Factory acceptance testing is the most visible quality activity in a capital equipment programme. It has a date, a protocol, travel, witnesses, and a pass or fail at the end. It is also, structurally, the worst place to discover a manufacturing quality problem.
Not the worst place overall — discovering it in the field is worse. But FAT is the last stop, and treating it as the primary quality gate means every finding arrives at the moment it is most expensive to act on.
The problem
By the time equipment sits on a test stand, four things have already happened. The material has been consumed. The labour has been spent. The pressure boundary is welded, insulated and enclosed. And the schedule has been committed downstream — a site has a date, an installation crew, and often a load waiting.
Each of those raises the cost of a finding, but the last one is the dangerous one, because it changes the argument in the room. A finding at fabrication is a technical conversation. The same finding at FAT is a commercial conversation with a technical topic, and commercial conversations have a strong pull toward the answer that preserves the date.
There is a second, quieter problem. Some characteristics are no longer verifiable at FAT at all. A weld can be pressure tested at FAT, but you cannot examine its root after it has been enclosed. Internal cleanliness can be sampled, but not observed. If the only control on those characteristics was “we will see it at FAT,” then there was never a control — there was a hope that the failure mode would be one of the detectable ones.
A framework: where should the evidence be created?
For each characteristic that matters, ask two questions in order:
When does the evidence stop being obtainable? Some evidence has an expiry. Weld root condition expires at the next operation. Internal cleanliness expires at closure. Component configuration expires at assembly — once a pump is coupled and piped, verifying its impeller trim means taking it apart. Evidence with an expiry has to be created before the expiry, which means the inspection point is fixed by physics, not by convenience.
What does it cost to act on a finding here versus later? A wrong component found at receipt costs a return. Found at assembly, it costs disassembly. Found at FAT, it costs disassembly plus a retest plus a schedule conversation. The ratio between those numbers is the value of the upstream inspection point, and it is usually an order of magnitude per stage.
Those two questions between them tell you where the quality plan should put its weight. FAT then becomes what it should be: confirmation that equipment already believed to be conforming performs as specified — not a search.
Worked example
In a synthetic investigation I wrote up, a cooling skid reached 195 kPa differential pressure against a 250 kPa requirement at design flow. The cause was a pump impeller one trim size smaller than the design required — ordered wrong, because two trims shared a catalogue description, then received and installed without any control verifying the trim against the bill of material.
Trace where that could have been caught:
At order placement. Configuration-critical attributes stated on the purchase order line instead of relied on from a catalogue description. Cost of the control: a template change.
At receipt. Impeller trim verified against the BOM — explicitly not against the supplier packing list, which was consistent with the wrong order. Cost: minutes per pump.
At assembly. Independent confirmation of pump configuration before coupling, signed by someone other than the installer. Cost: minutes, and the last point at which the evidence is cheap to obtain.
At FAT. Detected as a 22% performance shortfall. Cost: investigation, disassembly, replacement impellers, full retest, containment across other units and stock spares, and a schedule impact.
The performance test did its job. It caught the defect before shipment, which is what a last line of defence is for. But the defect passed through three points where the evidence was available and cheap, and no control was looking for it. That is the real finding, and it is a finding about the quality plan rather than about the skid.
Engineering implication
The practical consequence is how you choose hold points. A hold point is expensive — it stops a supplier’s production line waiting for your release — so it has to be spent where it buys the most. My rule:
Put a hold point where the evidence becomes unavailable or unaffordable to obtain afterwards. Pressure testing, before the boundary is enclosed. Performance, before the equipment is crated. Documentation turnover, before the equipment ships and your leverage disappears with it.
Everything else is better served by a witness point, surveillance, or a record review — and by putting a supplier’s own verification step upstream where it is cheap. In the ITP for that study, three of twenty steps are hold points. If a quality plan has ten, the author has not decided what matters; if it has none, the author has decided that nothing does.
And a corollary worth stating plainly: if FAT is finding a lot, that is not evidence that FAT is working well. It is evidence that the upstream plan is not.
Takeaway
FAT should confirm, not discover. Place inspection where the evidence expires and where acting on a finding is still cheap, and reserve hold points for the handful of points where waiting is genuinely worth stopping a production line for.