What the five whys is
Most problems are first explained by their symptom. Reported sales fell because the purchase tag stopped firing; the campaign missed its target because midweek sold badly; customers complained because deliveries were late. Each explanation is true and none of them is useful, because fixing the symptom leaves whatever produced it in place. The five whys pushes past that first answer by asking why the symptom happened, then why that happened, and so on, so that the fix lands on the cause rather than on its latest effect.
The method is deliberately simple. It needs no software, no statistics and no special training, which is why it turns up in campaign reviews, incident procedures and complaint handling alike. A team can run it in a short meeting, and the output is a short chain of cause and effect that anyone can read and check.
What the method makes visible is the distance between the failure people noticed and the condition that allowed it. A broken tag is noticed; the release process that let an untested change reach the live site is not, until someone asks why the tag broke. What the method hides is everything off the chain. It follows one line of cause, and a problem with several contributing causes will only show the one the team happened to follow.
The chain
| Step | Question | Evidence that supports the answer |
|---|---|---|
| State the problem | What happened, where, when, and how do we know? | The report, the date range and the size of the gap against what was expected |
| First why | What directly caused this result? | A direct observation: a tag test, a log entry, a changed setting, a screenshot |
| Further whys | Why did that cause exist, and why was it not caught? | Change logs, release notes, tickets, messages and the account of the people involved |
| Stopping point | Is this a cause the team can change, and would changing it stop a repeat? | A missing or skipped process, rule or check, confirmed by whoever owns it |
| Countermeasure | What change will stop it happening again, and who owns it? | An owner, a date and a way of verifying that the change holds |
The problem statement deserves more care than it usually gets. "Tracking is broken" is not a problem statement; "the purchase Key EventA GA4 event marked as important to the business, the property's measure of success. recorded almost nothing from the morning of the launch, while the payment system recorded normal sales" is. A precise statement names the KPIA metric tied directly to an objective. or measure affected, the period and the size of the gap, and it rules out explanations early: if the payment system shows normal sales, the problem is in measurement, not in demand.
Each answer in the chain should be a fact that someone has checked, not a plausible story. When an answer cannot be supported yet, mark it as a hypothesis and go and check it before asking the next why. A chain built on guesses produces a confident fix for the wrong cause.
Running a session
Run the session soon after the failure, while logs, messages and memories are fresh, but after the immediate problem is contained. In a tracking failure, for instance, the tracking outage response SOP covers detection, containment and the fix, and requires a root cause before the incident is closed; the five whys is one way to meet that requirement.
Invite the people closest to the work, including those who made the change or handled the complaint, and ask someone not directly involved to run the session. Agree at the start that the chain will stop at processes and checks, not at people. If an answer is "the developer forgot", the next question is why the process relied on nobody forgetting.
Write the chain where everyone can see it, one step at a time. When an answer has more than one plausible cause, draw a branch for each and follow both, rather than choosing the more comfortable one. Stop when the last answer is something the team can change and when changing it would prevent a repeat, then read the chain backwards with "therefore" between the steps to check that each link holds.
Finish with a countermeasure for each root cause: a change to a process, a rule, a check or a tool, with an owner, a date and a way to confirm it has worked. Record the chain and the countermeasures where future teams will find them. The campaign post-mortem template has a root-cause section for exactly this, and the guide to post-campaign analysis for media plans shows how delivery shortfalls are explained before the causes are pursued. For a single error that reached customers, the tutorial on reviewing a marketing error after the fix puts the chain inside a full review, from the timeline to owned actions and an error logOne list across every channel of the errors that reached customers: when found and contained, what customers saw, the checks that missed it, causes and actions. shared across channels.
The same method works on data problems that keep coming back. The guide to marketing data quality lists the usual root causes, from tags to naming, and a short chain of whys usually shows which of them applies. In customer service, the guide to service recovery and complaint handling tags complaints by root cause rather than symptom, and a five whys session on the largest tag turns that review into a change upstream. For everyday support contacts, the tutorial on tagging support contacts by reason keeps the reason agents record separate from the root cause, which a chain of whys then finds for the largest reasons each month.
Worked example (illustrative)
A ticketing platform traces a collapse in reported sales
Quillacre Ticketing, an invented event-ticketing platform, opens sales for a 12-date concert tour at 10:00 on a Tuesday. By 13:00 its analytics shows 310 purchases against a forecast of about 2,500, while the payment system records 2,430 completed orders. Problem: purchases recorded in analytics fell by about 88% from launch morning while real orders were normal. Why? The purchase key event stopped firing at 09:40. Why? A site release at 09:35 renamed the data layer value the tag reads, from "purchase" to "order_complete". Why was it not caught? The release went out without anyone checking the tracking. Why? The release checklist has no tracking step, and the analytics team is not told about releases in advance. The fix is not to blame the developer: the team adds a tracking test to the release checklist, sets an alert when recorded purchases fall below half of payment-system orders for an hour, and agrees a data layer specification that changes only with analytics sign-off. The tag had been repaired the next day, 26 hours after the release.
Limits and criticisms
The method follows one causal path, while most marketing failures have several. A campaign that missed its target may have suffered from a weak plan, late creative, a competitor's promotion and a tracking gap at once, and a single chain will find whichever cause the group pursued first. Branching helps, but for problems with many interacting causes a broader tool, such as a cause map that lists every contributing factor, or an analysis of the data before any questioning, works better.
The answers depend on who is in the room. The chain can only reach causes that someone present knows about or can check, so a session without the developer, the agency or the customer service lead will stop short of causes in their areas. Different groups asking about the same failure can produce different chains.
It can stop at a person rather than a process. "Someone forgot" is the easiest answer to reach and the least useful, because people will always forget sometimes. A chain that ends in blame also teaches people to hide mistakes, which makes the next session less honest.
The number of whys is a convention, not a rule. Treating it as a quota leads teams either to stop too early because they have reached the count or to invent extra links to make the chain long enough. The right length is whatever reaches a changeable cause.
Related frameworks
Many tracking failures are found late because nobody wrote down what should be measured. The measurement plan framework defines the events, the Data LayerA structured object on a web page that passes values to a tag manager for tags to use. values and the owners in advance, which gives a five whys chain a specification to check against. When the problem is a drop in a headline number rather than a single failure, the KPI tree breaks the number into its drivers, so the team knows which branch moved before it starts asking why.
Common mistakes
- A vague problem statement. "Tracking is broken" invites guesses; name the measure, the period and the gap.
- Answers without evidence. Each step should be checked, or marked as a hypothesis and checked before moving on.
- Stopping at the first why. The symptom's immediate cause is rarely the one worth fixing.
- Ending at a person. Ask why the process allowed the mistake, and fix that.
- Following one branch only. Where there are several plausible causes, follow each of them.
- No owner for the fix. A root cause without a countermeasure, an owner and a date will happen again.
