Using AI for 8D problem solving: what it drafts, what stays with the team, D2 to D8
In an 8D, an AI assistant can structure the D2 is / is not description, check that the containment covers every suspect part, test the evidence behind the D4 root causes, plan and track the verification of corrective actions, list what must change to prevent recurrence and draft the customer report. It must not name the root cause, choose the corrective actions, lift the containment or close the 8D.
This guide goes through the disciplines one by one and gives a copy-ready prompt for each. The prompts work in Claude in Excel, Copilot in Excel, or with the notes pasted into any assistant your company approves for customer data.
Which disciplines of an 8D can AI help with?
AI helps where the 8D asks for structure, coverage and writing; the team keeps every decision. Discipline by discipline:
- D0–D1, emergency response and team: human decisions, no prompt needed. D0 is the planning and emergency-response step that Ford’s Global 8D added.
- D2, problem description: AI structures the is / is not from your observations and lists the gaps.
- D3, containment: the team decides the actions; AI checks that they cover every place suspect parts can be, and that a clean point is defined.
- D4, root causes: the team proves them; AI sorts occurrence from non-detection (Ford’s escape point) and flags causes without evidence.
- D5, permanent corrective actions: the team chooses them and proves they work; AI maps each one to its cause and drafts the verification plan.
- D6, implementation and validation: the team implements, validates on production and removes the containment; AI compares results with the acceptance criteria.
- D7, prevention: AI lists, as questions, the standards, FMEAs, control plans and similar products to review.
- D8, closure: the team closes the 8D and recognizes the people who did the work.
Prompts for each discipline, D2 to D7
These 8D prompts for an AI assistant use only the notes and data you give them, say what is missing instead of filling it, and never propose a root cause. Replace the parts in [brackets]. The last prompt drafts the report sent to the customer; it is updated as each discipline is completed, often containment first and root cause later.
From the complaint and our first observations pasted below, write the D2 problem description in 'is / is not' form: what, where, when, how many. Fill an 'is not' cell only with what the observations state; otherwise write 'to check: …'. List the data missing to complete it. Remove customer names and personal data from your output. [paste the complaint and observations]
Here are our D2 description and our containment actions: [paste]. Check the containment against the D2: does it cover every place suspect parts can be (our stock, work in progress, in transit, at the customer, at other sites making the same part)? Do our notes state a clean point (first certified lot or date) and how certified parts are identified? For each action, say whether our notes state how its effectiveness is checked; if not, list it as a gap. List all gaps as questions. Do not add actions or checks yourself.
Here is our D4 analysis: [paste]. Check it against the data in this workbook if one is open; otherwise say which data would confirm each cause. Sort the causes we gave into occurrence (why the defect was made) and non-detection, Ford's escape point (why it was not caught before the customer). If our analysis gives no cause on one side, say it is missing; do not supply one. For each cause, state the evidence that supports it, how it was verified, and what would disprove it. Flag any cause stated without evidence. Do not propose a root cause yourself.
Here are the root causes we verified and the permanent corrective actions we chose: [paste]. Map each action to the cause it addresses (occurrence or non-detection) and flag any root cause without an action. For each action, draft how we will prove it works before rollout: what we measure, the sample or duration, and the acceptance criterion. For a non-detection action, include a challenge with known-bad parts. Mark every proposed measure and sample size [TO CONFIRM], and leave the criterion empty where our notes do not give it.
Here are our corrective actions as implemented and the results since: [paste or point to the range]. For each action, compare the results with the acceptance criterion we set in D5, treating empty cells as missing. Say whether the evidence supports removing the containment, and list what is still missing before that decision. Do not decide it yourself.
Our D4 and D5 are pasted below. Draft the D7 as questions: which standards, process FMEAs, control plans or work instructions must be updated, and which other lines, parts or sites share the same cause. Leave the owner of each question empty. [paste D4 and D5]
Turn our working notes for D1 to D7, pasted below, into the customer-facing 8D report: one factual paragraph per discipline completed so far, containment actions with their dates, permanent corrective actions with how their effectiveness was verified. Use roles instead of names unless the customer's format asks for the team list. Mark with [TO CONFIRM] anything our notes do not establish. [paste the notes]
The trap AI makes worse: the plausible root cause
An 8D closed on an unverified root cause comes back as the next complaint. The classic test of a root cause is whether the team can switch the defect on and off with it, where that is safe and feasible; a cause that reads well but was never tested fails that test.
Language models are very good at plausible causes. That is why none of the prompts above asks for one: the D4 prompt checks the evidence of the causes the team found, sorts them into occurrence and non-detection, says when one side is missing, and asks what would disprove each cause.
Customer data in an AI assistant
Paste 8D data only into an assistant your company’s AI policy approves, and anonymize the complaint first if in doubt. An 8D carries customer names, part numbers and sometimes contractual details; the D2 prompt also asks the assistant to keep customer names and personal data out of what it writes, and the report prompt uses roles instead of names unless the customer’s format asks for the team list.
When 8Ds repeat: a signal for the strategy
One 8D deals with one defect that reached the customer. Several 8Ds on the same product family say something the strategy should hear: a quality objective in next year’s X-Matrix, with an initiative and a KPI, rather than a succession of fire-fights.
In hoshin.app, an initiative can point to the 8D or A3 behind it, so a recurring quality problem shows up where the year’s priorities are decided. The other lean prompts, for the X-Matrix, the A3 and the PDCA, are gathered in our guide to AI prompts for lean.
Frequently asked questions
Can ChatGPT or Claude write an 8D report?
They can draft it from the team’s notes: the is / is not, the containment check, the verification plan and the customer-facing report. They should not name the root cause or choose the corrective actions, and the report must only state what the notes establish.
Do automotive customers require an 8D?
Many do: automotive manufacturers commonly ask suppliers for an 8D on formal complaints, often in their own format. The prompts name the disciplines, not the boxes of a form, so they work with any 8D layout.
When should I use an 8D rather than an A3?
Use the 8D when a quality problem needs containment and a documented answer, most often a customer complaint; use the A3 for improvement problems you choose to work on. Our guide to AI prompts for lean compares the A3, the 8D and the PDCA side by side.
What are D0 and D1?
D1 forms the team that will solve the problem. D0, added by Ford’s Global 8D, is the planning step that includes emergency response actions. Both are human decisions; there is no prompt for them on this page.