The integration nobody budgets: one bundle, five names

The situation

By the end of March the integration between Sumilon's execution layer and its planning system is six weeks late, and the invoices have passed USD 100,000 against a quotation of USD 19,600. Talvin Systems is not being unreasonable, and nothing has gone wrong technically. Every hour on the invoice went into one thing: agreeing what a bundle is.

One bundle, five names

Take a single physical thing. It is 24 cut front panels of style FL-318 in colour 407, size L, tied with tape, sitting on a trolley.

Five names for one trolley. The link between the two systems that had to talk is the cut ticket number. It is typed by hand into the execution layer when the cutting room confirms a lay.

That single hand-typed field is the whole integration. Sumilon's cutting room issues 264 bundle tickets on an average day. Measured over a fortnight, 2.4% of them arrive at the execution layer with a code that matches no work order, or matches the wrong one. On the FL-318 order, which is 4,000 bundles, that is 96 orphan bundles and 2,304 pieces of production that landed in a holding account nobody reads.

The four cases no amount of typing accuracy fixes

Getting the code right is the easy problem, and a barcode solves it. These are the ones that cost money.

1. Re-cuts. A panel is damaged, the cutting room cuts a replacement, and it issues a new bundle number. The execution layer opens a new bundle against the same work order. The planning system still holds the original order quantity. Unless somebody voids the original bundle, the same 24 garments are now counted twice. On FL-318 there were 214 re-cut bundles, and 71 of them were never voided. That is 1,704 pieces of production that never existed. Lesson 6 meets them again.
2. Splits. A bundle handed to two operators to catch up becomes two bundles in the execution layer and stays one in the cutting room's sheet. The systems now disagree about a count they were built to reconcile.
3. A colour change inside a lay. Cutting numbers by lay. The buyer buys by colour. One lay carrying two colours is one number upstream and two downstream, in both directions.
4. Three different levels of detail. The planning system holds quantity by size. The execution layer holds single-size bundles. The packing system holds cartons packed to a size ratio. Nothing links a carton to a bundle without going up to the order and back down, and the moment a carton is short-packed the link breaks.

None of these is a defect. All four are ordinary garment manufacturing, and all four have to be decided, written down and built. That is what the invoice was for.

Where the money actually went

Against a quotation of USD 19,600. The programme finished at USD 329,014 rather than USD 213,714, and the integration is 41.0% of it rather than the 9.2% the quotation implied.

And it could have been estimated. Not to the dollar, but to roughly the right size, on a whiteboard, in an afternoon, before anybody signed anything.

The count that predicts the bill

Here is the estimate Rowena now makes before an integration is quoted.

1. Take one physical thing that moves — a bundle, a roll, a carton.
2. List every system that holds an identity for it, and every place a human creates one of those identities. Sumilon's bundle: five identities, two of them created by a person.
3. List every event that changes the identity: split, merge, re-cut, re-number, cancel, colour change, size change.
4. List every place two systems hold the same quantity at a different level of detail.

The bill is driven by lines 2 to 4, and the quotation is almost always priced on line 1. A supplier quoting an integration is quoting the pipe. The cost is in the events and the levels of detail, and those belong to your factory, not to the supplier. That means you are the only party who can count them, and you can do it before you ask for a price.

The cheapest thing was done last

The rebuilt cut-ticket entry cost USD 19,300 of the invoice. What it replaced was a barcode on the bundle ticket. That needed a thermal printer in the cutting room at USD 640, and labels at USD 0.004 each: 264 bundles a day over 260 days, or USD 274.56 a year.

It went in eleven months after the sensors, because it was not a sensor, was not software, and belonged to a department that had not been in the meeting. The key that links the two systems was the cheapest component in the programme and the most expensive thing to get wrong, and it was the last thing anybody looked at.