Our store had size variants our warehouse never stocked.
A customer bought a 3-4Y in a style that doesn’t come in 3-4Y. It was live. It did not exist in our warehouse.
We cancelled and refunded the $16.24 within the hour.
How we found out
Three sources, same answer:
Our inventory database: no 3-4Y
That day’s export, 3,971 rows: no 3-4Y
Shopify: by then the variant was gone
The order line still carried the SKU. The variant it pointed to was already deleted.
Why it happened
An audit found 231 size options across 64 styles that never appear in our warehouse. We flagged 84 of them on 10-21, then removed them. This particular 3-4Y was one of them.
What we changed
Deleted the variants that never existed. Not flagged. Deleted.
Freshness gate. If source data is older than 24 hours, we refuse to write to the store at all. Stale data never reaches a customer.
Banned fallback costs. We used to substitute a default unit cost when the real one was missing. With it in place, 227 SKUs looked below our floor. With it banned, 98 actually are. 129 of those alerts were never real.
The part I’m not proud of
We deleted the variant before checking whether any open order referenced it.
The customer had already paid. We found out because the warehouse said “no such size.”
And the ad math
The same week, the ad report said 2 paid orders, $27.33, ROAS 0.45.
One of those orders was the cancelled one. Over 4 days:
Spent: $61.00
Clicks: 128
Paid orders: 1
Paid revenue: $11.09
Free-display order: $35.18
ROAS: 0.18
CPA: $61.00
A cancelled order was still counted as revenue. I changed the ledger: cancelled is not revenue.
The number got worse and more true at the same time.
One refund, three systems re-checked. Worth it.