1,000 dining chairs, 940 of them 60 mm too tall, and a $897 inspection that found the problem before it cost half a million dollars

The order

A contract furniture supplier in Northern Europe had ordered 1,000 upholstered dining chairs from a factory in Foshan. One model, one colour, delivered flat-packed. The order was worth $128,000 at $128 a chair, and the buyer had paid a deposit of $38,400. The remaining $89,600 fell due when the factory asked for the vessel booking.

The chairs were for the restaurant and the dining room of a hotel group, and the buyer's own contract with that group carried a liquidated damages clause: $165,000 for late delivery of the furniture package. The buyer's project manager had signed that clause. The factory had never seen it, because it sat between the buyer and its customer.

Production ran seven weeks. The factory sent photographs of the finished chairs and asked for the balance.

Two documents, one seat height

The buyer's purchase specification ran to three pages. Clause 4.2 read: seat height 450 mm, tolerance plus or minus 5 mm.

The buyer's dimensioned drawing, issued with that same specification and used by the factory for cutting, dimensioned 450 mm from the floor to the top of the seat frame. The 60 mm seat cushion was called out on the upholstery sheet as a separate item.

The factory cut the legs to 450 mm, built the frames and upholstered them with the 60 mm cushion. Finished, the seat height was 510 mm.

Both documents sat in the factory's job folder and both were used correctly. Neither was wrong on its own terms. The number 450 was in both of them and it meant two different things, and nothing in a factory's process compares a written specification against a drawing, because in the factory there is nothing to compare. The drawing is the drawing.

A chair that is 510 mm tall looks like a chair

The buyer specified 450 mm for a reason. Their dining tables are 740 mm, and they sell into hospitality, where the diner at the top of the range is a 95th percentile man with a popliteal height of up to 490 mm. A 450 mm seat is the height at which that diner has their feet on the floor. At 510 mm their feet are on the chair ring and their knees are above the table.

The European seating standards measure seat height as the vertical distance in the median plane between the floor and the highest point on the front of the seat, cushion included, not the frame. EN 1728:2012 is the test method standard the industry works to for seating.

None of that is visible in a photograph. 510 mm of seat height on a chair that photographs well is a chair that photographs well. The defect was invisible, which is why it survived seven weeks of production and the buyer's own approved sample.

Six chairs

The pre-shipment inspection was a Level 3 full verification, three man-days at $299, so $897. The buyer asked for it because this was a repeat order and there was one thing they wanted settled from the previous shipment.

Seat height was on the measurement list. The sampling plan for a lot of 1,000 is code letter K under ISO 2859-1 at general inspection level II: a sample of 80, and at AQL 2.5 a lot is accepted with 5 major defects and rejected at 6.

The inspector measured seat height on the first chair at the loading door and got 510 mm. The second gave 510. The third gave 509. The inspector stopped at the sixth chair, because six non-conforming units is the reject number and every measurement after that changes nothing about the lot.

That is the first thing worth taking from this case. A sampling plan is built to find scattered defects. A defect that is in every chair turns up in the first handful of measurements, and the sample size of 80 was never the thing that mattered.

But the count is what prices the remedy, so the inspector measured the rest. A seat height takes about 25 seconds to check: chair on a level floor, unloaded, measured in the median plane to the highest point on the front of the seat. A thousand of them is seven hours of work, most of one man-day, and it came to 940 chairs at 509 to 511 mm and 60 chairs at 449 to 451 mm.

Why 60 chairs were right

The 60 chairs that measured 450 mm were the last 60 of the run. The factory's line supervisor had found the leg-cutting setting wrong against the drawing three days before loading, corrected it and put those 60 through the corrected setting.

Nobody told the buyer. The factory's position was that it had found and fixed its own error before the goods shipped. The buyer's position was that the order was 940 chairs of a non-conforming product with no notice of anything.

This is the part of the case that does not fit neatly into a report, and it is worth being precise about it. The 940 is not a defect rate and the 60 is not a pass. All 1,000 chairs are non-conforming to clause 4.2, because the specification says 450 mm and the lot contains two seat heights. The factory's internal correction did not make the lot conforming. It made the lot less uniform, and the uniformity is the thing that made the error invisible in the first place.

What the report recorded

The report on this order was four pages and it did six things.

• It named both documents and both figures, the specification clause and the drawing dimension, so the buyer was not left to argue about which one the factory had followed.
• It stated the measurement method and the instrument, because seat height measured to the cushion and seat height measured to the frame are two different numbers, and only one of them is 450.
• It gave the sampling result with the code letter, the sample size and the accept and reject numbers, and recorded that the seat height measurement stopped at the sixth unit because the reject number had been reached.
• It gave the census as a separate finding, 940 and 60, with the range of readings rather than an average, so that the 60 cannot be presented as a second conforming product.
• It recorded what it could not establish: why the last 60 differed, whether the 940 were the earlier or the later units, and whether the frames were otherwise correct. Those are questions for the factory's own production records, not for an inspector with a tape measure.
• It said plainly that the discrepancy sits between two of the buyer's own documents. The report states the facts. It does not decide who is at fault, because the report cannot.

What it cost

Two columns. The left one is what happened. The right one is what the same order would have cost if the chairs had been loaded without an inspection.

Read the two columns together and the argument is in the first line. $897 spent at the loading door against $500,000 of exposure that never happened. The inspection did not make this order good. It made this order a $46,000 problem instead of a $500,000 one.

