SKILL.md
RMA Failure Analysis Skill
Raw RMA data lies: it mixes buyer's remorse, user error, and shipping damage in with real design and manufacturing defects. This skill turns returns into decisions — triage first so the Pareto is of verified failure modes, run the top modes through 8D discipline (contain now, root-cause properly, correct permanently), and price each mode in cost-of-quality terms so the fix competes for resources on money, not anecdote.
What This Skill Produces
- A triaged breakdown of returns using a standard RMA taxonomy
- A Pareto of verified failure modes with rates against units shipped
- An 8D-structured analysis for each top failure mode
- Cost-of-quality framing: cost per return, per mode, and the fix ROI
- Prioritised corrective actions with owners and cut-in points
Required Inputs
Ask for these if not provided; analyse whatever slice exists, but state the denominator caveats plainly:
- RMA records — return reasons, dates, symptoms, any teardown/FA findings
- Units shipped per period — the denominator; return counts without it are useless
- Product age mix — manufacture date or batch, to separate infant mortality from wear-out
- Cost inputs — per-return logistics, refurb/scrap cost, support cost per case (estimate and label if unknown)
- Known changes — ECOs, factory or component changes that bracket the data in time
Analysis Framework
Step 1 — Triage taxonomy. Bucket every return before any Pareto:
| Bucket | Meaning |
|---|---|
| NTF / CND | No trouble found — unit passes full test; count separately, it's a UX/expectation signal |
| CID | Customer-induced damage (drop, liquid) — a robustness signal, not a defect |
| OBF / DOA | Failed out of box — points at outgoing quality or transit |
| SW-resolvable | Fixed by update/reset — cheapest class to kill |