Choosing what to buy, and the four blanks

Three mechanisms, and the thing that actually decides between them

There are only three ways a sewing floor records what it has made, and the choice between them is almost never a technology question.

An overhead carrier system. Garments hang on carriers that are routed past each station. A reader at each station records the carrier as it arrives, because it has to pass anyway. Capture costs nobody a second. It is complete by construction, and it is piece-level for nothing.

A ticket coupon and a clerk. The operator tears a coupon off the bundle ticket she is already handling and drops it in a box. A clerk keys the box at the end of the shift. It costs the operator about a second and a half extra, and the data arrives fourteen hours late.

A terminal at each station. The operator stops and records. The cost of that is the whole of lesson 2.

Course 23.5 covers the reading technology itself: code schemes, check characters, read rates. None of that decides this. What decides it is how often the factory changes what it makes, and that number is on the order book, not in the quotation.

On a floor that runs long batches, a carrier system costs about six times a terminal system a garment. On Elmarj's own mix it costs 12.5 times: thirty-eight styles a year, with one and a half line-days lost to re-routing at each change. The whole of that difference sits in a column the quotation does not have, because re-routing is a production loss rather than an invoice.

And the honest reading of that table

A carrier system is never bought for the data. Its data benefit, on two lines, is the constraint capture cost it removes: USD 4,500 a year, against USD 62,000 a year of capital, which is 7.3%. Any business case that justifies overhead carriers on production visibility is short by a factor of about fourteen and should be sent back.

Carriers are bought to move garments, and to hold down the stock of half-finished work sitting on the floor. Course 5.6 prices that side of it. The data comes free afterwards, and that is exactly why the data is good:

The best production data in any factory is the data that nobody paid for, because it is a by-product of something the factory was doing anyway. Data that somebody had to be persuaded to produce degrades the moment they are busy, and they are busiest when you most need it.

What Elmarj actually bought, in the end

Three options were costed against the ten lines' baseline: one decision point a shift and an hourly hand count.

All three columns are US dollars a year. The second option is the one the original business case described, and it is the one most factories buy. It loses money against doing nothing but changing a standing instruction. It is not a bad system. It is a good system asked to earn its keep in a place where three quarters of the available gain was free.

The third option is the same technology bought differently. Terminals go only at the three stations that most often set the line's speed, and the record is made by the helper who feeds the station rather than by the operator, so the capture is a hand-off that was happening anyway. It captures most of the station-level detail, all of the decision-point gain, and its capital is a tenth.

Same equipment, same supplier, same floor, and a difference of USD 53,743 a year between the two ways of buying it.

The four blanks

Before any of this is quoted for, write four things down. If any of them is blank, the data has no value yet. Not low value. None. The honest thing is to fill the blank first, because filling it is usually free.

1. The decision. Name it, in a sentence, in the form we would move X. Not "improve visibility".
2. The person. Who takes it, by name and by role. If it needs two signatures, name both.
3. The frequency. How often that person is allowed to take it. This is the number from lesson 4, and it is the one that caps everything.
4. The value. What one instance of that decision is worth, taken earlier. A number, with its arithmetic.

Elmarj's four, filled in: we would move a floater onto the station setting the line's speed · the line supervisor · four times a shift, none in the last hour · USD 109.52 an event, 412.5 events a year.

That is a page. It is also the only part of the specification that anybody will still be reading in three years.

Where this course stops

This course is about the capture, and what it costs the people making it. What the machines themselves can report without anybody's help, and how those signals are carried, is course 24.2. What to do with the numbers once they exist, what belongs on a screen, what belongs in a report, and what nobody should be shown at all, is course 24.3.

The order matters. A screen built on a record that thins out under pressure will be confidently and consistently wrong in the same direction, and no amount of presentation repairs it.