Guide
What should a supplier scorecard measure for a custom metal parts supplier?
A supplier scorecard is a short, regularly updated set of measures that tells a buyer whether a supplier is delivering conforming parts on the agreed date, and whether that is getting better or worse. For a custom metal parts supplier it should measure delivery to the confirmed date, lot acceptance and repeat nonconformances, first article first-pass results, how fast and how well corrective actions close, document accuracy and responsiveness, with thresholds and escalation steps agreed in advance, so drift is visible before it reaches the assembly line.
By Saola Manufacturing engineering team · Published · Updated
Most published scorecard templates come from software vendors and high-volume automotive programs. They weight on-time delivery and defective parts per million, add a cost line, and stop there. That works when a supplier ships tens of thousands of identical parts a month. It works badly for a buyer of machined, cast, stamped and finished parts who receives a few dozen lots a year, each in small quantities and each with its own drawing. This guide is for the supplier quality engineer or buyer in that position: which metrics still carry signal at low volume, how to calculate them so the supplier cannot argue about the number, how to read them without being fooled by small samples, and what to write into the purchase order and quality agreement so the scorecard can actually be filled in.
On this page
What does supplier drift look like on a custom metal parts account?
Supplier drift is a slow slide in delivery or quality that no single shipment makes obvious: a few more late lines each quarter, the same characteristic rejected again, certificates that need chasing, corrective actions that stay open. It usually becomes visible only when it stops a line or forces an expedite.
The early signs are rarely dramatic. A supplier starts confirming ship dates and then re-promising them a week before they fall due. Receiving inspection begins to sort the occasional lot rather than reject it, because the parts are needed. A burr or a thread gauge failure on one feature turns up on two different part numbers. Material certificates arrive after the parts, or name a heat number that does not match the tag. Answers to engineering questions take longer. None of these events alone triggers a formal complaint, and that is exactly why they need to be counted.
The automotive quality standard treats this as a monitoring obligation, not a courtesy. Clause 8.4.2.4 of IATF 16949:2016 requires the buying organization to monitor supplier performance with at least four indicators: conformity of delivered product, disruptions at the receiving plant (including yard holds and stop-ships), delivery schedule performance and the number of premium freight occurrences. The general quality-management standard used across industries, clause 8.4.1 of the 2015 edition, carries the same idea: the organization sets and applies criteria to evaluate, select, monitor and re-evaluate external providers, and keeps records of the results. Even a buyer who works to neither standard can borrow the logic: decide the measures before the problems, and review them on a fixed rhythm.
Why do generic supplier scorecards miss drift on low-volume parts?
The most common reason is that the metrics were designed for high volume: one rejected part in a small lot swings a PPM figure wildly, so people stop trusting it. Close behind are delivery measured against a moving date, no count of repeat problems, and scores that are reviewed too rarely to catch a trend.
Ranked by how often they hide drift on custom parts accounts, the causes are these:
- Volume-based quality metrics. Defective parts per million divides rejected parts by received parts. With one reject in a lot of forty parts the figure is twenty-five thousand PPM, and with none it is zero, so the number jumps around and tells you little about the trend. Count-based and lot-based measures behave better at low volume.
- Delivery measured to the latest promise. If on-time delivery is calculated against the most recent re-promised date, a supplier that slips every order but warns you in advance scores perfectly. Measure against the first confirmed date and record re-promises separately.
- No memory of repeats. A scorecard that only totals nonconformance reports treats a new problem and the third return of an old one as equal. The repeat is the sign that a corrective action did not work.
- Missing soft signals. Late certificates, wrong revisions on inspection reports and slow answers cost real time but never appear in delivery or defect counts.
- Reviewing once a year. An annual review finds drift after it has already hurt. A monthly data update with a quarterly review is a common rhythm in practice.
A scorecard for low-volume parts therefore swaps some ratio metrics for counts and ages, measures delivery against a fixed reference, and treats repetition as its own signal.
