AI can help a manufacturer write an 8D report by drafting each step, D1 to D8, from the complaint, inspection records and notes your team supplies. It structures the evidence, finds gaps and proposes questions. Your team decides the root cause from tests and data, because a cause the AI guesses is worth nothing to the customer.
An 8D is a response format with eight numbered disciplines: team, problem description, interim containment, root cause (including why the defect escaped), choosing permanent corrective actions, implementing and validating them, preventing recurrence, and recognising the team. Some customers add a planning step, D0. It goes wrong in two ways. Either it is late, because the team was gathering evidence while the clock ran, or it is wrong, because someone wrote a plausible root cause to fill the box. AI helps with the first problem and can easily cause the second, so every step below starts from an evidence folder.
The rules that keep an AI-assisted 8D honest
- Evidence first. Only material in the evidence folder can support a statement. Every paragraph the AI drafts cites a file name from that folder.
- Causes are hypotheses until tested. The AI may list possible causes, always labelled as such. The report's root cause is the one the team has confirmed by a reproduction, measurement or record.
- No invented numbers. Sample sizes, dates, quantities and measurements come from your records, never from the model.
- A named person owns each step. Names go in the report header, and the quality lead signs the whole thing.
- The customer's format and deadlines win. If their portal or form differs from D1 to D8, copy your content across; do not argue about layout.
If you already read complaint trends across many customers, the tutorial on spotting patterns in complaints and defects covers that side. Here the job is answering one complaint well.
Main example: cracked mounting bosses on a plastic housing
Illustrative case, with invented details. A small injection-moulding company supplies a plastic control-box housing to an equipment maker. On a Monday the customer's quality contact emails: 14 of 300 housings from batch 4417 have cracked screw bosses at assembly; they have quarantined the batch and want an 8D. Their supplier terms ask for containment confirmation within 24 hours and the full report within 20 working days.
Step 0: build the evidence folder and the header
Before any drafting, collect everything into one folder: the complaint email and photos, the customer's drawing and revision, the batch traveller (production record), moulding machine settings for the run, material lot certificate, inspection records, gauge IDs, and stock counts for that part. Name each file clearly (E01-complaint-email, E02-photos, E03-traveller). Then ask the AI to build a header and a gap list:
I am the quality manager at a small plastics moulder. Below is a customer
complaint and a list of files in our evidence folder. Draft the 8D header
(customer, part number and revision, batch, quantity affected, complaint date,
interim response due, full report due, owner for each D) and a table of
deadlines from the customer's terms. Then list what evidence is missing for
D2 (problem description) and D3 (containment). Do not guess any value that is
not in the material.
[paste complaint text; paste file list; paste the relevant terms]Illustrative output: header filled with the dates from the terms (interim due Tuesday, full report due 20 working days from the complaint), and a missing-evidence list: "No photo showing crack location against the drawing; no record of which cavity produced the cracked parts; no stock count of batch 4417 or neighbouring batches at your site or in transit." Those three gaps shape the first 24 hours. Send nothing to the customer from the AI draft until a person has checked it against the folder.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowD1: the team
List the people with the knowledge and the authority to change things: quality manager (lead), moulding supervisor, toolroom technician, materials buyer and a customer-facing account owner. Add the customer's contact as the reporting line. AI adds nothing here except a table of roles, a meeting schedule and who answers which question; keep it brief. A team of three to six works best.
