Why growth slowed in week 27.

1. The week in question

The headline drop is 10.1 points. Splitting it matters, because nearly three quarters of it is arithmetic rather than performance:

2. More customers came in. Fewer of them bought.

Sixteen stores report door count across the whole six-week window. Their traffic, transactions and conversion, week by week:

That is not a demand problem. The customers arrived; the sale did not happen.

2.1 What the conversion drop cost, in one week

Had week 27 converted at week 26's rate, its 214,994 visitors would have produced 39,707 transactions rather than 37,539. At the week's own average ticket of $46.48 that is roughly $101,000 of sales, in one week, across sixteen stores — before counting the nine sites without door counters.

2.2 And the baskets themselves held

If shoppers were being turned away by empty shelves, the baskets that did get rung should show it — thinner, fewer add-ons, more single-item sales. Across 214,818 transactions in the window, they barely moved.

There is a compositional shift underneath — the share of baskets containing each department moved more than the totals suggest:

The same test at store level returns the same answer. Across the ten largest stores, in-stock fell between 3.0 and 6.7 points, while units per basket moved between +0.26 and −0.16 — noise, with no relationship to the availability change. One store lost 6.7 points of in-stock and its basket was flat; another lost 4.9 points and its basket grew.

3. The root cause: the buy has not kept pace with the sell

For every four units sold this year, three have been bought. The difference came out of inventory. The warehouses absorbed it for most of the year — which is precisely why nothing appeared in the sales line until the buffer thinned.

Store cover is units flowing into stores divided by units sold. Above 1.0 the network builds stock; below it, drains. The warehouses drained nearly three times faster than the stores — −19% against −7% — which is the signature of a distribution network working correctly against an inbound supply that is too small. The goods were not stuck in the warehouse. They were never bought.

4. Where it bites: the A stores

The seven highest-selling stores carry the trade, and every one of them was short-fed over the six weeks to 8 August.

Meanwhile a mid-volume store sits on 11.6 weeks' cover and 37 units per square foot, roughly double the chain norm. The surplus is not where the trade is.

4.1 And the stock the A stores hold is the wrong stock

At the busiest stores, 19–22% of units sit in styles with a broken size run, and only about a third of styles are complete on both size and colour. Four in five of the missing sizes do not exist anywhere in the network — which is the replacement ratio showing up on the shop floor. Runs break as they sell through, and there is nothing behind them to repair with.

5. What we had to build to answer the question

We rebuilt it. Anchoring on the current snapshot and walking the movement ledger backwards produced a daily stock position for every SKU at every site: 3.8 million rows across 43 days, with the anchor day reconciling exactly to the live snapshot and zero day-over-day discontinuities. Every in-stock and cover figure in this document comes from it.

6. What it is worth

From the companion distribution analysis, measured on the same database:

A 21-point improvement in service level using stock the business already owns. No purchase order, no additional working capital.

7. Proposed engagement

Four workstreams. Each stands alone and delivers on its own; none waits on the one after it.

8. Basis, and what this analysis does not claim

• Full dataset, no sampling. Every figure is measured against a full copy of the live trading data. Trading, availability and movement figures cover retail weeks 22–27, complete to the end of week 27.
• Availability is measured, causation is inferred. The chain of evidence — receipts below sales, warehouses draining, in-stock falling, traffic rising while conversion falls — is consistent and each link is measured. It is not a controlled experiment, and the per-store correlation between availability and sales is not strong enough to attribute a precise revenue figure to any single store.
• Traffic is measured within the six-week window only, on a fixed panel of the sixteen stores reporting door count throughout. Year-on-year traffic comparison is excluded because counter coverage changed during the year. The $101,000 figure covers those sixteen stores for one week and is not annualised.
• Unmet demand is not lost revenue. Customers substitute, defer and return. The recoverable figures above are gross exposure; a measured substitution rate is a workstream 1 deliverable rather than an assumption baked in here.
• Full size and colour runs improve availability, not obviously sell-through. Across all seventeen stores, complete styles are in stock materially more often but do not demonstrably sell through faster. The presentation rule is recommended on the availability evidence, and the sell-through question is left for a controlled pilot.