Five Whys and Business Root-Cause Analysis

“Customers are complaining because the packers need training” already contains a diagnosis and a proposed remedy. It may be right, but repeating it five times does not make it evidence. Start with what happened, trace how it could have happened, and identify what would support or contradict each proposed connection.

This article develops two fictional business cases: damaged grocery deliveries and water-pump defects after a software change. The numbers are illustrative, not measured company results. The exercise is to turn a plausible story into an investigation, an action, and a check of whether that action worked.

What Five Whys can establish

Five Whys is a questioning technique: take a stated problem, ask why it occurred, and use the answer to form a more specific question. Five is a reminder to look beyond the first explanation, not a required chain length or a test that a cause has been proved. ASQ’s guidance explicitly allows fewer or more questions.

Root cause analysis is the broader investigation of how an unwanted outcome arose and which underlying conditions contributed. A root cause is not necessarily one deepest cause shared by every incident. There may be interacting causes and separate paths to failure. Five Whys can organize candidate explanations; observations, records, comparisons, and suitable tests must evaluate them.

Ask people who see different parts of the work: packing, dispatch, customer support, training, and equipment maintenance. The Institute for Healthcare Improvement notes that different perspectives may reveal multiple causes. Agreement in the room is useful for planning the next check, but is not independent confirmation.

Define the symptom before explaining it

For the grocery case, define the symptom as distinct delivered orders with a damage complaint within seven days of delivery. Count each order once, use the same complaint definition, and allow both periods the full seven-day reporting window. In two illustrative cohorts of 1,000 delivered orders each, 16 and 40 orders have such complaints: 1.6% and 4.0%, an increase of 2.4 percentage points. This is a complaint rate, not the damage rate among every delivered item.

First check whether reporting changed. A new complaint button, reclassification of “late” as “damaged,” or a different product mix could change the recorded rate without the proposed packing mechanism. Check delivery dates, duplicate tickets, order coverage, and category definitions before starting a causal chain. Keep examples that do not fit the story, including experienced staff’s damaged orders and new staff’s undamaged orders.

Build a chain with evidence beside every answer

The following is one candidate branch, not a finding. Its purpose is to show how the packaging-and-training explanation can be investigated. Each row depends on the earlier links holding; if a check contradicts an answer, revise or split the branch rather than continuing to a convenient fifth answer.

Question	Candidate answer	Check the connection
Why were damage complaints recorded?	Goods arrived physically damaged.	Read the complaint records and available photos; separate actual damage, late delivery, and duplicate reports.
Why might goods have been damaged?	Fragile items moved inside boxes.	Inspect packing and delivery handling; consider supplier damage, material failure, and route conditions too.
Why might items have moved?	Some packing did not follow a suitable padding rule.	Check whether the rule fits the products, materials were available, and actual work matched the rule.
Why might staff miss that rule?	Some staff packed independently before practical preparation was complete.	Compare assignment dates, observed packing practices, training records, and order types; a missing training record alone is insufficient.
Why could assignment happen before preparation?	The training rollout had no effective readiness check linked to shift assignment.	Inspect the actual handoff and scheduling controls, including exceptions and who could approve them.

Suppose 35% of the packing staff are new hires. That describes workforce composition; it does not show what share of orders they packed or whether their orders had more complaints. If they handle more fragile products or busier shifts, an unadjusted comparison mixes training with those differences. Compare relevant exposure and working conditions, and record what cannot be separated with the available data.

A one-page interim guide may be insufficient, sufficient, or simply not the procedure people actually used. Check its content, task demands, practical demonstrations, supervision, and material availability. Do not end at “HR failed” or “the worker was careless.” A useful explanation names the mechanism and conditions, including how the system allowed or failed to detect the problem.

Choose when to stop, branch, or gather evidence

Stop extending a branch when you reach an unsupported link and collect evidence for it. A candidate cause becomes actionable when it is specific, supported by the observations, and connected to a change that can be evaluated. That is a practical stopping point for this investigation, not proof that nothing deeper exists. A condition being outside the team’s control does not make it non-causal; it may require another owner, a protective control, or a narrower improvement goal.

Ask a counterfactual question: “If this condition were absent, could the same damage still occur through another route?” This can expose an incomplete explanation, but answering it in a meeting is not an experiment. Draw separate branches for packaging, handling, and supplier condition when evidence points to them. For complex or high-consequence failures, use a fuller process and timeline analysis with the relevant specialists rather than forcing every event into one chain.

From suspected cause to corrective action

Keep immediate protection and recurrence prevention distinct. Replacing a damaged order corrects that customer’s immediate problem. Inspecting fragile orders before dispatch contains current exposure. Changing a readiness check or packing control aims to prevent recurrence. None of those actions, by itself, proves that the proposed training cause was correct.

If evidence supports the training branch, a proposed action might require demonstrated packing competence before independent assignment, with a named operations owner and a workable exception process. Define what will be observed after the change: compliance with the readiness check, damage-complaint orders per eligible delivered orders, and possible side effects such as dispatch delays. Attendance at training is an implementation measure, not proof of improved deliveries.

Compare periods with consistent definitions and reporting windows, and examine changes in product mix, materials, routes, and staffing. Where feasible and appropriate, use a planned comparison to distinguish the intervention from concurrent changes. An uncontrolled before-and-after improvement supports monitoring, but cannot by itself isolate the action’s causal effect. Do not withhold necessary protection merely to create a comparison.

A second case: calibration after a software change

A pump manufacturer sees more failed quality checks after a software update. A plausible chain is: changed machine settings, calibration performed with old instructions, and a release handoff that did not deliver the new procedure. Treat each as a claim to test. First distinguish a real pump defect from a changed test threshold or a miscalibrated measuring instrument.

Trace the software version, maintenance time, instruction version, calibration results, and affected production lots. Look for counterexamples: unchanged machines with the same failure, or updated machines that pass. If the instruction mismatch is supported, have the responsible engineers verify the correct procedure, assess affected stock, and strengthen the update handoff so the applicable version and readiness are checked before release. Sending an email is not evidence that calibration was performed correctly.

If the next 200 tested pumps have no observed defects, report “0 defects in 200 tested pumps under the stated test,” not “defects eliminated.” Preserve test coverage, operating conditions, and the follow-up period. The cause statement, action record, and outcome evidence are three separate parts of the conclusion.

Practice: challenge the explanation

Complaint orders rose from 16 of 1,000 deliveries to 40 of 3,000. Has the complaint rate increased? What must be checked before comparing?

One possible response

The count rose, but the rate fell from 1.6% to about 1.33%. Confirm identical definitions, deduplication, complete reporting windows, and comparable coverage or product mix. Neither the count nor the rate alone identifies a cause.

A team writes: “35% of staff were new, so incomplete training caused the damage.” Add one missing comparison and one alternative explanation.

One possible response

Compare complaint rates per relevant order exposure for staff with different preparation, while checking product and shift differences. A packaging-material change could affect new and experienced staff alike. Training records and staff shares alone do not show either damage exposure or the mechanism.

After updated calibration instructions were emailed, 200 pumps passed inspection. Rewrite “communication fixed the root cause and eliminated defects,” and name a follow-up.

One possible response

“Following the instruction update, 200 tested pumps passed the specified inspection; this does not establish zero future defects or isolate the effect of the email.” Verify that the correct calibration was actually performed on the relevant versions, then monitor later lots and other changes. The release control, not the sent-message count, is the proposed prevention mechanism.