D2: describe the problem, precisely
A good problem description says what, where, when, how many, and what is not wrong. Use the "is / is not" table, and ask the AI to draft it from facts and flag blanks:
| Question | Is | Is not | Evidence |
|---|---|---|---|
| What | Cracked screw bosses on housing, part number and revision from the drawing | Other features; other part numbers | E02 photos, E04 drawing |
| Where on the part | Two of four bosses, both on the same side | The two opposite bosses | E02 |
| When found | At customer assembly, during screw fit | At our final inspection | E01 |
| How many | 14 of 300 (4.7%) | Not yet known for other batches | E01 |
| Which batch | Batch 4417, one material lot | Earlier batch 4409 (no complaint so far) | E03, E06 |
The sum is simple: 14 / 300 is 4.67%, written 4.7%. The useful detail is the "is not": two of four bosses, on one side only, points towards a tool cavity or gate feature, but that is an inference to test and not a finding. A line saying "likely a cavity problem" does not belong in D2. Write facts only; the AI's job is to keep it that way, and your job is to read the D2 draft for any word like "caused", "due to" or "because".
D3: interim containment, within the customer's deadline
Containment protects the customer now and stops more bad parts leaving. Think in layers: the customer's stock (they hold the batch), your stock, material in transit and the process itself. For the example:
- Customer batch 4417 is quarantined by the customer; agree whether you sort on their site, replace, or take the batch back.
- Your warehouse: quarantine any remaining parts from 4417 and 4409 and the next batch; 100% visual and screw-fit check on a sample bench until the cause is known.
- Production: add a boss check to the first-off and hourly check on the running tool; record every result.
- Certify the new shipments: mark containment-checked parts with a coloured label so the customer can tell old from new.
Ask the AI to draft the containment email to the customer from your actual actions: "We have quarantined X parts at our site and Y in transit, started 100% inspection on [date and time], and will ship replacements marked with a green label on [date]". The Tuesday response goes out on time; the full cause analysis comes later. Do not write that containment is complete until someone has seen the quarantine tags.
D4: root cause, where the AI must stay in its lane
This is the step that is most often written to fill the form. Use two columns: possible causes (hypotheses) and the test that would confirm or rule each one out. Ask the AI to build the list from the is/is-not table and process data, and to make the tests concrete:
From the is/is-not table and the process records below, list possible causes
of the cracked bosses, grouped as Material, Machine, Method, Man, Measurement,
Environment. For each, give: why the evidence points towards or away from it,
and one test or record that would confirm or rule it out. Label every item
HYPOTHESIS. Do not state any cause as the root cause. Do not use any
information that is not in the data supplied.Illustrative output: Material, "Lot differs from the previous batch; test: check certificate and run a moisture test on retained pellets." Machine, "Packing pressure changed at hour three per the traveller; test: compare the cracked parts' cavity numbers with the time of the change." Method, "Boss cracking at screw-fit can follow a low-cavity-fill or a weld line; test: section two cracked bosses." Measurement, "Torque setting at the customer's assembly may differ from the drawing; test: ask for their torque tool record." The tests are what the team does over the next week. The AI's list is a prompt for the meeting, and half of it may be wrong.
Now the team runs the tests. Illustrative result: the cracked parts all come from cavity 3, one cavity of four; the cavity's gate shows wear; the moisture test is normal; the customer's torque tool is within spec. The root cause for the report is supported by tests: "Gate wear on cavity 3 caused short-shot weakness at the boss, confirmed by sectioning (E09), cavity traceability (E10) and a reproduced crack when the same tool was run with the wear unchanged (E11)." Only now does the sentence enter the report. A "5 whys" chain written afterwards, with a record for each step, explains why the wear went unnoticed, and is the escape point: no cavity-level check at first-off.
D5: choose and prove corrective actions
The permanent action addresses both the cause and the escape point. For the example: repair or replace the cavity 3 gate insert, and add a cavity-level boss inspection at first-off and every shift change. List two or three options with pros, cons and cost, and have the team choose; AI can set out the options table quickly, but cost figures come from the toolroom's quote, not from the model.
D6: implement and verify
Implementation needs a date, an owner and proof. Proof for the example: ten consecutive shots from the repaired cavity sectioned and measured, 300 consecutive parts screw-fit checked at a sample bench with zero cracks, and a pass in the customer's own assembly on the first replacement lot. Record results in a table with dates; ask the AI to format the table from your numbers, then check them against the originals.
