Where Does a Supplier Scorecard Get Its Inspection and Audit Evidence?

A scorecard is an argument about capability, and an argument is only as strong as what it rests on. Most arguments a buyer makes about a production partner rest on two or three sources because those are the two or three that arrive without being asked for: the inspection report, the delivery date and whatever the last email said. That is not because the other sources do not exist. It is because nobody wrote down which evidence the programme would collect.

This piece takes each of the eight in turn: what it measures, who produces it, how often it is refreshed, what it can be used to conclude, and how it fails alone. It closes with a default weighting table and three combination rules.

What Counts as Evidence Behind a Scorecard Number?

Four attributes make a field scoreable: dated so it can be placed on a timeline, attributable to the party that produces it, method-bound rather than judgemental, and comparable across lots. A summary without a method is an impression, and an impression has no place in a score. The working test is simple: two analysts reading the same file should produce the same score.

What Does a Pre-Shipment Inspection Result Actually Measure?

One lot, at one moment, drawn under a stated sampling plan. On this programme that is AQL 2.5 under ISO 2859-1 at General Inspection Level II, with the sample size read off the lot size and defects classified against the written class list. What it supports is a statement about that lot. It does not support a statement about the process that built it, and it supports nothing at all about units outside the sample.

Read with the defect counts rather than the verdict alone, the same report supports a much stronger statement, because a trend across lots says something the single outcome cannot. The practical requirement follows from that: the retained report must carry the lot size, the sample size, the accept and reject numbers, the counts by class and the date.

What Does In-Process Checking Add That an End-Point Result Cannot?

Timing, and therefore remedy. The same defect found at a station can be corrected for the remainder of the lot; found at the end point, it is dispositive and the whole lot carries it. An in-process record converts a defect from a verdict into an intervention, which is why it deserves weight in the scorecard even though its sample sizes are small and its method is looser.

It also measures a different object. An end-point result measures units; an in-process record measures whether the controls were in place during the build: the parameter set recorded at the machine, the reference unit present at the station, the operator working to the current version of the specification. That is closer to capability than any attribute count, and it is the only one of the eight that can be acted on inside the window. Its failure mode is informality: a verbal check cannot be scored, because it has no denominator and no definition of what was examined.

What Weight Can a Third-Party Laboratory Report Carry?

The highest of the eight on reproducibility, and the narrowest on scope. A laboratory report states the submitting party, the sample description, the method, the result, the limit it was judged against and the date, and any reader can re-read those six things and reach the same conclusion. It therefore carries real weight where the question is whether a material met a stated limit.

Its limit is that it describes what was submitted. A report on one fabric lot, in one colourway, does not cover another lot or another colourway, and dye lots within the same colour differ. The rule that makes laboratory data scoreable is therefore about attachment: the sample description on the report must tie to the material lot used in the order, through the roll or lot number recorded at receipt. Where it does not tie, the report is evidence of something, but not of the goods. Its failure mode is coverage drift: one current report treated as covering every colourway and every later lot. Recording two fields per report, the material lot covered and the interval before the next test, converts that drift into a routine line item.

What Can a Third-Party Audit Report Support?

A statement about a system rather than about a lot: whether the controls described in it exist, whether non-conformities raised at the previous cycle were closed, and whether the closure evidence holds up. The report carries a scope, an auditor identity and credentials, a classification of findings, and a stated validity period, and every one of those limits what can be concluded from it.

Social findings inside such a report are read one way only. The findings are assessed by a qualified third party against the law of the production country and against the buyer code of conduct, and the programme records that assessment as a governance fact rather than as a scored lot attribute. Where the buyer needs to act on a social finding, the action belongs in the corrective action file, not in the monthly score. An audit report should therefore carry modest weight in a monthly score and substantial weight in an annual review, because it does not change in between. Scoring it every month is the most common double-count in this field: the same audit counted twelve times.

How Should Delivery Records Be Scored Against the Agreed Window?

Scored against the trigger, not against the order date, and against the specific event the window runs from. Bulk production runs 35-50 days from written sample approval, and sampling runs 6-10 working days from a signed-off tech pack. Those are two different triggers and conflating them produces a variance that belongs to nobody.

Four fields per lot make the record usable: the date the trigger fired, the agreed goods-ready date, the actual goods-ready date, and the variance with a cause noted. The fourth field is the one usually left out and the reason a delivery score cannot be acted on. A late lot caused by a late buyer approval and a late lot caused by a production shortfall are different events that look identical in a percentage. The remedy is to score net of buyer-side delay while recording both figures.

What Do Sample-Round Counts Tell You?