Which metrics should the scorecard measure, and how is each calculated?
Use six to ten metrics in three groups: delivery, product quality and the quality system around the product. Write the formula for each one into the quality agreement so the supplier calculates the same number you do.
The table lists the measures that keep their meaning when lots are small and part numbers change often. The usual template headings (delivery, PPM, corrective actions) are still there, but each is defined for custom parts. Not every buyer needs every row; a buyer of simple turned pins may drop first article metrics after launch, while a buyer of cast and machined housings will want all of them.
| Metric | How to calculate it | Why it suits custom parts |
|---|---|---|
| On-time delivery (OTD) | PO lines received complete inside the agreed window around the first confirmed date ÷ PO lines due in the period × 100 | Line-level and fixed reference date, so re-promising cannot hide slips |
| Re-promise count | Number of confirmed dates changed by the supplier in the period | Early warning: a rising re-promise count often comes before late deliveries |
| Lot acceptance rate | Lots accepted at receiving inspection without sorting or deviation ÷ lots received × 100 | Stable with small lots, unlike PPM |
| Defective PPM (rolling) | Rejected parts ÷ received parts × 1,000,000, over a rolling window long enough to hold several lots | Still comparable across suppliers once the window is long enough |
| NCR count and repeat NCRs | Nonconformance reports opened in the period; repeats = same part and characteristic as an earlier NCR | A repeat shows that a corrective action did not work, or was never applied to that part |
| FAI first-pass rate | First article reports accepted on first submission ÷ first article reports submitted × 100 | New part numbers are frequent on custom accounts |
| Corrective action response and closure age | Days from NCR to containment confirmed; days from NCR to verified closure; count of open actions past due | Measures the system, not only the parts |
| Document accuracy | Shipments whose certificates, inspection report and packing list are complete and correct on arrival ÷ shipments | Late or wrong certs block receiving as surely as bad parts |
| Disruptions and premium freight | Count of line stoppages, yard holds or expedited shipments caused by the supplier | Captures the cost the buyer actually felt |
| Responsiveness | Median working days to answer an RFQ or an engineering question | Slow answers delay launches and changes |
Two definitions decide whether the scores can be trusted. First, “on time” needs a window, such as a set number of days early or late, and a rule for partial shipments; without them every review becomes an argument. The reference date matters as much as the window. Take twelve PO lines in a quarter, three of which the supplier re-promised two weeks late and then shipped on the new date: measured against the latest promise, on-time delivery is 100 percent; measured against the first confirmed date, it is 75 percent, and only the second number matches what your planners lived through. Second, “rejected” needs a rule for parts accepted under a deviation or sorted at your dock. Counting those as accepted makes the supplier look better than the parts were, so most buyers count them as rejects for scoring and record the deviation separately, as described in the supplier deviation guide.
Cost and price competitiveness often appear on templates too. They belong in a separate commercial review, because mixing price with quality and delivery in one weighted total lets a cheap supplier hide a quality problem behind a good price score.
How do you read scorecard numbers without being fooled by small lots?
Look at trends and counts over a rolling window, not at one month’s percentage, and put a confidence range on any rate built from a small number of parts or lots. A change is real when it persists across several periods or crosses a control limit, not when one lot goes bad.
Small samples make rates noisy. The NIST/SEMATECH e-Handbook shows that a proportion from a small sample has a wide, asymmetric confidence interval; its worked example of 4 defective units in 20 gives an exact 90% interval of roughly 7% to 40% (NIST, confidence intervals for a proportion). A supplier whose lot acceptance falls from perfect to a single rejected lot in a quarter has not necessarily changed at all. The same reasoning works in reverse: a clean quarter on a handful of lots does not prove the process is capable.
