Decision guide
Pre-shipment inspection or an accountable supply model: which control does the order need?
The decision is usually framed as a budget question and answered by comparing an inspection fee to a purchase-order value. That comparison is easy, quick and structurally misleading: it prices the control and ignores the exposure. The useful question is narrower — what kind of failure is this order actually capable of, and at what point in its life would that failure still be cheap to fix?
What a final random inspection is built to do
A pre-shipment inspection is a terminal conformity gate, and it is good at being one. An inspector attends near completion — in most published service definitions, at or beyond 80% production, with goods substantially packed — draws a random sample, works through a checklist covering quantity, workmanship, labelling, dimensions, packing and basic function, and returns a decision on the lot.
Two properties of that instrument matter for the decision. First, it is acceptance sampling: ISO 2859-1 defines AQL-indexed lot acceptance sampling, and the statistical literature is explicit that acceptance sampling decides lot dispositionrather than estimating lot quality — producer and consumer risk are recognised features of the method, not defects in it. Second, its intervention point is late by design, which compresses the correction window against the ship date. For the goods it was designed for, both properties are acceptable trade-offs. For an engineered machine, both become structural limits.
The coverage arithmetic worth doing once
Physical inspection coverage is n/N, and the ratio is usually smaller than buyers picture. A sample of 200 units from a lot of 4,000 is 5% coverage: 95% of the shipment is never physically examined. That is not a criticism of sampling — it is what sampling is — but it should be a conscious choice rather than an assumption.
For how AQL levels and sample sizes are actually chosen, and what a given plan does and does not promise, see AQL sampling explained.
The two models side by side
| Evaluation metric | Final random inspection | Accountable supply model |
|---|---|---|
| Control objective | Verify sampled finished goods against the PO and specification; issue a lot decision | Deliver conforming output by controlling supplier, process, evidence and recovery — inspection is one gate inside it |
| Intervention point | Near completion, often with goods substantially packed; correction window compressed against the ship date | Supplier approval, specification freeze, tooling, pilot, in-process, then final release — risk attacked before value is locked in |
| Decision basis | Checklist, random sample and an AQL accept/reject rule | Product evidence plus process readiness plus open-risk status; release only when defined gates close |
| Physical coverage | n/N — a fixed sample regardless of which characteristics carry the consequence | Risk-based mix: samples, process data, 100% checks on critical characteristics, serial or batch traceability |
| Change and configuration control | Detects substitutions only where they are visible or land in the sample; golden-sample-to-production drift can escape | Approved BOM, specification and tool baseline; change and deviation approval; configuration locked from sample to mass production |
| Root cause and corrective action | Records non-conformities; root cause, corrective action and verification sit with supplier and buyer, and re-inspection is commonly a new engagement | Owns containment, root-cause analysis, action dates, effectiveness evidence and recurrence prevention, with closure timing capable of being contractual |
| Release authority and leverage | Late-stage leverage — hold or reject the shipment; strength depends on the unpaid balance and control of goods and documents | Leverage preserved across deposit, tooling, pilot, production, balance and shipment gates, with contracted stop-work and deviation authority |
| Evidence integrity | Time-stamped report, photographs and sampled-unit results, often relying on the supplier-presented lot, records and access | Population lock, independent randomisation, serial or batch linkage, signed gate records, raw data and an audit trail retained for claims |
| Cost structure | Low visible fee per visit or man-day; extra tests, travel, re-inspection and follow-up billed separately | Higher upfront engineering and project cost, structured against milestone gates |
| Best fit | Stable, repeat, lower-consequence goods where a terminal conformity gate is sufficient | Custom, engineered, high-value, high-consequence, tooling, machinery, multi-supplier project packages, or weak supplier maturity |
Where the difference actually bites
Three of those rows carry most of the practical weight, and they are the three that never appear in a fee comparison.
- Production readiness versus a snapshot — a final inspection photographs finished units at one moment. It offers limited proof that the process will repeat at rate. Where that proof is required, the mechanism is production-run evidence in the manner of a PPAP submission: demonstrating that the supplier understands the requirement and can consistently meet the design and specification during an actual production run at production rate.
- Measurement assurance— calibrated tools may be used during an inspection, but gauge suitability, fixture correlation, gauge repeatability and reproducibility, and measurement uncertainty commonly sit outside a discrete inspection scope. On tight-tolerance work, the measurement system is itself a source of accept/reject error, and it needs a plan rather than an assumption.
- Where liability attaches — under an inspection service contract, liability attaches to the service and its report, and is normally capped as a multiple of the service fee with consequential loss excluded. That is standard and reasonable for a service engagement; it simply means the remedy is scaled to the fee rather than to the cargo. Under an accountable supply contract, remedies attach to named control duties — re-inspection, rework or tool correction, supplier recovery, chargeback — with caps and exclusions negotiated explicitly. Neither structure is an insurance policy, and any model presented as one should be read with suspicion.
