The direction nobody builds

A button specified for three seasons and bought for none

The Larkfall's bill of materials names seven corozo buttons, dyed to shade, from Vantry Trims. (Corozo is a hard nut, sometimes called vegetable ivory, cut and dyed into buttons.) It has named them for three seasons. Meranti has never bought one.

Corozo takes 45 days after shade approval. The tech pack's own trim window is 21 days. That is not a supplier failing or a planning failing. It is an arithmetic impossibility, present in the specification from the day it was written, and it has been resolved the same way three times running.

1. Purchasing finds the gap when it goes to buy.
2. A substitution request goes to Wildersmoor with a lab dip and a counter sample.
3. Bethan Coyle approves a polyester imitation, as a one-off, for this season only.
4. The order runs. Nothing changes in the specification.

Every step is correct. The substitution request was right to be raised. The approval was right to be given. Marking it as a one-off was right too, because a technologist approving a permanent substitution is doing something quite different and would want a different conversation. Three correct decisions, three times over, and the specification comes out the other end exactly as unbuyable as it went in.

Season cost, at Meranti's own rates: USD 186.00 in lab dips, samples and courier, plus 5.5 hours of merchandiser time at USD 9.40. That is USD 237.70 a season, and USD 713.10 over the three.

There is a second number, and it needs stating carefully, because it is not a loss. The cost sheet prices the specified button at USD 0.047 and carries USD 0.329 a garment. The substitute costs USD 0.038, so Meranti actually pays USD 0.266. Across the season's button-carrying styles, 118,400 pieces, that is USD 7,459.20 of difference between what Meranti quotes itself and what Meranti spends. It is not money lost, and it is not money found. It is a quotation built on a price nobody has paid since before the current season. You cannot say from here what it cost or gained, because that depends on orders that were and were not won. What you can say is that nobody knew.

Everything the order learns, it learns at the far end

Look at what the commitments system found out about the Larkfall in one season, and where each fact ended up.

The forward direction, specification to order, is the one every integration project is scoped around. It is visible, it has a name, and it fails loudly: somebody is chasing cloth by Thursday. The reverse direction has no name, no owner and no failure event. Nothing goes wrong on the day it does not happen.

One season, both directions

The forward column is larger, and it is the smaller problem. Here is why.

A forward-flow error has an owner. Somebody's name is on the purchase order. Somebody chased the cloth. There is an email thread. So it gets investigated and it gets fixed, one case at a time. Next season those three specific causes are unlikely to repeat.

A reverse-flow absence has no owner at all. No job description in the building contains tell the specification what purchasing learned. So it is not a failure that recurs. It is a run rate. The same four absences produce the same class of cost next season, and the season after, without anybody's performance getting worse. Three seasons at USD 2,733.94 is USD 8,201.82, and it grows with the number of styles rather than with the number of mistakes.

The closing step, and why it wins

The forward fix from lesson 3 was one field on a requisition: USD 301.00, returning USD 7,349.63 in its first season. That is 24.4 times.

The reverse fix is three fields at the other end. When a purchase order is closed, the person closing it answers three questions.

1. What was actually bought: the same item, or something else?
2. How long did it actually take, from order to receipt?
3. What was the real minimum or price break?

Where any answer differs from what the specification assumed, the item goes on a short list. The product team clears that list before the next season's tech packs are released. Meranti raised 212 purchase orders that season. At 40 seconds each, that is 2.36 hours a season, or USD 22.14 of office time, against USD 2,733.94 of recurring cost. That is 123.5 times.

Which is 5.1 times the return of the fix everybody agrees to build.

The method, in three questions per fact

You will not redesign your systems. What you can do, for every fact that matters on your order, is answer three questions and write the answers down.

1. Which system is the record of this? Use lesson 1's two tests: does a change move money, and can it be true at two versions at once.
2. When does it cross the seam, and what is stamped on it when it does? Lesson 3: at minimum, the revision it was copied from.
3. What does the other side learn afterwards, and who carries it back? Lesson 5: name a person, or accept the run rate honestly.

Three questions, and the third is the one nobody is asking.

Act as a product-development manager. You assume that everything the factory learned this season is still trapped in purchasing, and you want it measured rather than asserted. Facts: factory or brand [NAME], the season and the styles in it [LIST], the number of purchase orders raised [NUMBER], every substitution approved during the season [LIST EACH ONE WITH THE SPECIFIED MATERIAL, THE MATERIAL ACTUALLY BOUGHT, THE REASON AND THE DATE], every material whose real lead time or minimum turned out to differ from what the specification assumed [LIST WITH BOTH FIGURES], every production failure traced to a specified material [DESCRIBE], and the loaded hourly cost of an office hour and a factory minute [AMOUNTS]. Do the following. First, for each item, state what was learned, where it was learned, where it should have been written, and where it actually went. Be blunt where the answer is nowhere. Second, open the current specification for each affected style and say whether it still names the material that was substituted. Report the count as a fraction, because that fraction is the size of the reverse flow. Third, price the recurrence. For every absence, what did it cost this season, and what will it cost again next season if nothing is written back? Remember that nobody's performance has to get worse for it to repeat. Fourth, separate the items that are a genuine loss from the items that are only a difference between an assumed price and a paid one. Refuse to add the second kind into a loss total. State it separately, and say what is and is not knowable about it. Fifth, design a closing step for a purchase order that captures the three or four fields that would have prevented all of it. Tell me how many seconds each purchase order would take, multiply it out over the number of purchase orders raised, and compare that cost with the recurring cost. Sixth, rank every fix by return. Then tell me which one has an incident behind it and which one does not, because that is what decides which gets funded.