8D problem solving with AI — the eight disciplines fed by live data, not by a template filled in after the fact.
The classic 8D dies in a Word document someone fills in three days later, from memory, to hand to the customer. With iLEAN, all eight disciplines feed on live plant data: D2 starts from the incident captured by photo or voice in the moment; D4 cross-references real batches, curves and shifts from the central memory; D7 updates the standard and the FMEA. An agent drafts the 8D. The person signs each discipline.
The 8D doesn't fail as a method. It fails because it lives in a Word file filled in after the problem is forgotten.
The 8D is one of the best problem-solving disciplines industry has produced: it forces you to contain before correcting, to find the root cause before acting, and to prevent before closing. The problem isn't in the eight disciplines — it's in the medium. And the usual medium is still an office template someone fills in after the fact:
- D2 written from memory — the incident happened Tuesday on the night shift; the report is written Friday. What's left is a vague description ("marks appear on the part") with no batch, no time, no photo and no process condition.
- D4 without a single data point — the Ishikawa is drawn in the meeting room from what people remember. Nobody has gone down to look at the process curves, or at what maintenance was done the week before, or at whether the same raw-material batch shows up in other incidents.
- D6 with no evidence of effectiveness — the "action verified" box gets checked because the problem hasn't reappeared in two weeks. There is no downstream data to back it.
- D7 that never arrives — the prevention discipline is settled with a sentence of good intentions. Neither the work standard nor the FMEA gets touched. That's why the same problem comes back six months later under another 8D number.
- Engineering time badly invested — the quality engineer spends more time writing, formatting and chasing data by email than thinking about the root cause.
The result is familiar: an archive of 8Ds that look impeccable to the customer, and a recurrence rate that doesn't drop. If the 8D isn't fed with plant data at the moment the incident happens, it isn't a problem-solving process: it's a deliverable.
iLEAN doesn't replace the 8D — it puts the whole plant underneath it.
The 8D asks for exactly what a system like IRIS can give: facts, not memories. iLEAN doesn't change the eight disciplines or the order they're worked through; it changes where the information that fills them comes from. Connect is the putty that fills the cracks between the MES, the ERP, the SCADA, the lab and the maintenance report — which is exactly where the evidence a Word-based 8D never incorporates happens to live.
The 8D stops being a document written at the end. It becomes a file that fills itself from minute zero, and that the person reviews and signs discipline by discipline.
The iLEAN pieces applied to the 8D:
- Edge — captures the incident at the moment and at the point where it happens: a photo from the terminal, the operator's voice note, the sensor or machine-vision reading. D2 is born with the real batch, machine, shift and time. It works without a network.
- Connect — brings the context D4 needs and nobody has time to hunt down: process curves, machine parameters, the raw-material batch history, prior maintenance orders, lab results and earlier claims for the same failure mode.
- Agents — the brain. They draft each discipline, propose the D1 team from who was actually on the process, scope the containment in D3, rank the D4 root-cause hypotheses by evidence, and track D6 effectiveness with downstream data instead of a checked box.
- Central memory — every closed 8D feeds the next. When a similar incident comes in, the system retrieves the previous 8Ds for the same failure mode and flags the recurrence before another file is opened from scratch.
- Three safety rings — the agent proposes; the responsible person reviews, corrects and signs each discipline. No discipline moves forward without a human signature.
8D in a Word template vs. 8D fed by live data with iLEAN
| Aspect | 8D in a Word template | 8D with iLEAN |
|---|---|---|
| Origin of D2 (description) | Written from memory days later, no batch, no time | Incident captured by photo or voice in the moment, with batch, machine, shift and time |
| Basis of D4 (root cause) | Ishikawa drawn in a meeting from what the team remembers | Real cross-reference of batches, process curves, shifts and maintenance from the central memory |
| Evidence for D6 (effectiveness) | Box checked; "it hasn't happened again" | Verified against the line's downstream data over the agreed period |
| Scope of D7 (prevention) | A paragraph of intentions; the FMEA and the standard go untouched | Action opened and signed on the living FMEA and on the work standard |
| Recurrence detection | Manual, if someone remembers a similar 8D | Automatic: the system retrieves previous 8Ds for the same failure mode |
| Quality-engineering time per 8D | Mostly writing, formatting and chasing data | Administrative time cut on the order of ≥40% (estimate to be validated) |
Impact estimate for your plant — to be validated with your numbers.
