The router optimises the parcel and pays for it in margin

The week the DC ran dry

A website sells something the warehouse has run out of. Many retailers now hand the order to a shop that still has one. It looks like a free rescue: an order that would have been refused becomes a sale. This lesson prices that rescue properly. It turns on a question the software never asks. What would that shop have done with the garment?

Week 7, the week starting 18 May 2026. Corvedale's website took 412 orders for CRV-4120 mid indigo that week. Hartswood could fill 268 of them. The other 144 were all size 12 or size 14, and Hartswood had none left in either.

Before omnichannel, those 144 orders were cancelled and that was the end of it. Corvedale's order-management system does something better. When the DC cannot fill a line, it looks for a shop that can, and sends the order there to be picked, packed and posted by a colleague. This is ship-from-store, and it is the single most valuable thing a retailer gets from putting shop stock on the website. A chain with 62 shops has 62 more places to send parcels from, and the stock is already paid for.

The router filled all 144. Monday's dashboard reported 144 orders saved and GBP 8,496.00 of revenue rescued — 144 pairs at GBP 59.00, which is arithmetically true.

The margin effect of that same week was GBP 883.20. It should have been GBP 1,238.40.

Price the decision, not the parcel

The unit of decision is one order line the DC cannot fill, with at least one shop holding the size. There are exactly two things Corvedale can do with it.

Ship from the store. A colleague leaves the shop floor, finds the pair, packs it by hand, and puts it in the carrier's evening collection.

A shop parcel costs GBP 5.45 more than a DC parcel. That gap is real. It is also the least interesting number here, because it is not what the decision turns on.

Cancel the order. Corvedale prices a cancellation at GBP 6.80: GBP 2.30 of contact-centre handling and refund processing, plus GBP 4.50, which is its own valuation of the measured drop in repeat buying among customers whose order it cancelled. That second figure is Corvedale's own, from Corvedale's data, and you can argue with it. The sensitivity test at the end of this lesson shows exactly how much rests on it, which is the honest way to use a number like that.

Now the part that decides everything. Where does the unit come from, and what would that shop have done with it?

There are two answers, and they are not close.

Every figure in that table is in pounds, and each branch uses exactly one pair. So the GBP 22.42 the pair cost appears on both sides and cancels out. Comparing only what differs is not a shortcut. It is the whole method. Put the cost of goods in and you end up comparing two numbers that both contain it, and concluding something about neither.

Read the two rows in words.

Shipping from a shop that was going to sell the pair anyway does not create a sale. It moves one, from the shop to the website, and charges GBP 10.50 for the move. Against that it saves the GBP 6.80 cancellation. The net is a loss of GBP 3.70, every time, on every such pair.

Shipping from a shop that was going to end the phase with the pair unsold turns a GBP 29.50 clearance into a GBP 59.00 full-price sale. After the cost of fulfilling it and the cancellation avoided, that is GBP 25.80 — nearly seven times the size of the loss on the other row, and the reason ship-from-store exists.

The router does not know which shop it is taking from

Corvedale's router ranks candidate shops by distance to the delivery address, then by whether the shop has room in tonight's collection. That is a sensible rule and somebody sensible wrote it. The nearest shop means the cheapest carriage, the fastest delivery promise and the lowest chance of a missed collection. All of that is true.

It also has nothing to do with the only thing that matters — and it is worse than that, because online orders cluster where people live, and Corvedale's biggest and fastest-selling shops are in exactly those areas. Ranking by distance is, in practice, ranking by rate of sale, the wrong way up.

Corvedale grades its shops: 10 grade-A shops in the largest catchment areas, 22 grade-B, 30 grade-C.

Cover is the forward-looking figure from track 17: units held divided by forecast weekly sales. It says how many weeks the stock will last. A shop with 1.8 weeks of cover and 3 weeks of phase left will sell everything it holds in that size before the phase ends. A shop with 5.4 weeks of cover will not.

So the test is one comparison — is this shop's cover for this size less than the weeks remaining? — and it sorts the 144 pairs cleanly:

• 96 pairs came out of shops with cover under 3 weeks: all 88 from the grade-A shops, plus 8 from grade-B shops selling ahead of their band. Every one of those was going to sell at full price where it stood.
• 48 pairs came out of shops with cover at or above 3 weeks: 33 grade-B and all 15 grade-C. Those were heading for clearance.

Two pulls in three came from the wrong shop. Not because anyone chose badly, but because the router was asked to keep carriage cheap, and it did.

The week, in money

Three readings of the same week, all correct. The difference between them is what this lesson is for.

The dashboard says GBP 8,496.00 of revenue rescued. The programme is worth GBP 883.20 against not having it. A router ranking by cover instead of distance would have been worth GBP 1,238.40 — 40% more, from changing one sort order.

How much of this rests on the GBP 6.80

Enough to state plainly, which is why it is stated.

The GBP 4.50 repeat-buying part is a judgement Corvedale made about its own data. Move it and the answer moves with it.

If a cancellation costs nothing, the whole week's programme is negative: 144 parcels, GBP 8,496.00 on the dashboard, and GBP 96.00 of margin destroyed. The week breaks even at a cancellation cost of GBP 0.67. So Corvedale does not need to defend GBP 6.80 to justify having ship-from-store at all. It needs GBP 6.80 to be above GBP 0.67, which is not a hard case to make.

What GBP 6.80 does not do, at any value below GBP 10.50, is make a pull from a fast shop worth taking. The programme is right and the routing is wrong. Those are two separate arguments, and they get lost in each other every time somebody defends the whole thing as one thing.

Act as a retail merchandiser who is paid to protect margin and has no attachment to a fulfilment metric. I want my ship-from-store programme priced properly for one style over one period. My inputs: style [CODE], retail price [AMOUNT], landed cost [AMOUNT], first markdown [PERCENT], clearance price [AMOUNT], distribution-centre fulfilment cost [AMOUNT] a parcel, store fulfilment cost [AMOUNT] a parcel, and my own costing of a cancelled order [AMOUNT] with its parts shown as [WHAT IS IN IT]. The period is week [NUMBER] of [NUMBER], so [NUMBER] weeks remain. Orders the distribution centre could not fill: [NUMBER]. Routed to shops: [NUMBER]. For each pull, here is the shop, its grade, and its weeks of cover for that size at the time: [PASTE THE LIST]. Do the following. First, build the per-unit comparison of ship against cancel for the two cases — a shop that would have sold the unit, and a shop that would have cleared it — and show that the cost of goods appears on both sides and drops out. Second, split my pulls into those two groups using cover against weeks remaining, and give me the period total, the best available total if the router had ranked by cover, and the difference. Third, put beside those the number my dashboard reports as revenue rescued, and say in one sentence why they are all correct and all different. Fourth, run the sensitivity test on my cancellation cost: give me the value at which a pull from a fast-selling shop breaks even, and the value at which the whole period breaks even. Fifth, write the routing rule I should ask for, in one sentence an engineer could build. Do not tell me the programme is good or bad as a whole. Separate the programme from the routing.