D7: prevent recurrence
Look sideways. Which other tools show gate wear? Which other parts rely on a boss feature? Which control plan, work instruction or preventive-maintenance task needs an update? Illustrative actions: add gate-wear inspection to the preventive-maintenance schedule for all multi-cavity tools, update the control plan to include a cavity-level boss check, and revise the work instruction. Update the documents, and keep the revision numbers in the report. If the change alters a drawing or bill of materials, see the engineering change order tutorial.
D8: close the loop
Recognise the team in a sentence, confirm the customer accepts the report and archive the folder. Add a lessons-learned line for other sites or products. Keep the folder for as long as your quality system or the customer requires.
The "why" chain for the escape point, filled in
The root cause explains why the defect was made; the escape point explains why it was not caught. Write both, and keep the "why" chain short and sourced. For the housing example, an illustrative chain: why did cracked parts reach the customer? Because first-off checks looked at one boss per part, on the side that was fine. Why? Because the control plan lists "boss present, no visible damage" without naming the position. Why? Because the plan was copied from an earlier two-boss part. Why was it not reviewed? Because tool changes did not trigger a control-plan review. Each answer has a document behind it, and the last one becomes a D7 action. Stop when you reach something the team can change; do not go on to "people are busy".
Ask the AI to turn rough meeting notes into this chain and to underline any link with no record behind it. Those underlined links are the ones to verify before the report goes out.
The cover email that goes with the report
A short, factual email helps the customer's quality contact find what they need. Illustrative text to adapt: "Hello [first name], attached is our 8D response for batch 4417. In short: cracks originate at cavity 3, where gate wear reduced fill at the screw bosses; we have repaired the insert, added a cavity-level check at first-off and every shift change, and verified 300 consecutive parts with no cracks. Replacement parts shipped on [date] carry a green label. Our effectiveness check on the next three batches will be sent on [date]. Please tell us if you need anything else in a different format. [your name]". Ask the AI to draft it from the finished report, then check that each claim matches D4 to D6, since the email is the part everyone reads.
Deadlines: how to stay ahead of the customer's clock
Customers set clocks that look generous until evidence takes longer than expected. Put the dates in a table at the top of the report and in your own task list on day one. To show the kind of numbers you will meet, the middle column below comes from two large equipment makers' published supplier procedures, checked in November 2026; one of them says every timeline can be adjusted by the customer's quality engineer. Copy your own customer's numbers in their place:
| Milestone | Example timing from published procedures | What goes out |
|---|---|---|
| Acknowledgement | Not always written down; same or next working day is good practice | Short email: received, owner named, containment starting |
| Containment (D3) | Within 24 hours of notification, in both procedures | Containment actions and dates |
| Root cause (D4) | 15 calendar days in one procedure | Confirmed cause, tests and results |
| Corrective action in place (D6) | 30 calendar days in the same procedure; the other sets 10, 20 or 30 days by priority | Implementation and verification evidence |
| Closure (D8) | 45 calendar days in the first procedure | Complete report with evidence |
| Effectiveness check | Agree it with the customer, for example the next three batches | Short follow-up with data |
If a root-cause test needs a tool to be pulled or a lab to section parts, ask the customer for a dated extension in writing as soon as you know, not on the due date. Ask the AI to draft the request from the facts: "Sectioning is booked for Thursday; results Friday; request full report on [date]".
Where AI-drafted 8Ds fail, and a one-machine shop's version
An invented root cause that sounded right
Say an assistant, asked to "complete D4", writes "operator error during assembly". Nobody has tested that, and the customer rejects the report. The fix is a rule in the prompt ("do not state any cause as the root cause") and a reviewer who checks every causal verb against a file name. Reports that blame an operator, a supplier or the customer without evidence are the quickest way to a second complaint.