How well the requirement was understood the first time. A count runs from first submission to written approval, with the interval per round recorded, and on this programme each round sits inside the 6-10 working day sampling window, longer at 12-15 for complex constructions. It measures clarity at the specification stage and responsiveness on both sides.

It is also the source most easily read backwards. A round count of one can mean an unambiguous specification or an approver who signed whatever arrived, and those are opposite conclusions from the same number. The field that separates them is what changed afterwards: where the specification was revised within two weeks of approval, the low count described an uncritical approval rather than a clear document. Adding a reason code per round makes the field scoreable: requirement unclear, buyer-side change, material substitution, construction not feasible as drawn.

What Belongs in a Corrective Action Record?

Seven fields, and the last two decide whether it can be scored. The finding reference, the root cause statement, the action, the named owner, the due date, the evidence of closure, and the date on which somebody verified it. Without closure evidence and a verification date the item is open whatever the calendar says, and the only defensible definition of closure is verification rather than assertion.

The scorecard derived from it should be a closure rate and a closure cycle time, both measured only against verified items. Counting items opened rewards activity; counting items closed rewards completion, and adding the median days to closure discourages the pattern where every action is closed in the week of the next audit and nothing in between. One further derived field earns its place: the recurrence rate of findings on the same clause across consecutive audits, which is the strongest available predictor of whether the next cycle produces the same conversation.

Why Is Buyer-Side Goods-In Data the Strongest Source?

Because the buyer produces it, on the buyer side method, at the point goods enter the buyer side operation. It measures what arrived rather than what was reported: count received against count shipped, carton condition, labelling and marking conformance, damage found at unsealing, and the result of any receipt inspection the buyer runs. Nothing in that list has to be taken on trust.

It has two weaknesses and both are manageable. It arrives late, weeks or months after the lot closed, which means it cannot drive an intervention in that lot, only the next one. And it mixes supplier-caused with carrier-caused damage, so it needs a split: damage photographed at unsealing can be attributed, damage discovered days into storage cannot, and unattributed damage should be recorded as unattributed rather than assigned.

The dominant failure mode is absence. Where goods-in records nothing per lot, the strongest source in the entire file is permanently missing, and no amount of work on the other seven recovers it. Two minutes per receipt recording count, condition and exceptions is the highest-value data-collection step in the whole framework.

How Often Should Each Evidence Source Be Refreshed?

Each has a natural cadence and a staleness limit, the age at which it stops counting. Inspection is per lot and stale when the next lot closes without it. Laboratory is per material lot and stale when the programme interval passes. Audit runs to the scheme validity period. Delivery, sample rounds and goods-in are per lot or per style. Writing the staleness limit next to the weight is what stops a scorecard drifting into a museum: a field carrying 10 per cent on data fourteen months old is not evidence, and the honest treatment is either to refresh it or to drop it from the score for that period and say so.

How Should Weights Be Set Across the Eight Sources?

Start from the table above and change it for one stated reason at a time. A programme where the dominant risk is chemical conformity carries more laboratory weight; one where the dominant risk is launch timing carries more delivery weight. The default exists so that every deviation has to be justified in writing. Two constraints keep the exercise honest: no single source above 30 per cent, and weights set before the quarter begins rather than after the results are in.

How Do You Combine Sources Without Counting One Event Twice?

Three rules, applied in order. First, one event is entered once, under the source closest to it: a defect found at in-process and again at the final inspection is one event, not two defects across two sources. Second, an action that fixes a finding does not score as good performance elsewhere; the corrective action file records the repair, and the affected lot keeps its original result.

Third, a document does not score twice because it appears in two places. An audit report is filed in the governance record and referenced in the review; it scores once, in the review. The discipline that makes all three rules work is a single register of scored events, with one line each and a cross-reference rather than a second entry.

1. Evidence, not impressions

Dated, attributable, method-bound and comparable

2. Counts, not verdicts

Defect counts by class rather than a pass line

3. Trigger, not order date

Variance measured from the event the window runs from

4. Closure, not activity

Verified closures only, with recurrence tracked

5. Attachment, not existence

A laboratory result must tie to the material lot

The commercial frame sits underneath every one of the eight: MOQ 500 pcs per style and colourway, sampling 6-10 working days from a signed-off tech pack, bulk production 35-50 days from written sample approval, inspection at AQL 2.5 under ISO 2859-1 General Inspection Level II, T/T 30/70, quoted FOB Xiamen with CIF and DDP available. Where a source cannot be read against those figures it usually cannot be read against anything.

Frequently Asked Questions

People Also Ask