Where the scarce counts go

Ranking is the decision, not the counting

A store can only count so much. So every counting plan is really a ranking. This item gets looked at often, that one twice a year, and somebody decided which was which. The decision is usually made on annual spend, because spend is the number every system already holds. This lesson asks whether spend is the right axis, and builds the alternative from a store's own records.

Where lesson 1 left the problem

Sambok can release 1,150 hours a year to counting without adding a head. At the measured 9.5 minutes a count, that is about 7,263 counts a year for 1,480 items.

Lesson 1 derived a counting interval from two things: the rate at which records break, and a target for how many items may be wrong at once. It also said, at the end, that the target was the weakest thing in the calculation. It was the same 5% for a roll of shell twill and for a polybag. Shell twill is the outer cloth of the trouser; a polybag is the plastic bag one finished garment goes into. Being wrong about those two things does not cost the same, so holding them to the same exposure spends the scarce counts in the wrong place.

Everything below is about fixing that. The counting arithmetic is unchanged. Only the question of which items get the short interval is on the table.

The usual answer, and where it comes from

Nou Ratanak inherited a plan built on annual issue value, which is what a year's worth of an item is worth. The items carrying most of the money get counted most often. Sambok's stock resolves into ten families that between them hold 512 of the 1,480 items.

That is the familiar shape. Shell twill alone is 55.8% of the money, and the top three families are 77.5% of it. That is why the value ranking feels so convincing: a small number of items really does carry most of the spend.

The inherited class A, the 220 items counted every four weeks, is exactly the top four of that table: 62 + 38 + 24 + 96 = 220. Nobody at Sambok chose those four families. The value ranking chose them, years ago, and nobody has looked since.

The other axis: what being wrong costs

Value answers the question where is the money. A counting plan exists to answer a different question: what will stop me, or cost me, if the record is wrong. Those are the same question only when every item's error costs the same, which is never true in a garment factory.

So build the other axis from Sambok's own 2027 record. Of the 673 discrepancies the year raised, 118 were consequential, meaning big enough to change something rather than to be absorbed. Each one was priced at what it cost if it was found at the point of use rather than by a count. That figure is called the exposure.

Read the third column before the fourth. The trouser zip costs more per error than the shell twill does, USD 2,840 against USD 1,960, even though a zip is worth a fraction of a metre of cloth. The reason has nothing to do with the zip's price. The zip is named by Nipigon Outfitters, it is imported, and it is on a 34-day lead time. So a shortfall cannot be covered locally at any price, and the line either waits or the order ships short. The twill can be had from a converter nine days away at a premium.

The cost of an error is set by what you can do about it, not by what the thing is worth. That sentence is the whole of this lesson, and every re-ranking below is a consequence of it.

Nine of the ten families move, by 2.8 places on average. Only shell twill is in the same place on both lists, and that is exactly why the value ranking survives. The one item everybody would have picked anyway comes first either way, and that is taken as evidence the method works.

The unit matters, and it is not the family

There is a step here that is easy to skip and changes the answer again. A count is spent on an item, not on a family, and the families are wildly different sizes: 96 thread items against 19 interlinings. Ranking families by total exposure quietly rewards a family for being large.

Divide through.

Fusible interlining is seventh by value and fourth by family exposure, and it is first per item. Fusible interlining is a stiffening cloth glued onto a panel with a heated press, to hold the shape of a waistband or a pocket flap. It is the cheapest thing on the list that can stop a line: nineteen items, a fusing press that will not run without them, and no substitute that passes a bond test. It is also, at Sambok as at most factories, kept on a shelf nobody signs for.

The two exposure rankings pick the same families for the top of the list and disagree about their order, which is the correct relationship between them. The family total says how much is at stake in a category. The per-item figure says where a count goes.

What it is worth changing

Now the comparison that decides whether any of this was worth doing. Hold the counting hours roughly constant and change only the membership of the four-week band.

inherited class A, by value:      220 items    exposure covered  USD 77,320  = 56.4%
by exposure per item:             234 items    exposure covered  USD 125,360 = 91.4%
                                  14 more items, 182 more counts a year, USD 40.92

The second band is fusible interlining, shell twill, lining taffeta, trouser zip and the woven main label. Those are the first five rows of the per-item table, and they come to 234 items. For USD 40.92 a year of extra counting, the share of Sambok's measured exposure sitting inside the four-week band rises from 56.4% to 91.4%.

Thirty-five points of exposure for forty dollars. That is the finding, and it is not a clever one. It is what happens when a plan that was never questioned is questioned once.

The whole plan, rebuilt

inherited: 6,240 counts = 988.00 h, leaving 162.00 h of the 1,150 in reserve
adopted:   6,090 counts = 964.25 h, leaving 185.75 h in reserve
difference: 150 fewer counts, USD 33.73 less labour a year

The adopted plan is cheaper than the one it replaces and covers 35 points more exposure. It also looks identical from a distance: three bands, the same intervals, the same shape of table. Only the membership changed, and the membership was the entire decision.

