Sampling Plan
Definition
Sampling plan: how QA teams pick what to inspect
A sampling plan is a documented framework that specifies which items are pulled from a larger batch for inspection, how many, and what results will trigger acceptance, rework, or rejection. Every serious quality assurance programme runs on one, because inspecting every unit is rarely affordable and rarely necessary.
The plan is the bridge between statistics and shop-floor decisions. It turns “we checked a few calls” into a defensible, auditable process.
Key takeaways
- A sampling plan defines sample size, selection method, and accept/reject criteria before inspection starts.
- Probability sampling gives every item a known chance of selection; non-probability sampling doesn’t.
- BPOs, factories, and auditors use standards like ANSI/ASQ Z1.4 to keep sampling defensible.
- The plan is worthless if the sample isn’t representative of the population.
How it works
A sampling plan answers four questions in writing: what population is being inspected, how large the sample will be, how the sample is chosen, and what defect rate will fail the batch. The quality analyst then executes it the same way every cycle, so results are comparable week to week.
The maths behind sample size comes from acceptance sampling theory. ANSI/ASQ Z1.4 — the American national standard for attribute sampling, first issued in 1950 as MIL-STD-105 and updated in 2018 — publishes lookup tables that pair batch size with an “acceptable quality limit” and spit out the exact sample count. The NIST/SEMATECH e-Handbook of Statistical Methods walks through the underlying operating-characteristic curves for teams that want to build their own.
Two families of sampling exist, and they aren’t interchangeable.
| Type | Method | When to use |
|---|---|---|
| Probability | Simple random | Baseline QA audits where any bias is unacceptable |
| Probability | Systematic (every nth item) | High-volume call call center monitoring |
| Probability | Stratified | Multi-team or multi-shift operations where each subgroup must be represented |
| Probability | Cluster | Geographically distributed sites, sampled by branch |
| Non-probability | Convenience | Quick spot checks, never for compliance sign-off |
| Non-probability | Quota | Market research where demographics matter more than randomness |
| Non-probability | Snowball | Rare-event studies (fraud, safety incidents) |
Scribbr’s methodology primer on sampling methods is a clean overview of the two types of sampling plans if you need to explain the split to a stakeholder.
The plan also fixes the acceptance number, the maximum defects a sample can contain before the whole batch is rejected. Set it too tight and good batches get scrapped; set it too loose and defects slip through. The American Society for Quality publishes reference tables that most auditors treat as gospel.
Examples
Contact centre QA. A BPO with 400 agents pulls 5 calls per agent per week using systematic sampling. Calls are scored against a rubric covering compliance, empathy, and first call resolution. The plan sets a 90% pass threshold, so any agent below it triggers coaching.
Pharmaceutical batch release. A contract manufacturer producing 100,000 tablets per lot uses ANSI/ASQ Z1.4 Level II inspection. The lookup table says pull 200 tablets, and the batch fails if more than 3 are out of specification. Results feed straight into the regulatory dossier.
Insurance claims audit. A back-office outsourcing provider handling 12,000 claims a month runs stratified sampling — 30 claims per adjudicator, per month. Stratification stops one high-volume adjudicator from dominating the sample and hiding weaker performers.
Ecommerce packaging QC. A fulfilment centre uses simple random sampling — 50 parcels a day pulled by barcode scan. The error rate target is under 0.5%, and any breach kicks off a root-cause review before the next shift.
Related terms
- Quality assurance: the broader discipline that sampling plans serve.
- Quality analyst: the role that designs, runs, and reports on the plan.
- Error rate: the output metric a sampling plan is built to measure.
- First call resolution: a common metric sampled inside contact-centre QA.
- Compliance: regulated industries lean on sampling plans for audit defensibility.
- Business process outsourcing (BPO): where sampling plans are used at scale across offshore teams.
FAQ
What’s the difference between a sampling plan and a sampling method?
The method is the technique: random, systematic, stratified. The plan is the full document, covering the method, the sample size, the acceptance criteria, and the reporting cadence. One plan uses one or more methods.
How large should the sample be?
It depends on batch size, acceptable defect rate, and the confidence level you need. ANSI/ASQ Z1.4 gives standard lookup tables; most BPO QA teams sample 5–10% of transactions, but regulated work follows the tables strictly.
Can I change the plan mid-cycle?
Not without documenting why. Changing sample size or acceptance rules partway through a review period breaks comparability and invalidates trend data. Change it at cycle boundaries only.
Is convenience sampling ever acceptable?
For informal spot checks and pilot studies, yes. For anything that feeds a compliance report, a client SLA, or a regulatory filing, no; the lack of randomness makes results indefensible under audit.
Who owns the sampling plan in a BPO?
Usually the QA manager, signed off by operations and the client. In regulated work, a compliance officer countersigns. The plan sits in the quality management system and is reviewed at least annually.
Need help designing a QA sampling framework that stands up to client and regulator scrutiny? Browse the outsourcing knowledge base at Outsource Accelerator’s hubs to compare providers and frameworks.
Independent