Where a supplier ships enough lots, a p-chart of the lot reject fraction separates noise from change. NIST gives the p-chart limits as the average fraction plus or minus three times the square root of p(1−p)/n (NIST, proportions control charts), so the limits depend on n, the number of lots in the period; when that number changes from period to period, recalculate the limits for each period. Where there are too few lots for a chart, use simpler rules that do not need statistics:
- Any repeat nonconformance on the same characteristic is a signal on its own, whatever the rates say.
- A metric below target for consecutive review periods is a trend; a single dip is a conversation.
- Re-promised dates rising while on-time delivery still looks fine is an early delivery warning.
- Corrective actions aging past the agreed date tell you more about the supplier’s system than any defect rate.
Diagnose before you escalate. When a quality score falls, sort the NCRs by characteristic, part family and production site. Drift concentrated on one feature points to a process or gauge problem; drift spread across many part numbers points to a system problem such as staffing, planning or a changed subcontractor. For a supplier that manages several plants, scoring by production site is the only way to see which plant is moving.
A monthly drift check that catches most of this takes one person less than an hour, provided the data already sits in receiving and purchasing records:
- Pull every PO line due in the month and compare the receipt date with the first confirmed date, not the latest one. List the lines that moved, and by how much.
- List every NCR opened in the month. For each, search earlier NCRs for the same part number and the same characteristic; any match goes straight to the escalation table below.
- Group the month’s NCRs by characteristic type (threads, a datum-related position, surface finish, coating, burrs, documents) and by production site. Two part numbers failing the same type of feature in one month deserves a question even if neither is a repeat.
- List corrective actions past their agreed dates, with their age in days.
- Check the shipments received against the document list on the PO: certificates present, heat or lot numbers matching the tags, inspection reports citing the current drawing revision.
- Note any change the supplier announced, or that you noticed without an announcement: new packaging, a different certificate format or mill, a new ship-from address. These often come just before a shift in quality.
Send the supplier the lists, not just the scores, and ask for a written answer on each item before the quarterly review. Most drift is easier to reverse at this stage than after a formal corrective action.
What thresholds and escalation steps turn a score into action?
Set a target and a trigger for each metric from your own history and the cost of a miss, then tie each trigger to a named action and owner. Escalation should step up from a conversation to a formal corrective action, then to controlled shipping or a sourcing review, and step down again when the data recovers.
Published templates show targets, but a target copied from someone else’s program means little for your parts. A safety-critical fitting deserves a stricter quality trigger than a cosmetic cover, and a part with weeks of buffer stock can tolerate a delivery slip that a just-in-time part cannot. Write the thresholds as colors (green, amber, red) with the boundary values in the quality agreement, and leave them unchanged for at least a year so the trend is comparable.
| Signal on the scorecard | Action | Exit condition |
|---|---|---|
| One metric amber for one period | Raise it at the regular review; ask for the cause in writing | Metric back to green next period |
| Any repeat NCR, or a metric red for one period | Formal corrective action request with 8D or equivalent; containment confirmed; receiving inspection moved to tightened sampling | Root cause verified and no recurrence over the agreed lots |
| Red for consecutive periods, or corrective actions past due | Supplier-performed 100 percent inspection at its cost (controlled shipping); management review call | Clean lots under containment plus a closed corrective action |
| No improvement under controlled shipping | Third-party inspection or on-site audit; new business on hold | Audit findings closed and scores recovered |
| Sustained failure | Sourcing review and resourcing plan | Decision recorded |
If your receiving plan follows ANSI/ASQ Z1.4, note that the standard has its own switching rule: under normal inspection it moves to tightened inspection when two of five consecutive lots are rejected on original inspection. Moving to tightened sampling after a single repeat NCR, as in the ladder above, is a stricter rule you choose, so write it into the quality agreement. Every step needs an exit. A supplier who sees no route back to green has little reason to invest in fixing the cause. The 8D report guide covers how to judge whether a root cause is real before you close the step.
What should the PO, quality agreement and RFQ say so the scorecard can be scored?