One thing it deliberately gives up. Lesson 1's target was that no more than 5% of items in a class should be wrong at any moment, and the bottom band at 26 weeks sits at 5.93%, outside it. The adopted plan accepts that, because the items now in the bottom band are the ones a wrong record costs almost nothing on. A uniform exposure target was the thing this lesson set out to reject, and honouring it here would have undone the argument.

Where the ranking stops being trustworthy

This is the section most rankings do not have, and it is the reason to trust the rest.

The event counts are small

Every exposure figure above is an event count multiplied by a cost. The event counts are single- and double-digit, which means they carry a lot of sampling error. These are counts of rare, independent events, and there is a standard way to put a range around such a count.

Look at what overlaps. Shell twill's 26 and lining taffeta's 14 have ranges that share everything from 16.98 to 23.49. The first and third families on the exposure list are not separated by their event counts. The four families in the middle, at 11, 12, 13 and 14 events, are one undifferentiated cloud.

Carry that through to the exposure figures themselves, holding the cost of a late discovery fixed:

shell twill  USD 50,960   range 33,281 to 74,656
trouser zip  USD 25,560   range 11,701 to 48,507
polybag      USD    380   range      104 to      973

The zip's range runs from well below the twill's point figure to well above it. What survives is the separation of the top of the list from the bottom. The polybag's entire range sits below every other family's lower bound, and nothing else survives.

So the ranking sorts items into bands. It does not sort them into places. Use it to decide who is in the four-week band and who is in the six-month band. Do not use it to argue that the zip should be counted more often than the lining, because on this evidence you cannot tell.

And the ranges above are too narrow, not too wide. They treat the cost of a late discovery as exact, and lesson 1 showed shell twill's USD 1,960 is the mean of six numbers running from USD 210 to USD 6,840. A range that accounted for both would be wider still.

Most items have no evidence at all

968 of Sambok's 1,480 items produced no consequential error in 2027. Not one. A strict exposure ranking scores every one of them at zero and would never count any of them again.

That is not conservatism. It is a trap that closes on itself. An item that is never counted never produces the discrepancy that would have raised its score, so its score stays at zero because it stays at zero. The evidence and the plan feed each other, and after two years the store is confident about exactly the items it was already confident about.

The floor is the answer, and it is not negotiable. Every item is counted at least twice a year whatever the ranking says, and the floor is costed first, before any band is allocated:

floor:      1,480 items x 2 counts   = 2,960 counts = 468.67 h
close watch:  234 items x 11 more    = 2,574 counts = 407.55 h
middle:       278 items x  2 more    =   556 counts =  88.03 h
                                       6,090 counts = 964.25 h

The floor is 48.6% of the counting the store does. It buys almost nothing on the evidence available, and it is the part of the plan that finds the thing the evidence could not have predicted.

The evidence is a year old and the order book has moved

The 2027 events are about 2027's styles. Sambok's 2028 book moves NP-3306 onto a nylon shell with a different zip supplier, and the interlining changes with it. Re-derive the ranking every year, and treat the first year of any new material as unranked. That is precisely what the floor is for.

And the cost column is not innocent either

The cost of a late discovery was priced from what actually happened. So it is measured only for items that have gone wrong at Sambok, and it is measured under Sambok's own recovery options: a converter nine days away, an air freight lane it already uses. Change either and the column changes. This is the number a reader must not copy. It is a measurement to take on your own operation, and there is no shortcut to taking it.

Help me decide where my scarce counting attention goes, and then tell me where the answer stops being trustworthy.

Ask me for two lists. The first is my items grouped into families, with the number of items in each and the annual issue value of each. That is the ranking I probably already have. The second is harder, and it is the one that matters. For each family, ask how many times in the last year a discrepancy in it was consequential, meaning it changed something rather than being absorbed, and what each of those events actually cost when it was found at the point of use rather than by a count. Ask me to price those from real events, and if I have never priced one, tell me to start now rather than letting me estimate.

Then build both rankings. Rank the families by annual issue value. Rank them by consequential events multiplied by the cost of a late discovery. Show me the two orders side by side, with the rank movement of each family. Then divide the exposure by the number of items in the family and rank a third time, because a count is spent on an item and not on a family, and tell me which families change place when you do.

Then make it a decision. Hold my counting hours roughly constant, tell me which families should be in the short-interval band under each ranking, and give me the share of my total measured exposure that each choice covers. Give me the difference in counts a year and in money.

Then do the part I did not ask for. Give me a confidence range for each event count, treating them as counts of rare events, and tell me plainly which families are genuinely separated by the evidence and which are indistinguishable. Say explicitly that the ranking sorts items into bands rather than into places, and refuse to argue for one family over another when their ranges overlap.

Finally, three refusals. Refuse to score an item at zero just because it has no events yet, and insist on a floor frequency that every item gets whatever the ranking says. Refuse to fill my cost column with unit price or lead time as a stand-in for a consequence I have not measured. And tell me that the ranking is about last year's styles, so it has to be re-derived every year, and a new material is unranked until it has a history.