PCBA Acceptance Criteria: What to Agree Before the Line Starts

Article

October 26, 2026

In a public buyer forum this year, a purchaser running about 150 assembled boards a year wrote that reliability and transparency mattered to him far more than unit price. His previous order had come back with a failure rate of 25 percent. He was not looking for a cheaper supplier. He was looking for one that could tell him, before the boards were built, what would count as a failure.

That is a contract question, not a manufacturing question. And most small-batch orders never answer it — not because the two sides disagree, but because the question was never put on the table while it was still cheap to answer.

A Defect Rate Without a Base Is a Rumor

“25 percent failed” sounds precise and, on its own, says almost nothing. Twenty-five percent of what — boards, placements, joints, or orders? Within one batch or across a year? Counted by the factory’s final inspection, or by the customer when the units came back from the field?

Two numbers decide whether an acceptance clause means anything: the base, and the sample size. If you agree to inspect thirteen boards out of fifty, you have agreed to a statistical statement rather than a guarantee. At that sample size, a lot carrying a genuine five percent defect rate will pass a zero-defect criterion roughly half the time. That is not an argument against sampling. It is an argument against treating a passed sample as proof.

For small batches the honest structure is usually not the automotive AQL table. On fifty boards, full visual and full electrical inspection of every unit is affordable; what is not affordable is a sampling plan designed for ten thousand. Write down which of the two you are buying. A supplier who says “we check everything” and a supplier who says “we apply AQL 0.65 at the final station” are selling two different products at two different prices, and only one of them can be verified afterwards.

What AOI Sees, and What It Only Appears to See

Automated optical inspection photographs the board and compares it against what the program expects. It is genuinely strong on presence and absence, offset, rotation, billboarding and tombstoning, insufficient or bridged solder, and wrong markings on the top surface. It catches the class of error a tired operator makes at three in the morning, and that is worth paying for.

It cannot see under a BGA. It cannot see a cold joint that happens to look like a proper fillet. It cannot see a component with the right shape, the right marking and the wrong die inside. And it will not tell you that the reel was a refurbished lot with a fresh label.

So “AOI 100 percent” is a statement about the placement process, not about the product. If your board carries a BGA, a QFN with a thermal pad, or a power device whose behaviour depends on voiding, the inspection that matters is X-ray — and X-ray is normally sampled rather than run on every unit. If your product fails in the field rather than at the station, the inspection that matters is functional test under load, and possibly burn-in. Ask which stations run on your order and which are skipped, and ask for the list in writing. That answer is worth more than any certificate on the wall.

The First Article Is the Only Build You Can Still Change

First article inspection is often treated as paperwork that delays the shipment. It is the last moment when a mistake is still cheap. After that, the mistake has fifty copies.

A useful FAI is not a signature on a drawing. It is line-by-line confirmation that the board in front of you matches the file you sent: the manufacturer part number of every BOM line, the orientation of every polarised component, the keying of every connector, the firmware or test program revision actually loaded, and the panel or single-board format actually used. It takes an hour and it catches exactly the class of error that no amount of final inspection can recover, because the whole batch shares it.

The failure mode to watch for is the FAI signed before the board exists, filled in from the design file rather than from the physical unit. If the sheet is complete and identical to the last four orders, it was not an inspection.

Four Documents That Decide Who Is Right

When a batch fails, the argument is rarely about whether it failed. It is about what was promised. Four documents settle most of it, and all four have to exist before the build, because none of them can be reconstructed afterwards.

The first is the inspection record with identifiers: which station, which program revision, which operator, which time window, which lot. A summary PDF that says “passed” is not a record; it is a claim.

The second is the first article sheet, filled in from the physical board.

The third is the material traceability list — for each BOM line, the manufacturer, the lot or date code, and the channel it came through. Without it, a batch-wide failure cannot be narrowed to a suspect lot, and the supplier cannot be asked whether other orders used the same material.

The fourth is the remedy clause. More on that below, because it is the one that is almost always written badly.

“We Will Replace It” Is Not a Remedy

Replace is the word suppliers reach for, because it sounds generous. What it usually means is: we will build another batch and ship it, and you will absorb four weeks, the freight, your own rework hours, the conversation with your customer, and the delay to whatever the boards were supposed to go into.

A remedy clause worth the paper it is written on answers five questions. Who pays return freight on the failed units. Who pays your labour if you sort or rework them locally. Is the fix a repair of the existing units or a rebuild, and who decides. What is the deadline, counted from which event. And what happens if the second batch fails the same way.

Buyer forums carry a real case of a batch of modules that failed in the field, where the settlement offered was a voucher worth about thirty pounds against a far larger loss. The complaint was not the amount. It was that the voucher assumed the loss consisted of parts, when the loss consisted of a schedule and a customer relationship. A credit note cannot be spent on a missed launch date. Write the remedy in units of time, not only in units of money.

The Sheet: What to Agree Before the Line Starts

The table below is the version we use in our own quotations. It is deliberately short. Every line can be settled in a single conversation before the order is confirmed, and every line is expensive to settle afterwards.

One column deserves a second look: “when it has to be settled”. Almost every item on this list is cheaper to agree before material is purchased, and the cost of agreeing it later is not administrative. It is a re-quote, a rebuild, or a dispute that neither side wins.

What We Promise, and What We Do Not

SPI, AOI, X-ray and functional test run in the same plant as the SMT line, and we will tell you which stations your order passes through and which it does not. You get the first article sheet filled in from the physical board, the material traceability list with lot codes, and a remedy clause that names a deadline and a person.

We will not promise a zero defect rate. Nobody who has run a line for a year would. What we can promise is that the number is defined before the order, that the evidence exists if the number is missed, and that the first response to a failure is a technical explanation rather than a credit note.

If your current supplier cannot tell you the base behind their defect rate, that is not a confidentiality problem. It is the answer.