The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.
- Plant opening on the order of 20 to 60 8D reports a year between customer claims and internal incidents, on an office template, with one or two quality engineers as the bottleneck.
- Edge + Agents pilot on incident capture and the 8D flow (plant terminals, MES/ERP and lab integration, central memory). First 8D drafted with live data within a few weeks.
- Indicative payback between 4 and 9 months, resting on two levers: fewer recurrences (the same failure mode coming back because D7 never closed the loop) and less quality-engineering time spent writing and hunting evidence.
- Administrative time per 8D cut on the order of ≥40% (conservative estimate), plus a shorter lead time for delivering the 8D to the customer — which in many plants is itself a contractual commitment.
And the quality engineer's reasonable doubt
“What if the AI invents a root cause?” — hallucination is a problem of free generation, not of anchored tasks. Here the agent doesn't opine about the world: it writes from a specific batch, a specific curve and a specific history, and every D4 hypothesis comes with the evidence that supports it linked alongside. In this kind of anchored task the best models are below 1.5% error [1]. And even so, no discipline closes until the responsible person reviews it and signs: if the hypothesis doesn't hold, it gets discarded the way it always has been — with engineering judgment.
[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.
You may also be interested in: Multichannel incident capture with AI · Digital A3 thinking with AI · DMAIC Six Sigma with AI statistical analysis
What people ask about 8D problem solving with AI
Why does the classic 8D end up dying in a Word template?
Because the template is filled in after the fact: the incident happens on Tuesday, someone reconstructs it from memory on Friday, and the document is closed to be handed to the customer, not to solve the problem. The symptoms are always the same: a vague D2 (the part comes out wrong), a D4 with an Ishikawa drawn in the meeting without a single piece of process data behind it, D5 and D6 with no real evidence of effectiveness, and a D7 that never gets around to touching either the standard or the FMEA. The 8D doesn't fail as a method — the medium fails: a dead document disconnected from the plant.
How is each 8D discipline fed with live plant data?
D1 (team): the system proposes the team from who was on the affected shift, line and process. D2 (description): it starts from the incident captured in the moment by photo or voice from the line itself, with the real batch, machine, shift and time. D3 (containment): the agent scopes the extent by cross-referencing batches and production times. D4 (root cause): it cross-references the real data in the central memory — batches, process curves, parameters, shifts, prior maintenance — and proposes hypotheses ranked by evidence. D5 and D6 (actions and effectiveness): each action is verified against downstream data, not against a checked box. D7 (prevention): it updates the work standard and the FMEA. D8 (closure): the team's recognition and the lesson learned stay in the system's memory.
Who signs each discipline? Does the AI close the 8D on its own?
No. The agent writes the draft of each discipline with the data it already has; the responsible person reviews, corrects and signs discipline by discipline. Without a signature, the discipline doesn't advance and the 8D doesn't close. It's the same principle as iLEAN's three safety rings: the system proposes, the person decides and signs. The gain isn't taking judgment away from the quality engineer — it's freeing the hours spent writing and hunting data so they can be spent thinking about the root cause.
How does the 8D connect to the FMEA and the work standard?
That's the point where the Word-based 8D almost always breaks. In iLEAN, D7 isn't a paragraph of good intentions: when the discipline closes, the system opens the concrete action on the living FMEA (reviewing the associated failure mode, its occurrence and its detection) and on the work standard or affected instruction. Until that action is signed, the 8D stays open. That way the lesson from an incident enters the prevention system instead of sitting in a shared folder nobody ever reopens.
What does the auditor see when they ask for a specific 8D?
They see the complete 8D with something the Word template almost never has behind it: data traceability. Each discipline links to its evidence — the original photo or voice note with its time, the batch, the process curve, the standard version before and after, the FMEA revision, and who signed each step and when. The audit stops being a reconstruction exercise and becomes a lookup. For a scheme like IATF 16949 or ISO 9001, or for a formal customer claim, that trail is exactly what's being asked for.
Tell us your case and in 48h we'll send you the estimated ROI of an 8D fed by live data for your plant.
We work on your plant's real data, not ours. Diagnostic with no commitment.
Request estimated ROI in 48h ‹ See all Lean methods