Sampling Plans: How Much Should You Actually Test?

A buyer in Guangzhou lost $47,000 last Tuesday.

Full container. Laptop stands. She tested 5 pieces out of 3,000.

All 5 passed.

The other 2,995? Garbage. Cracks in the hinge. Wobbly base. One fell apart when her client opened the box.

She’s filing for bankruptcy next month.

This is what happens when you don’t know how sampling works. You gamble. The house always wins.

I’ve been doing this for 6 years in Shenzhen. I’ve seen every trick. The sample size debate isn’t academic theory. It’s the difference between profit and legal hell.

So let’s talk about how much you should actually test.

What Suppliers Say vs What They Mean

First, the lies.

See the pattern?

They want you to check less. Always. Because checking reveals problems. Problems delay payment.

And they’re betting you don’t know the math.

The AQL Thing Everyone Gets Wrong

AQL stands for Acceptable Quality Limit.

Think of it like food poisoning odds at a restaurant.

AQL 2.5 means you’re okay if 2.5% of diners get sick. You walk in knowing that 25 out of 1,000 people will be vomiting by midnight. You accept this risk.

AQL 1.0? Only 10 people puke. Better odds.

AQL 0? Nobody gets sick. Costs way more to guarantee. Most restaurants can’t afford it.

Now replace “diners” with “products” and “getting sick” with “defects.”

That’s AQL.

The question is: what’s your stomach for bad units?

If you’re shipping baby bottles, AQL 0 or you’re getting sued. If you’re shipping keychains for a dollar store, AQL 4.0 might be fine. The client expects some duds.

But here’s the trap.

Most buyers pick AQL 2.5 because it sounds “standard.” They never ask: standard for what? A $2 item or a $200 item? A safety product or a novelty pen?

I watched a guy lose his Amazon account over this. Phone cases. AQL 2.5. That meant 1 in 40 cases cracked on arrival. Amazon doesn’t care about your sampling plan. They care about returns.

His account got suspended in 6 weeks.

So before you pick a number, ask: what defect rate kills my business?

Then set your AQL lower than that.

How Many Pieces to Actually Pull

This depends on batch size.

Not your feelings. Not what the supplier suggests. The math.

If you order 500 units, you check around 80 pieces. That’s roughly 16%.

If you order 5,000 units, you check around 200 pieces. That’s 4%.

Why less percentage for bigger orders?

Statistics. The bigger the batch, the less you need to check to detect patterns. It’s called confidence intervals. Don’t worry about the formula. Just follow the chart.

Most inspectors use ISO 2859 tables. It’s free online. Plug in your order size and AQL. It spits out a sample size.

Do not negotiate this number with the factory.

I’ve seen buyers agree to “check only 50 instead of 200 to save time.” Guess what? The 50 were perfect. The other 4,950 were assembled by drunk temps.

You want to save time? Order from better factories.

Red Flags That Mean Pull More Samples

Here’s when the standard plan isn’t enough.

• The factory just changed a key component (cheaper plastic, different motor, new paint supplier)

• It’s your first order with this supplier

• The last batch had issues and they “promised to fix it”

• They’re rushing the order to meet your impossible deadline

• The boss isn’t at the factory during production (he’s “traveling”)

• You see new workers on the line who don’t know the process

• The packaging materials look different from the sample

• They keep pushing back your QC visit date

• You smell solvent or chemicals stronger than last time

• There’s a new subcontractor’s name on the shipping docs

Any of these? Double your sample size.

The factory will whine. Let them.

Because here’s what happens when you don’t: you find the problems after the container lands. Now you’re stuck with junk. You can’t send it back—the shipping cost is more than the goods. You can’t sell it—it’s defective.

Your money is gone.

I had a client ignore this last year. Bluetooth speakers. The factory swapped the battery supplier mid-production to save $0.30 per unit. We caught it because we doubled the sample size after seeing the new supplier name on an invoice.

Half the batteries were swelling after 2 hours of charging.

If we checked the standard 125 pieces instead of 250? We’d have missed it. The defect rate was 11%. You need a bigger sample to catch rates that low.

She would’ve shipped 10,000 fire hazards to California.

The Destructive Testing Problem

Some products need destruction to test properly.

Drop tests. Stress tests. Pulling wires until they snap.

This creates a nasty dilemma.

You can’t test every unit—you’d have no product left. But testing too few means you miss systemic issues.

The rule: destructive tests run on a separate, smaller sample. Usually 10-20 pieces depending on risk.

I’ve seen suppliers lose their minds over this. “You want to destroy 20 units?! That’s $500 of product!”

Cool.

Pay them for the 20 units. Add it to your sourcing cost. Because the alternative is shipping 5,000 units that shatter on impact, then refunding $125,000 to angry customers.

Which one sounds cheaper?

For stuff like electronics, you also need electrical safety tests. Voltage, grounding, insulation. These often destroy the unit.

Don’t skip them.

I remember a guy who skipped electrical tests on USB chargers. “They’re just phone chargers, what could go wrong?”

One caught fire in a Best Buy.

His company doesn’t exist anymore.

Random vs Stratified Sampling

Random: You walk into the warehouse, grab units from random cartons across the batch.

Stratified: You grab units from the start, middle, and end of the production run.

Always do stratified.

Why?

Because factories “improve” as they go. The first 500 units are made by the A-team, carefully, to impress you. The last 2,000 units? Made by whoever showed up that day, rushing to finish.

Or the opposite: they start sloppy, you visit and yell, they tighten up for the middle batch, then get lazy again at the end.

If you only pull random samples, you might grab 80% from the good middle section.

Stratified forces you to check all phases.

I saw this blow up on a toy order last month. Plush animals. The inspector grabbed random samples—mostly from the middle pallets. All passed.

Guess which batch had stitching defects? The last 800 units.

They didn’t catch it until the client opened boxes in Ohio.

$22,000 in replacements.

When You Should Test 100%

Very rare.

But it happens.

• Medical devices

• Baby products with safety risks

• High-value items (watches, jewelry)

• Orders where the defect rate is historically insane

• Products going to litigious markets (Germany, USA)

For the rest? Statistical sampling is fine.

Testing every unit costs too much time and labor. It also doesn’t guarantee perfection—inspectors get tired, they miss things on unit 2,847.

Better to test smart than test everything poorly.

The Real Trick

Sampling plans don’t prevent defects.

They detect them.

If your factory is making junk, no amount of testing will fix that. You’re just documenting the disaster.

The real trick is picking suppliers who don’t produce high defect rates in the first place.

How?

Audit their process before you order. Check their training, their machines, their material sourcing. Look at their scrap rate. If 15% of their daily production goes in the trash, that’s your future.

A good factory has defect rates under 2% before you even show up.

A bad factory is scrambling to hit 10%.

No sampling plan saves you from the second type.

This is why we push clients toward supplier audits before placing orders. You spend $500 now to avoid $50,000 in junk later.

Walk the line. Check the QC station. Ask workers how long they’ve been doing this task. Inspect their raw materials.

If the factory resists? Walk.

Because they’re hiding something worse than you’re imagining.

The Line You Don’t Cross

Final rule.

If your defect rate hits 5% during inspection, stop the shipment.

Not 4.9%. Not “close enough.”

5% is the edge of the cliff.

Ship it and you’re gambling that the rest of the unseen units are somehow better. They’re not. If 5% of your sample is bad, the real batch defect rate is probably 8-12%.

The factory will beg. They’ll offer discounts. They’ll promise to sort and rework everything.

Don’t.

Rework in China is usually “put the worst ones on the bottom of the pallet.”

Make them reproduce or refund. Over 5%? That batch is trash.