A DPI report with twenty findings helps only when the buyer knows which three cause refunds. This fifth part of the chargeback series ranks mid-run defects by refund risk, verifies corrections before packing, and feeds return codes back into the next inspection plan. The method pairs sampling discipline with acceptance-test verdicts: fair pulls across the real output, defined criteria from the specification sheet, and a pass, rework, or hold decision while corrections still work. Return economics sets the priorities, so factory effort lands on the defects that actually cost money.
Key takeaways
- Rank every DPI finding by refund risk, multiplying defect rate by customer visibility and return cost, and fix in that order.
- Trustworthy mid-run samples spread across lines, sizes, colors, and hours, applying the representative-sampling idea at the heart of acceptance sampling[1].
- Verdicts should follow acceptance-test discipline with defined criteria, recorded evidence, and a clear pass, rework, or hold per line[2].
- No correction counts until fresh output from the fixed station is re-measured before cartons are sealed.
- Aggregate return codes monthly and convert the top patterns into purchase-order clauses so each inspection starts from a stronger baseline.
Not every defect deserves the same urgency. A scratched hangtag and a dead motor both appear as lines in a DPI report, but only one of them generates chargebacks by the hundred. Buyers who treat all findings alike exhaust the factory's correction capacity on trivia while the refund-makers sail through to the customer. Prioritization is therefore the skill that converts inspection data into fewer returns.
The ranking logic starts from the buyer's own return history. Which complaint codes cost the most in refunds, freight, and listing penalties? Which defects do customers photograph in one-star reviews? A mid-production visit planned around those answers checks refund risk rather than abstract quality, and its corrections land where money is actually lost. Sampling discipline and acceptance-test verdicts supply the method; return economics supply the order.
This fifth part of the chargeback series builds that prioritization system. It shows how to rank findings by refund risk, how to verify corrections before packing, and how to close the loop from return codes back into inspection criteria. Acceptance sampling provides the fairness logic for partial-lot pulls, and factory acceptance testing supplies the pass-or-hold discipline.
Ranking findings by refund risk
A practical risk score needs only three inputs. The first is the defect rate observed mid-run, expressed per hundred pieces so lines can be compared. The second is customer visibility: a dead function or wrong size is noticed by every buyer, while a slightly uneven hem escapes most. The third is return cost, combining refund value, return freight, and any marketplace penalty attached to the complaint code. Multiplying the three gives a ranking that survives arguments about taste.
The ranking changes factory behavior because it speaks money rather than quality points. A seam failure tagged as a certain return with a known refund value concentrates a production manager's attention in a way that a workmanship note never does. Inspection instructions should therefore require the risk tag on every major finding, and the closing meeting should work through findings in risk order rather than page order. Low-risk trivia gets a written note; high-risk refund-makers get same-day containment.
Planning a visit around during production inspection in China makes this ranking concrete from the start, because the instruction sheet can name the buyer's top three historical return codes as priority checkpoints. The inspector then spends the deepest effort where the buyer's money has historically gone. That focus is the difference between a generic quality visit and a chargeback-reduction visit, and it costs nothing extra to specify.
From defect counts to correction order
| Finding example | Observed rate mid-run | Customer visibility | Correction priority |
|---|---|---|---|
| Dead function on arrival | 4 per 100 power-tested[2] | Every buyer notices immediately | Contain lot, quarantine period, repair first |
| Wrong size across a bundle | 9 per 100 measured[1] | Fitting complaints with photos | Re-measure bundle, re-cut, segregate |
| Open seams at stress points | 6 per 100 tug-tested | Fails within first uses | Reset machine tension, re-sew bundle |
| Missing accessory in set | 5 per 100 counted | One-star incomplete-set reviews | Complete sets, add counting sheet |
| Uneven hem finishing | 11 per 100 viewed | Few buyers photograph it | Written note, monitor next visit |
Rate per hundred — visibility weight — return cost — risk-ordered correction
Sampling and verdicts that support the ranking
A risk ranking is only as good as the numbers behind it, and mid-run numbers depend on fair pulls. Representative sampling means spreading pieces across every active line, size, color, and production hour in rough proportion to output, so the examined part fairly reflects the growing whole. This is the central concern of acceptance sampling theory, which studies how conclusions about a larger quantity can be drawn defensibly from an examined part[1]. Staged baskets near the packing table violate the method and flatter the numbers.
Verdicts then apply acceptance-test discipline to each line or style separately. Defined criteria come from the specification sheet and the approved reference; evidence consists of measurement sheets, function logs, and dated photos; the decision reads pass, rework, or hold rather than a vague call for attention. Factory acceptance testing established this pattern for proving requirements are met before goods advance to the next stage[2], and mid-run visits borrow it directly. A hold verdict on a high-risk line protects more revenue than a pass verdict on a trivial one.
Traceability ties the two together. Every sampled piece carries a recorded line, hour, and size so defect clusters map to their source period. When most functional failures come from one evening shift, the correction is a supervisor conversation about that shift rather than a factory-wide campaign. Numbers without trace support ranking; numbers with trace support fixing.
Verifying fixes and feeding returns back in
- Agree the correction at the stationWalk the failing station with the supervisor, demonstrate the defect on current output, and agree the concrete correction with a named owner and an hour. Findings that never reach the station become paperwork.
- Re-verify fresh output before packingPull fresh pieces from the corrected station after the fix and re-measure or re-test them against the same criteria. Unverified corrections drift back within days, especially under overtime pressure.
- Re-count and record the new rateCount flagged defects per hundred again on the re-verified output and record the new rate beside the old one. Improvement you cannot count will not survive the next rush.
- Write the pattern into the next orderTranslate the top patterns into purchase-order language: tighter tolerance, an explicit lot-approval step, or a mandatory hourly check. The next inspection instruction then names these clauses as checkpoints.
- Aggregate return codes monthlyAggregate return codes monthly, rank them by refund cost, and update the inspection priority list every two cycles. The loop keeps factory effort aimed at the defects customers actually punish.
Inspect refund risk, not abstract quality. Rank by rate times visibility times cost, verify every correction on fresh output, and feed the top return codes into the next order. That loop is what turns inspection spending into steadily fewer chargebacks.
Sources
- Wikipedia: Acceptance sampling —— Overview of acceptance sampling including AQL concepts used in product inspection.
- Wikipedia: Factory acceptance test —— Overview of factory acceptance testing performed before shipment or installation.
Frequently asked questions
How do I prioritize when the DPI report lists many defects?
Multiply the defect rate by the likelihood the buyer notices and the cost of the resulting return, then fix in that order. A visible functional failure outranks a minor finishing mark even at equal rates. The report should carry this ranking so the factory works on refunds rather than abstract quality points.
What makes a mid-run sample trustworthy?
Representative sampling means pulls spread across lines, sizes, colors, and hours so the examined part fairly reflects the whole, which is the central idea acceptance sampling theory studies. Staged baskets of neat pieces prove nothing. Traceability of each pull to its line and hour matters as much as the count.
How is factory acceptance logic relevant mid-production?
An acceptance test proves defined requirements are met before goods advance, with recorded evidence and a clear verdict. A DPI visit applies the same discipline mid-run: criteria from the specification sheet, evidence in photos and sheets, and a pass, rework, or hold decision per line. The habit keeps doubtful lots out of the container.
How do I build a feedback loop from returns to inspections?
Tag each defect with the return code it would have become, aggregate monthly, and convert the top patterns into purchase-order clauses and inspection criteria. Share the ranking with the factory so improvement targets refunds rather than scores. Review the trend every two cycles.