Turn a Process Failure into a Five-Whys Investigation with ChatGPT

A process failure can produce an immediate explanation that feels convincing: someone forgot a step, a review took too long, or a message was missed. An investigation should examine what the evidence supports before turning that explanation into the final cause. Repeated why questions can help organize inquiry, but a fluent chain is not proof.

This guide offers an original five-whys worksheet using ChatGPT for a fictional editorial process failure. It is an adaptable investigation aid, not a validated diagnosis or a statement of a formal quality standard. Current OpenAI documentation was checked on October 6, 2026. The process owner must review evidence, alternative explanations, and any proposed corrective action.

Write a factual problem statement

Describe the observed failure, affected process, relevant period, and known consequence. “A reviewed handout was circulated after the planned workshop” is a bounded observation if records support it. “The reviewer caused the workshop failure” combines an event with an unverified causal judgment.

Identify the expected process separately from what occurred. Use the approved process description and dated records. If the normal sequence is undocumented, mark that limitation before analyzing departures from it.

Keep blame and motives out of the problem statement. A person may be involved in the event without being the cause of the failure. Start with observable actions and conditions so the investigation can examine the system rather than defend an initial accusation.

Assemble the evidence packet

Collect the relevant task history, messages, document versions, timestamps, and approved process instructions. Assign source IDs and preserve the distinction between an event date and the date it was reported. Mark incomplete exports or missing attachments explicitly.

The current OpenAI prompting reference discusses supplied context and checking uncertain information. In this worksheet, ChatGPT organizes questions and candidate explanations; the underlying process records provide the evidence for any causal claim.

Separate observations from retrospective interpretations. A later account can suggest a useful hypothesis, but it should not replace a contemporaneous record without review. Keep both sources and explain their relationship when they differ.

Create the question-chain worksheet

Use five initial why rows as a practical scaffold. Do not force the investigation to stop at the fifth row or fill every row when evidence runs out. The worksheet should expose an unanswered question rather than fabricate a complete chain.

Allow branching where several conditions may contribute. A missed review can involve an unclear assignment and an unavailable source. Forcing a single linear story may conceal the relationship between those conditions.

Ask ChatGPT for questions before conclusions

Supply the factual problem statement, expected process, evidence packet, and worksheet schema. Require the model to distinguish a proposed explanation from a supported finding and to identify missing evidence.

Draft a five-whys investigation worksheet for the supplied process failure.
Use five initial question rows, but do not force answers when evidence is missing.
Separate observed facts, candidate explanations, and reviewed findings.
Cite source IDs and preserve contrary evidence and alternative explanations.
Allow branches when several conditions may contribute.
Do not infer motives, assign blame, or claim a validated root cause.
For each unresolved explanation propose a specific evidence check.

Inspect whether each question follows from the preceding evidence. A model can leap from a delayed document to a claim about poor management without a supporting record. Replace that leap with an explicit question or remove it from the evidence chain.

Work through a fictional investigation

Suppose a fictional handout was circulated late. Records show that the draft reached a review inbox after the planned review slot. The first question can ask why submission occurred after that slot. The evidence establishes timing, not yet the reason for it.

A candidate explanation might be that a required source arrived late. Check the source request and receipt records before accepting it. Another candidate could be that the writer did not know the review cutoff. That requires evidence about the communicated schedule and assignment.

If both conditions are supported, represent both. The worksheet should not choose one merely because it makes a cleaner story. If the packet contains no evidence for either, record the unresolved question and identify who can supply the relevant records.

Test explanations against alternatives

Ask what observation would weaken each candidate explanation. If the source arrived before drafting began, that may contradict the claim that its lateness caused the submission delay. If the cutoff was clearly communicated and acknowledged, examine another explanation rather than repeating the same claim.

Check whether the explanation accounts for the specific event or only describes a general concern. “The process is complicated” may be true as an opinion but does not establish why this handout was late. A useful finding connects a documented condition to the observed failure within the review scope.

Keep correlation and cause distinct. Two events occurring close together do not alone establish that one produced the other. The reviewer should identify the evidence supporting the connection and retain any unresolved alternative.

Link corrective actions to reviewed findings

Once the owner reviews the findings, propose actions addressing the supported conditions. If assignment ambiguity is established, a clearer assignment record may be relevant. If source timing remains unresolved, do not present a new reminder system as a proven solution.

Separate immediate containment from longer-term process changes. Reissuing the correct handout may address the current consequence, while a revised review schedule may address a reviewed contributing condition. Record the purpose and owner of each proposed action.

Define how the team will check whether an approved change helps. Use actual follow-up evidence rather than declare the problem solved because a new checklist exists. The guide has not run such a check; the worksheet should record its planned method and later result separately.

Preserve the investigation record

Keep the problem statement, evidence packet, question branches, review decisions, and action status together. If new records change a finding, retain the earlier interpretation and explain the revision. This allows future reviewers to understand why the explanation developed.

Use the project timeline guide when chronology is unclear, and the assumption register guide for proposed actions that rely on unconfirmed conditions. These supporting records help keep evidence, causal hypotheses, and implementation assumptions distinct.

Close the investigation only under the owner's reviewed criteria. An unanswered question may remain open if the necessary evidence is unavailable. A transparent limit is more useful than a confident conclusion assembled to fill every why row.

Frequently asked questions