The operating measures follow from that split. An inspection programme is steered by defect counts, AQL pass rates, quantity verification and report turnaround. A supply-control programme is steered by first-pass yield, escape rate, process capability, non-conformity ageing, corrective-action closure, gate on-time performance, tooling changes and recovered cost — measures that only exist if someone is accountable for the process, not just for observing the output.
The risk-adjusted test
The commercial test is not which control is cheaper. It is which produces the lower total expected cost:
purchase price + assurance cost + expected delay cost + expected rework cost + P(failure) × impact − realistic recoveries
This is the ordinary cost-of-quality framing: prevention and appraisal costs evaluated together with internal and external failure costs. Written that way, the choice usually resolves itself. When the expected loss from a defect escape, a tooling error, a delay, a field failure or a weak supplier recovery exceeds the incremental engineering cost, buying only a terminal inspection is false economy. When it does not, the terminal inspection is the correct and proportionate answer, and paying for more control is waste.
One honest caveat belongs on both sides of the ledger: sampling risk does not disappear under the accountable model. Upstream control reduces how often defects occur and adds detection, containment and recovery, but it does not make a sample a census. What changes is the number of independent chances a defect has to be caught, and who is contractually obliged to deal with it when it is not.
Choosing, in practice
Four questions settle most cases without a spreadsheet:
- Is the risk created before the packing stage? Tooling, process capability, material substitution and configuration drift all are. If the answer is yes, a terminal gate is detecting late by construction.
- What does a failure cost relative to the order? A returnable consumer item and a plant-critical machine sit at opposite ends, and the same inspection fee buys very different protection in each case.
- How much leverage will still exist when the problem appears? Leverage is a function of unpaid balance and control of goods and documents. If the balance is nearly settled at the point of detection, the remedy conversation is already lost.
- Who owns the correction? If the answer is “we will work it out then”, the answer is the buyer.
For the mechanics of the inspection itself — what is checked, when it is booked and what the report should contain — see the pre-shipment inspection checklist and what a pre-shipment inspection costs in China. For the upstream gates this page argues for, see the controlled order package and execution partner vs sourcing agent.
Frequently asked questions
Is a pre-shipment inspection ever enough on its own?
Frequently, yes — and saying otherwise would be dishonest. For catalogue goods, repeat orders from a supplier with a known history, consumer products moving in volume, or anything where the consequence of an escaped defect is a return rather than a plant shutdown, a final random inspection at the right AQL is a proportionate and cost-effective control. The problems start when the same instrument is asked to carry an engineered machine, a first-article tooling programme or a multi-supplier project package, because those risks are not created at the packing stage and cannot be detected there.
What does "standard PSI" mean in this comparison?
A discrete final random inspection engagement: an inspector attends once, near completion, samples the finished lot against a checklist and an AQL, and issues a report. It does not represent every audit, testing, during-production or consulting service that a large inspection and testing network may separately offer — many of those services address the same upstream risks discussed here. The comparison is between two control models, not between organisations.
When is the accountable model not worth it?
When the order is small relative to the engineering cost, when the goods are stable and the supplier is proven, or when the buyer already has the internal engineering capacity and simply needs eyes on the ground. It is also not worth it when the contract will not actually carry the authority: without release rights, defined deliverables, escalation and negotiated remedies written down, the label is marketing rather than a mechanism.
Can we start with inspection and add controls later?
You can, but the sequence matters more than the budget. Most of the leverage in an order sits before money moves and before tooling is cut — supplier approval, specification freeze, DFM, first-article and in-process gates. Adding controls after the deposit is paid and the tool is machined means paying for detection at the point where correction is most expensive. Where budget is genuinely constrained, the higher-yield move is usually to move one gate earlier rather than to add one at the end.
How do we compare the two commercially?
Not by comparing inspection fee to purchase-order value, which omits most of the exposure. Compare total expected cost: purchase price plus assurance cost plus expected delay cost plus expected rework cost plus the probability of failure times its impact, less whatever you can realistically recover. This is the ordinary cost-of-quality framing — prevention and appraisal costs evaluated together with internal and external failure costs — and it is the only comparison in which a low visible fee and a high residual exposure appear on the same line.
How Sinospect structures the accountable model
In Sinospect’s standard supply model the buyer contracts with Sinospect as the principal supplier and accountable counterparty: Sinospect places and manages the factory order, applies the agreed quality gates, and keeps the factory’s final balance conditional on Sinospect’s own quality control passing. The buyer has one commercial interface for the order and one place where the correction obligation sits. Where direct purchasing suits the project better, the same gates run on the buyer’s own order as an execution mandate. Either way the terminal gate stays in place — pre-shipment inspection is one control inside the sequence, not a replacement for it, and factory acceptance testing carries the load where the goods are machinery. Send the specification or the order and Sinospect will say which model fits and why — including when the answer is a single inspection.
Not sure which control your order needs?
Send the specification or the purchase order, the supplier if one is already chosen, and what a failure would cost the project. Sinospect replies within one business day with the control model it would recommend, what it would cost, and where a lighter option would do.