succession.

Each level must also be verifiable. A chain built in a meeting room is a hypothesis, and it is checked on the shop floor — by looking at the station, reading the record, questioning the person who was present. A plausible and false chain looks exactly like a correct one.

Finally, an answer that blames a person almost always signals a skipped level. "The operator made a mistake" is not a cause: it is the place where one must ask why the error was possible, and why nothing stopped it before it produced its effects.

When to use it, and when to move on to something else

The tool suits one-off, well-delimited problems where a single causal chain is suspected: a line stoppage, a lost file, a missed delivery. It is quick, requires no preparation, and is practised standing in front of the place where the problem occurred.

It becomes unsuitable as soon as several causes contribute to the same effect, or the problem shows up in a scattered way. A single-line trace-back will then give a real but partial cause, and the resulting action will reduce the phenomenon by only a fraction — a disappointing result that will wrongly be attributed to poor implementation.

In that case, the fan-out of a cause-and-effect diagram or the ranking of a Pareto suits better. The two approaches in fact combine well: the Pareto points to the category that weighs the most, the trace-back by successive whys digs inside that category.

The common mistakes

Going down to a cause out of reach. "Because the market demands short lead times" closes the discussion without enabling anything. One should have stopped one or two levels earlier, where someone could act.

Accepting the first chain proposed. There are almost always several possible trace-backs, and the one that comes first is the one that suits the group present. Having two distinct chains proposed before choosing costs five minutes and often changes the conclusion.

Confusing chronological order and causation. What happened before is not necessarily what caused, and the chain looks all the more convincing for telling a coherent story.

On the shop floor — Industrial maintenance

A machine stoppage attributed to a wear part gave, tracing back, a lubrication not carried out, then a maintenance routing that did not mention it, then a routing revision made during an equipment change without review by maintenance. The useful action bore on the routing update procedure, four levels away from the observed symptom.

Frequently asked questions

Do you really have to ask the question five times?

No, the figure is indicative. Some problems deliver an actionable cause after three questions, others require seven. Imposing exactly five levels on oneself leads either to stopping too early, or to adding artificial tiers to reach the count, which weakens the chain instead of strengthening it.

What should be done when several answers are possible to the same why?

It is the sign that the tool is reaching its limit and that one must move to a representation that accepts branches, such as a cause-and-effect diagram. Arbitrarily choosing one of the answers and continuing gives a coherent but partial chain, and the resulting action will treat only part of the phenomenon.

Can the five whys be done alone?

It is possible for a rough cut, but the quality of the result depends on the presence of those who know the actual sequence of events. Alone, one builds the chain with one’s own representations, and stops where one ceases to know — that is, often just before the interesting level.

How do you check that an identified cause is the right one?

By checking each link rather than the conclusion. For each level, a trace is sought: a record, an observation, a direct testimony. A chain whose every link is observed counts as proof; a chain where only the last link is checked can be entirely false while pointing to a cause that really exists.