Containment that was only a promise
A draft says "100% inspection in place" while the sorting bench is not yet set up. The customer visits and finds nothing. Before sending, the quality manager walks to the bench, takes a photo and attaches it. Tone and wording can come from the AI, but the facts come from a person who was standing there.
Wrong sample, wrong conclusion
A team measures 10 parts from the end of the shift and finds all within tolerance. The cracked parts came from the start of the run. Ask the AI to review the sampling plan in D4: "Does the sample match when and where the failures occurred?" This catches a mismatch that the team, tired and under time pressure, may skip.
A smaller supplier with one machine and no quality department
A four-person machine shop gets a complaint about an oversized bore on one of 50 parts. There is no formal team; the owner is D1. The report is shorter, but each step still has evidence: gauge ID and calibration date (see the calibration log tutorial), the program offset history and a check of the tool wear record. AI helps the owner turn rough notes into the report format in an hour instead of an evening.
A repeat complaint six months later
The same defect on a related part returns. Search the archive: give the AI the earlier 8D and the new complaint and ask "where does the new complaint differ from the earlier root cause, and which D7 actions were not applied?". If D7 was never completed on the related part, the report says so, with the corrective action.
A review checklist before sending the report
- Every statement of fact has a file reference that exists in the evidence folder.
- Numbers in the report match the records, including quantities, dates and measurements.
- D2 contains no cause words, and D4 contains tests with results, not only hypotheses.
- D3 actions have been physically seen by someone who signs.
- D5 and D6 name owners and dates, and the verification data is attached.
- D7 lists the documents changed and their new revision numbers.
- The report uses the customer's format, names and part numbers exactly.
- One senior person has read the whole report and signed it.
For the quality system around it, the small manufacturers' quality control tutorial shows where complaints, inspection and corrective action fit together, and the ISO 9001 documentation tutorial covers writing the procedure that governs your corrective actions.
Confidential drawings and customer data
Complaints often arrive with drawings, specifications and part numbers that the customer treats as confidential. Check your agreement before putting them into any AI tool. Where in doubt, describe the problem in your own words, remove drawing content and customer names, and paste only the text of the report. Use a business plan that does not train on your content by default, and keep the evidence folder in your own storage. The AI drafts and organises; the people who made the part decide what happened to it.
Questions about 8D reports and AI
Can AI find the root cause of a defect?
It can suggest hypotheses and questions to test, drawn from the evidence you supply, but it cannot know your process. A root cause needs a test or measurement behind it. Treat any AI-suggested cause as a hypothesis until someone has reproduced or ruled it out on your own line.
How long does a customer give you for an 8D?
It varies by customer. Two large manufacturers' published supplier procedures both ask for containment within 24 hours, and one sets root cause at 15 calendar days and closure at 45. Your own deadlines sit in the customer's supplier quality manual or purchase terms. Read those first, put the dates in the report header, and ask for an extension in writing if evidence will take longer.
Is an 8D the same as a corrective action report?
They overlap. An 8D is a structured format for a corrective action response, with eight numbered steps, and some customers insist on it. Others accept their own form or a plain corrective action report. Use the format the customer names.
Should the report admit fault?
State facts and evidence. If your inspection missed a defect, say so plainly in the escape point section. Customers trust a report that names where the process failed more than one that blames the customer's use, unless the evidence supports that explanation.
Further reads
- 5S Audit Checklist: Build One for Your Shop Floor With AI — Shop-floor organisation that prevents repeat escapes.
- Can AI Handle Customer Complaints Without Making Them Worse? — Where AI helps and hurts with customer complaints.
- Small Factory CMMS With AI: Does the AI Layer Earn Its Cost? — Tool wear and maintenance records that feed D4 and D7.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ASQ's eight disciplines (8D) overview for the D0 to D8 structure; Graco's supplier corrective action request page and Oshkosh Corporation's 8D corrective action procedure for example deadlines (checked November 2026). Your deadlines come from your own customer's supplier requirements.