Inspection labour is the part of a mill's cost base that hides in plain sight. It appears as a headcount figure on a monthly sheet, and after a while nobody looks at it again.

That figure is the smallest part of the cost. Fatigue, missed defects and turnover all sit behind it, and none of them show up on the wage line.

This article splits the cost into three parts. It then says which part AI inspection reduces, and which jobs still need people.

What does the wage sheet leave out?

Most mills measure inspection labour as how many people stand at the table, and that figure is the only one that reaches the accounts.

Three costs sit underneath it.

Cost	How it shows up	Where it lands
Headcount	Inspectors per shift, per table	Wage sheet
Missed defects	Faults that pass the table and surface later	Claims, rework, freight
Turnover	Training time for each replacement	Quality variance, supervisor time

Only the first row is easy to count. The other two arrive months later, and they arrive as somebody else's problem.

First cost: how many people does a shift need?

Working back from your own speed

This one is arithmetic, and you can run it with your own numbers.

Take your daily output in metres. Divide it by the inspection speed you actually achieve. Manual inspection runs at 10 to 15 metres per minute, and the upper end assumes good light and easy fabric.

Daily output	At 15 m/min	At 12 m/min
6,000 m	6.7 h	8.3 h
8,000 m	8.9 h	11.1 h
12,000 m	13.3 h	16.7 h

A single eight-hour shift stops covering 8,000 metres as soon as the real speed drops below 15 m/min. The gap gets filled by a second shift, or by overtime, or by fabric that leaves the table unread.

The third row is the one that costs the most over time. A mill that replaces one inspector a year, across three tables, trains nine people in three years, and every one of them starts from zero.

Ask your supervisor what speed the table really runs at, on your worst fabric, in the last hour of a shift.

Second cost: how long does attention last?

Manual inspection is a visual task with a fixed attention budget. Backlit inspection of a difficult fabric is not the same job all shift, and the fourth hour is not the first hour.

Fatigue does not announce itself. An inspector who has missed a fault does not know it, and neither does the supervisor watching the table.

Industry practice puts manual defect detection well below what a first-hour test suggests. The drop is not a training problem, and no training programme removes it.

How to measure attention loss in your own mill

Take one roll of your hardest fabric. Run it twice with the same inspector, once at the start of a shift and once after four hours. Compare the two fault counts, and you have the attention loss as a number you can put in front of a manager.

AI inspection inverts that curve. The stated figure is 24-hour continuous operation at 100% attention, which is a description of what the equipment does rather than a claim about people.

Third cost: hiring, training and turnover

What turnover resets

Inspection is a job that takes weeks to learn, and it is often paid at entry level. Both of those facts cost the mill money, and neither one appears as a separate line.

A new inspector needs time on real fabric before their judgement is worth trusting. During that period the table runs with two people, or with one person and a higher miss rate. Neither version shows up in the accounts as a training cost.

Turnover also resets institutional memory. An inspector who has seen three years of your fabric knows which faults matter for which customer. That knowledge walks out of the gate with them.

Count the replacements over the last two years. Add the weeks each one needed. That is the third cost, and it is usually larger than mills expect.

One more figure worth finding is the average tenure of your inspection team. If it sits under two years, the training cost is not a one-off. It is a running cost, and it repeats every time somebody leaves.

Which part does AI reduce, and which part stays?

What AI changes, in three rows

AI inspection touches all three costs, but not equally.

Cost	What AI changes	What it does not change
Headcount	The specification states replacement of 3 to 5 inspectors per line	Re-inspection, grading and exception handling still need people
Missed defects	Consistent attention across the shift	No guarantee on a defect type never seen in training
Turnover	Fewer routine inspection posts to fill	Somebody still trains the people who remain

The labour cost reduction quoted for AI inspection is around 70%, and the missing 30% is the interesting part. It covers the jobs that need judgement: deciding whether a flagged area is acceptable, handling a customer claim, and agreeing what a new fabric style should look like when it is running well.

Those jobs do not disappear. They move up the skill ladder, and there are fewer of them.

Where the defect data goes afterwards is a separate subject, and how AI fabric inspection changes a quality control process covers it in detail.

What decides whether the payback holds

Three figures carry the labour-cost case, and you can work all three out before you ask for a price.

· Metres per minute on your fabric, not the catalogue maximum.

· Inspectors replaced on your line, with the number of shifts and tables stated.

· What the machine does not cover, in the supplier's own words.

Our ST-Thinkor AI fabric inspection machine is specified at speeds up to 80 m/min across fabric widths of 1,800 to 4,000 mm, with a detection rate of ≥90% and a minimum defect size of ≤0.1 mm on single-colour fabric. It is one of a range of fabric inspection machines, which differ mainly in working width and inspection speed.

Where the cost actually sits

The inspectors at the table are the visible part of inspection labour, and the smallest part. What sits behind them is the fabric that leaves the table unread, and the weeks of training that restart every time somebody leaves the job.

AI inspection changes the scanning share of the work, which is most of the hours. Judgement stays with people: fewer of them, at a higher skill level.

Run the sum on all three parts before you compare it with a price. A payback case that uses only the wage line will look better than the real one.