The right column breaks down like this: $19,000 to air freight 300 chairs and hold the opening date, $22,000 to return and dispose of 940 chairs, $11,400 of storage over five months, $19,400 of demurrage and handling on arrival, $263,200 to buy 940 replacement chairs locally at $280 each, and the $165,000 penalty clause the buyer had signed with its customer.

The stage is worth more than the fee

The most useful thing in this case is not the table above. It is that the same error carried four different prices depending on when it was caught, and none of those prices depended on how good the inspector was.

Why the cost moves with the stage

Read down the last column and a rule comes out of it. The cost of a mistake at a checkpoint is roughly the cost of the material already committed past that checkpoint, and the inspection fee at that checkpoint is a rounding error beside it. What moves the cost is not the quality of the check. It is how many things have already been cut, welded, upholstered and packed.

The pre-production row is the one worth arguing about. It says the cheapest check on this order was not an inspection at all. It was the buyer's own engineer spending two hours reconciling clause 4.2 of the specification against sheet 2 of the drawing, before the factory cut anything. That check costs nothing, needs no man-day, and would have caught this error on day one instead of day 49.

How to price the check before you book it

Most buyers think about a furniture inspection cost as a percentage of the order. That is the wrong unit, because the fee is not a percentage of anything. It is man-days at a published rate: $149 for a Level 1 order check, $199 for Level 2 order and quality, $299 for Level 3 full verification. Travel across Guangdong is included, the report comes within 24 hours, and there is no weekend or holiday surcharge.

Here is the same two checks priced against five different order sizes. The fee barely moves, so the percentage falls as the order grows, and that is the whole arithmetic of the question.

The number the fee has to be compared against

The table above is a pricing table. It is not the decision. The decision is made against one other number, and on this order that number was $500,000 of exposure at the buyer's end: the penalty clause, the replacement chairs bought at short notice, and the cost of chairs sitting in a warehouse for five months while a dispute ran.

$897 against $500,000 is 557 to 1. That is the number to hold onto, and it is not a ratio you can compute before you know what could go wrong. It is why the practical answer to how much inspection costs is: whatever it costs, compared with what the specific thing you are buying insurance against would cost you if it happened. A $149 man-day that finds a $46,000 error is a better purchase than a $897 man-day that finds nothing, and no pricing model tells you which one you are buying.

One more piece of arithmetic that buyers get wrong. The fee does not scale with the order. The same three man-days that cost $897 on a $128,000 order cost $897 on a $2 million order, where they are four hundredths of one percent of the goods. The man-days are driven by the number of SKUs, the number of characteristics you actually care about, the sample size your lot size produces and the number of containers you want watched being loaded. Not by how much the furniture is worth. A $2 million order with one SKU and four checks is a cheaper order to inspect than a $40,000 order with eleven SKUs.

When not to inspect

Not every order is worth a man-day, and an inspection you did not need is $149 spent to learn nothing.

Two cases where the money is better spent elsewhere. If the order is small and the goods are simple, the cost of the error is capped by the order value and you can absorb it. If the error is already visible before you pay, the inspection is confirming something you already know.

There is a third case, and it is the commonest one. If you have already approved a production sample and the factory built to that sample, you have paid for the information once. What the inspection adds is the certainty that what shipped is what you approved, and that is worth having when the sample approval was informal, or when the factory built the sample from a different production run than the goods.

The honest position, and the one our pricing page takes, is that there are orders where we will tell you that an inspection does not pay for itself. On this order it paid for itself 557 times over. On a $6,000 order of 40 chairs with one SKU and nothing unusual, it would not have.

Seven clauses for the purchase order

The cheapest thing in this case was not the $897 inspection. It was the line that would have been in the purchase specification. Seven clauses, none of which cost anything, and every one of them written to be checkable rather than to sound strict.

• A governing hierarchy. Where the written specification and a drawing disagree, the written specification governs. The factory shall confirm in writing which figure it will build to, for every dimension that appears in both, before material is cut.
• A datum for every controlled dimension. Seat height 450 mm, measured to the highest point on the front of the seat with the cushion fitted, unloaded, on a level floor. A dimension without a datum is an argument waiting to happen.
• Named checkpoints tied to production. The factory shall hold the order unpainted and unpacked at 20 per cent production for a buyer inspection, and shall not load any container before a pre-shipment inspection has been completed and reported.
• A change notification clause. The factory shall notify the buyer in writing of any change to materials, drawings, suppliers, jigs, settings or processes, and shall not ship after such a change without written agreement. This one clause would have caught the last 60 chairs.
• A price reduction, not a claim. For each unit that does not conform, the contract price is reduced by the amount set out in the schedule, currently $280 per dining chair, being the cost of replacing it in the destination market. A price reduction is taken off the invoice. A claim has to be proved and argued.
• A holdback until it passes. 10 per cent of the order value shall be retained until the pre-shipment report is issued and any non-conforming units have been corrected and re-verified.
• Who pays for the second look. If a pre-shipment inspection records non-conforming units, the factory shall bear the cost of re-inspection after correction, and of any storage or delay caused by its own failure to conform.

The number to hold onto

The inspection on this order cost $897. It was seven tenths of one percent of the order value. It found 940 chairs that were 60 mm too tall, on the sixth measurement, with the goods still on the factory floor and the balance of $89,600 still in the buyer's account.

It also found a defect that no photograph would ever have shown, in a batch that passed the buyer's own approved sample, in an order whose paperwork was complete and internally consistent. The defect was in the space between two documents the buyer had sent, and both of them said 450.

That is what a check buys. Not a guarantee, not a certificate, not a warranty. A second pair of eyes on the specific list of things you have decided are worth caring about, at a point in the production where the answer still changes something. On this order it changed $500,000.