The Same Defect Came Back: Why "We Fixed It" Is Not a Corrective Action

📨 "We Fixed It" Is Not an Answer You Can Ship On

The email is always short. Sorry for the issue, we already replaced the bad units and trained the workers. Next order will be fine. Six weeks later the next order lands, and the same defect is back — sometimes at double the rate, because nothing about the process changed. What changed was the sorting of finished goods.

We see this constantly when we are booked for an Amazon FBA inspection in China on a reordered SKU: the factory genuinely believes it fixed the problem, and the evidence for that belief is a pile of replaced units. Nobody ever asked the question that matters. What in the process produced the defect, and what in the process now prevents it?

🧩 Four Different Words That Factories Use Interchangeably

The vocabulary is where the money leaks:

When a supplier says "fixed", ask which column they are standing in. Nine times out of ten it is the first one. 🔍

📋 The Evidence Chain a Real Corrective Action Leaves Behind

This is what we ask for, in this order, and what we verify on site:

One — a problem statement with numbers. Not "some customers complained" but: 27 of 200 units sampled on 12 September failed the 60-second load test; all 27 trace to production week 36, line 2. Without a date code and a count, nothing downstream can be proven.

Two — a root cause at process level. "Operator was not careful" is not a root cause and it cannot be audited. A usable root cause points at a control: the torque driver had not been calibrated, the fixture pin was worn 0.4 mm, the second-source mould was not in the approved tool list. Five whys should end somewhere a gauge, a drawing or a checklist can be changed.

Three — the escape point. Why did it clear their own outgoing QC? Usually because their sampling plan was thin: run the numbers through the AQL calculator we keep public — with AQL 2.5 and n=200 you accept up to 10 majors, so a 2% defect rate will usually still pass. A factory QC on 32 pieces proves almost nothing. Their escape analysis tells you whether you can trust the next self-check at all.

Four — the change, with a record. A revision number on the fixture, a new check-sheet step, a specified component from an approved supplier, an ECN. Plus the same measurement taken on the affected units before and after, on the same instrument.

Five — verification of effectiveness. This is the step everybody skips. The fix is only proven by the next lots: tighter coverage on the first shipment after the change, and the specific failure mode measured again on the reorder — not merely "passed inspection".

Six — closure with an owner and a date, and a row in your own corrective-action log, so a repeat of the same defect after closure is visible instead of being re-litigated from memory. 📒

🔬 How We Verify a Fix on the Next Two Shipments

A closing report is a claim. Verification is a measurement plan, and it is cheap when it is designed in advance:

Sampling on the reorder stays at the same plan as before — AQL 2.5, level II, n=125 on a 1,201 to 3,200 unit lot — so this lot is comparable with the failed one. We pull units from different cartons and pallets, read the date codes, and check whether the affected cluster is still clustered. The specific failure mode gets a dedicated test at double sampling: if 27 units rattle at 60 seconds, we run 60 seconds on the full sample and log the sound file reference. Dimension claims are checked against the same measuring points and the same instrument, with the gauge in the photograph.

Then we look at the numbers, not the narrative: majors per 100 units against the previous three lots, and whether the same failure mode appears at all. A drop from 6.8% to 1.2% with zero recurrences is a closed corrective action. A drop from 6.8% to 5.1% on the same failure mode is a correction that did not reach the process, and it should go back to the supplier with the photos attached. Our inspectors report back typically in ~24 hours, so the next shipment is still open when the numbers land. ⏱️

💸 The Case Where Nobody Asked: 9.4% Came Back

A 2026 file from our records: 3,900 stainless steel pet water fountains, FOB $12.80. First lot passed pre-shipment at 1.4% majors, but a DPI run pulled 6.8% pump rattle — the impeller touching the housing at speed. The factory replaced 265 pumps, sent photos, and closed it. Because the replacement was a correction, not an accepted corrective action, no control on the pump seat was written, and the pump supplier quietly ran the part on a second mould after the fix.

On the reorder — 4,600 units, Q4 timing — the same rattle came back at 9.4%. 430 units were reworked at the factory, five days came off the calendar, and 200 units had to fly to keep the listing in stock. Cost of the intervention we recommended but was not booked: two man-days at $169 each. That is the whole invoice. The air freight alone was several times it. 🧾

❓ FAQs

What is the difference between a correction and a corrective action?

A correction sorts or replaces the affected units in the lot at hand. A corrective action changes the process that produced them. Replacements leave no evidence about the next order; a process change leaves a control, a record and something to verify.

Is training the operators ever enough?

Rarely on its own. Training is the most common substitute for a root cause, and it is a hope unless the process still stops the mistake. Ask what changed in the fixture, the gauge, the drawing, the approved supplier list or the check sheet, because those can be inspected on the next visit.

What should a factory's 8D or corrective-action report contain?

A named owner, a problem statement with counts, line and date codes, containment of suspect stock, a process-level root cause, an escape analysis against their own sampling plan, the change with a revision or ECN number, agreed verification criteria and a closure date.

How is the fix verified on the next shipment?

Keep the same AQL plan as the failed lot so the numbers are comparable, draw units from different cartons and pallets, test the specific failure mode at double sampling, and benchmark majors per 100 units against the previous three lots.

Can one clean lot close a corrective action?

Rarely. With AQL 2.5 and a sample of 200 you accept up to 10 majors, so a 2% defect rate will usually still pass the plan. Closure is much stronger when two consecutive lots show no recurrence and the same measured values hold.

✅ What to Put in Writing After Any Failed Lot

One: problem statement with unit count, date code, line and test method. Two: root cause that names a control, with a change record. Three: an escape analysis against their own sampling plan. Four: verification criteria agreed before the next shipment leaves — coverage percentage and the specific mode measured. Five: an owner, a date, and a log row you control, not a WeChat promise. When you book the pre-shipment inspection, hand us the corrective-action closed point and we will test that specific failure mode on the next batch instead of a generic pass. Pricing is on our pricing page, questions land at our contact page, and 2,000+ inspections of production from this side of the ocean tell us the same thing every quarter: the second lot is where fixes are proved or lost.