One catalogue, true on every channel.

The 60-count bottle goes up four dollars on a Monday because freight moved. Shopify takes the change the second you hit save. On Thursday, Amazon is still selling it at the old price, because that price arrives through a feed on a schedule somebody set eighteen months ago and nobody has opened since. Nothing broke. No alert fired. You have simply sold three days of units at the margin you raised the price to escape.

That is catalogue drift, and it is the quietest failure in ecommerce operations. A price, a stock level or a variant attribute diverges between Shopify and a channel, and there is no error state to find: the listing is live, the orders are landing, the dashboard is green. You find out when a customer buys something you cannot ship, or when the buy box goes to somebody who undercut a price you had already changed.

The catalogue is only a source of truth if something enforces it. Left alone, every channel becomes its own version of the product — and the version that is wrong is always the one making the sale.

Product data changes faster than the feeds do

A skincare line moves from a 50 ml jar to 45 ml and every net weight changes. A supplement brand rewrites the ingredient deck when the manufacturer swaps a binder. A roaster retires a single origin when the lot runs out and puts a new one in the same slot on the site. Each of those is a routine Tuesday in Shopify and a five-place propagation problem everywhere else: Amazon wants it in flat-file attribute names, Merchant Center wants unit pricing and a GTIN, TikTok Shop and Walmart each want their own category attributes. Somebody mapped all that once, by hand, in a spreadsheet, and then left.

The translation itself is the tractable half, and it is exactly the shape of work a model is good at: bounded, repetitive, and checkable against a source record. Reading a 45 ml jar, an ingredient deck and a pack quantity and producing the fields Amazon wants for that category — or picking which browse category a new SKU belongs in — is a job nobody will do by hand across a deep variant tree, and a job a person can review as a batch rather than compose one field at a time. What a model does not get to decide is the price, the floor, or which channels you should be on at all. Those are yours; the build turns them into rules so they stop being re-argued every week.

Subscription SKUs do not exist on a marketplace

On Shopify a subscription is a selling plan attached to a product — the same SKU, sold differently. No marketplace has that concept. So brands either create parallel SKUs and start maintaining two truths, or they push the subscription variant out as a one-off, where it quietly draws from the same inventory pool as next week’s autoship allocation. The bottle that was spoken for gets sold to a stranger, and the subscriber of fourteen months gets the delay email.

Variant-heavy bundles multiply every mistake

Six flavours in three sizes, plus a three-pack of each, is thirty-six sellable things before anyone has invented a gift set — each of them needing a GTIN, a net weight, a pack quantity and a category attribute per channel. One wrong net weight on a bundle is not one wrong listing: it is a unit-price error that gets the listing flagged in Merchant Center and a pack-weight error that follows the SKU into every fulfilment quote until somebody notices.

And nobody owns the reconciliation

Rarely does anything in the stack exist to ask whether the channel still says what Shopify says. Inventory syncs on a timer, prices sync on publish, and the two drift apart in between. Most tools do a sync. Almost nothing does the reconciliation that catches the sync when it silently stops. And there is a second reader of all this now. When a retrieval system finds three descriptions of one product it does not resolve the contradiction — it skips the product and names a competitor whose data agrees with itself. The catalogue stopped being only a sales feed the moment models started answering shopping questions out of it.