Why ask "why" five times
The method comes from the Toyota Production System. The idea is simple: when there is a problem you do not ask "how do we solve this" but "why is this happening", and you ask it again about the answer. Until you arrive at something somebody decided and somebody can change.
In marketing it nearly always goes wrong the same way. The first answer sounds like a cause and is not one. "The market is slower." "The algorithm changed." "The creatives are tired." Stop there and you buy new creatives for a tracking problem.
Data first, then the interview
If your problem is a number, look first at where the number moved. Which campaign, which funnel step, since which date. Asking why five times about a number nobody has looked at gives you five guesses.
Make the problem concrete too. Not "marketing isn't working", but: what happened, where, since when, how big.
The rules of the interview
- One question at a time. Ask five whys at once and you get a questionnaire back, and the other person starts guessing.
- Never answer your own why. Not even when you suspect the answer.
- Do not propose the fix during the interview. Once the solution is on the table everybody stops digging and starts defending.
- Five is the starting point, not the ceiling. You stop at bedrock, not at the number.
The three gates
Every answer has to get past these before it becomes a rung.
Gate 1: vagueness. An answer an outsider cannot check is not a cause. "Communication is difficult." "We had no time." The pushback: that's a category, not a cause. Which handover, between whom, and what did it miss?
Gate 2: "that's just how it is." Habit and nature are not causes. "We've always done it this way." "The client is just slow." The pushback: somebody chose that, once. Who set it up, and what were they solving for at the time?
Gate 3: blame. If the answer names a person or a team, you are not deep enough yet. People are the mechanism, never the root. The pushback: take that person out of it. What in the process made forgetting possible, and what would have caught it on a good day?
Does nobody really know the answer? Then that is the finding. "Nobody knows this today" is a rung, and you keep digging in that gap.
Bedrock: when may you stop?
You stop at one of these three:
- A process or system decision somebody made and somebody can unmake: a rule, a default, a template, a handover, a missing check.
- A structural constraint outside your control: a platform limit, a contract, a law. Then the fix targets the workaround, not the wall.
- A measurement gap. "We don't know because we never measured it" is an honest root cause. The fix is the measurement.
You do not stop at a person, a mood, the market, bad luck, or the symptom in other words.
In paid funnels, bedrock often sits in one of these places:
- a release or change process without a tracking check (site, form, consent banner, checkout)
- the handover from marketing to sales: who picks up a lead, how fast, and who reports back what became of it
- the conversion the platform optimizes for versus the outcome the business is paid for
- a budget or approval rule that was made for last year's situation
- a briefing process with no step that feeds results back into the next brief
- a report with platform numbers that nobody reconciles with the CRM or the shop
The output: four blocks
- The symptom, in the words it started with.
- The root cause in one sentence, with the chain underneath so anyone can audit the reasoning.
- One fix. It acts on the root, has an owner with a name, and can start this week. It says what changes in the process, not what people should try harder at.
- One signal after a week. Something you can check in seven days that separates "working" from "not working". A number with a direction and a threshold, a document that exists or does not. Never "it feels smoother".
An example
Made up, but you will recognize the pattern. CPL doubles from mid August. The chain:
| Rung |
Answer |
| Symptom |
CPL from 40 to 80 euros at stable clicks |
| Why 1 |
The platform counts fewer leads. "The market is slower" did not pass gate 2. |
| Why 2 |
Since the new form tool, the conversion event fires on fewer form fills. The CRM shows the same number as before. |
| Why 3 |
The event was not moved along. "The developer forgot" did not pass gate 3. |
| Why 4 |
The release checklist has no tracking check. |
| Bedrock |
The checklist was written by web for web. Nobody owns tracking across the two teams. |
The fix is one line on the release checklist, owned by the web lead. The signal: seven days from now the gap between platform leads and CRM form fills is under 10%. New creatives would have solved nothing here.
The skill
The skill runs this interview. It asks one question at a time, pushes back when an answer fails a gate, and only delivers the four blocks at bedrock.
Download the Five Whys skill (zip)
In Claude Code, copy the folder to ~/.claude/skills/. In the Claude app, upload the zip in the skills settings. The README in the zip covers both, and a full example session is included. No Claude? The method above also works with a colleague and a whiteboard. Agree up front who guards the gates.
When not to use this
- For a one-off incident with no pattern. There is no process to fix.
- When the cause is already known and the question is which fix to pick. That is a decision, not a diagnosis.
- When the answer is a data pull. Pull the data first and use the method on what the data shows.
One next step
Take the problem that already came back twice this quarter. Set aside fifteen minutes and stop only at bedrock.