Every metric needs data the supplier is obliged to send and a definition both sides signed. Put the definitions and review rhythm in the quality agreement, and put the per-order inputs (confirmed date, documents to ship, inspection basis) on the RFQ and purchase order.
- Confirmed date. “Supplier confirms a ship date for each PO line; on-time delivery is measured against the first confirmed date within the agreed early and late window.”
- Documents with each shipment. Name them: material certificates, certificate of conformance, dimensional report on listed characteristics, and finish or coating certificates where a finish is specified. Name the drawing revision they must cite.
- Inspection basis. State the receiving sampling plan and the rule that sorted or deviated lots count as rejected for scoring; the receiving inspection guide explains how to set it.
- First article. When a first article report is required and in what format, for example the forms of SAE AS9102C or a ballooned drawing report, as covered in the first article guide.
- Corrective action. The expected format, the time allowed for containment and for the full report, and how closure is verified.
- Change notification. The supplier must tell you in advance of any change of production site, subcontractor, material source or process, because unannounced changes are a classic cause of sudden drift.
- Review rhythm. Who sends the data, when the scorecard is issued and who attends the review.
Share the scorecard with the supplier every period. A score the supplier never sees cannot change behavior, and sharing it early turns the review from a verdict into a working meeting.
How does a supplier scorecard work on a Saola order?
Saola Manufacturing can be scored like any other supplier, with one difference worth building into the scorecard: the work comes from more than one plant. CNC turning and milling run at our lead CNC plant in Vietnam, while casting, stamping and surface finishing go to partner plants we manage, and each quote names the production site and country of origin.
Because of that, the most useful scorecard for an account with us records the production site for each lot, so a drift on finished parts is not hidden inside a good machining record or the other way round. We remain the contracting party for every step, so nonconformances, corrective actions and change notices run through one contact whichever plant made the part. Turned and milled features are worked through with the machining and inspection team at the lead plant, which runs CNC turning and CNC milling and has a 2D vision measuring machine, height gauge and hardness tester on site. Cast, stamped or finished features involve the managed partner, and the corrective action says which plant made which step; see surface finishing and die casting for how those processes are arranged.
ISO 9001 certification is in progress, not held, and the quality and inspection page lists the records available with each lot. If your scorecard depends on particular documents, a sampling basis or a change-notice rule, put them in the RFQ and they are confirmed in the quote. To start, send the drawing and your quality requirements for a quote.
Frequently asked
How many metrics should a supplier scorecard have?
Usually six to ten. Fewer than that misses soft signals such as late certificates; many more and nobody reviews them. Keep delivery, lot acceptance, repeat nonconformances and corrective action age as the core.
Is PPM a good quality metric for low-volume suppliers?
Only over a rolling window long enough to hold several lots. For month to month tracking, lot acceptance rate and the count of repeat nonconformances are steadier and harder to argue with.
Should a supplier scorecard be weighted into one overall score?
A single score is convenient for ranking, but review the individual metrics too. A weighted total can hide a red quality metric behind green delivery, so many buyers add a rule that any red metric caps the overall rating.
How often should supplier scorecards be reviewed?
Update the data monthly and hold a review with the supplier at least quarterly for active parts. An annual review alone usually finds drift after it has already caused a disruption.
Should the supplier see its scorecard?
Yes. Send it every period with the underlying data, so the supplier can check the numbers and act on them before the next review.
Sources
- IATF 16949:2016, Automotive quality management system standard (about the standard), International Automotive Task Force, 2016
- SAE AS9102C, Aerospace Series – First Article Inspection Requirements, SAE International, 2023
- NIST/SEMATECH e-Handbook of Statistical Methods, 7.2.4.1 Confidence intervals, National Institute of Standards and Technology, 2012
- NIST/SEMATECH e-Handbook of Statistical Methods, 6.3.3.2 Proportions control charts, National Institute of Standards and Technology, 2012