What an audit tests, and what a factory prepares for

Five weeks of paint

Brindlemoor's initial social-compliance audit was booked for 18 May 2026. Serapion had five weeks' notice and used all of them.

The entrance was repainted and landscaped. New uniforms were issued. Policy posters were laminated in two languages and hung on every floor. The fire extinguishers were re-tagged, the first-aid boxes restocked, the canteen deep-cleaned, and every supervisor was taken through a briefing on how to answer questions. Total spend: USD 9,180.

The audit raised four non-conformities. Preparation touched none of them.

This lesson explains why. It is the most useful thing in the course. Once you can see the gap between what a factory prepares and what an auditor tests, you can close it in two weeks.

An audit tests records against each other

An auditor cannot see last March. What an auditor can do is take three separate records of the same day and check whether they agree. That is the whole method. Once you understand it, the preparation writes itself.

Serapion's finding on 14 June 2025 is the clearest example in this course of correct decisions adding up to a bad outcome.

Line 7 recorded 1,006 pieces of a five-pocket trouser that day. The style's standard time is 21.5 minutes. So the line earned 21,629 standard minutes of work. (Standard minutes are the work content of a garment, at a normal pace.) The line has 48 operators. It has never run above 64% efficiency, and it was costed at 62%. Take the output at face value: 21,629 minutes of work at 62% is 12.1 hours per operator. The attendance record for that day shows 9.0.

Read it the other way and it looks worse. 48 operators for 9.0 hours is 25,920 attendance minutes. Producing 1,006 pieces in that time means an efficiency of 83.4%, in a factory whose best line has never passed 64%.

Nobody falsified anything. The line worked the Saturday. Payroll booked those hours into the next pay period, because that is how the payroll calendar has always been cut. Production booked the output on the day it was made. Security logged the gate honestly. Three departments each did their own job correctly, and the three correct records contradict each other.

The auditor does not have to prove wrongdoing. The finding is simply that the records do not agree. That finding gets written whether the cause is a payroll habit or something worse. From outside the building, the two look identical.

The other three findings

Undeclared production off site. Embroidery on three styles goes to a workshop 400 metres away. Nothing was hidden. Nobody had ever been asked. Serapion did not know that Brindlemoor's terms require every unit of production to sit at a declared and audited site. Closing it costs USD 2,400 and six weeks to get the workshop assessed, or USD 14,800 to buy a six-head machine and bring the work in-house.

Records retention. Brindlemoor's programme requires twelve months of time and payroll records. Serapion keeps six, because six is what the labour authority asks for. That is entirely correct locally, and still non-conforming to the programme. It is the only finding on the list that cannot be closed with money. You start keeping them today, and full conformity arrives when the calendar arrives.

Worker interviews. The auditor interviewed 18 people out of 740. That is a sample of 2.4%. What that number means is the subject of the last section.

What the preparation bought

Two readings of that table matter.

First, Serapion spent 1.64 times as much on things the audit does not examine as on the things that closed findings: USD 9,180 against USD 5,590.

Second, the guarding and electrical work sits in its own row on purpose. It was required by law, it was overdue, and it had to be done whether Brindlemoor existed or not. Counting it as a cost of winning the account would make the paint look better than it was.

Nine weeks, and what they cost

The corrective-action window ran from the audit on 18 May to the closure verification finishing on 10 August 2026: nine weeks.

Brindlemoor's autumn placement needed an active vendor code by 8 June. Serapion's was activated on 10 August. So the autumn programme went to another factory: 26,000 pieces at a contribution of USD 1.12 a piece, or USD 29,120. Serapion got the winter trial instead, at 14,600 pieces.

So the price of that audit was not the audit fee. Against the USD 3,900 invoice, the nine weeks cost 7.5 times as much in lost contribution. That is 5.21 times the whole cost of closing every finding.

A clean audit is a sample that found nothing

Eighteen interviews out of 740 people is 2.4% of the workforce, on one day, chosen by an auditor with a few hours. A factory that treats the audit as its measurement system is measuring itself at 2.4% coverage, once a year, by somebody else's hand.

This is not a criticism of auditing. Sampling is what makes auditing possible at all, and a good auditor knows exactly what their conclusions rest on. It is a criticism of using someone else's sample as your own control. The factory that matches four random days a month knows something about itself that no annual audit can tell it. It is also, usually, the factory that passes.

What to take away

1. An audit tests whether separate records of the same day agree. Prepare by matching them yourself, not by painting.
2. Three departments can each be correct and still produce a finding, because a habit in one system is a contradiction in another. Find your habits before an auditor does.
3. Undeclared subcontracting is a finding even when nothing is hidden. Declare every site that touches a unit of production, including the workshop over the road.
4. Some findings cannot be closed with money. A records-retention gap closes at the speed of the calendar, so the day to start is today.
5. The audit fee is the smallest number in the whole exercise. The cost is the weeks, and the weeks are paid in placements you were not offered.
6. A clean audit is a sample that found nothing. It is not a measurement of your factory, and it should not be your only one.