A messy catalog isn't a content problem you fix once — it's a state problem you govern continuously. This guide gets a 1,000+ SKU store from a raw export to a catalog the Catalog Specialist can run, in ten working days.
Days 1–2: get the ground truth
- 1Export the full catalog and join it to landed cost. SKUs with no cost are your Tier C problem — flag them now.
- 2Tag obvious dead inventory: zero sales in 180 days, no stock, discontinued lines still listed.
- 3Identify your top 20% of SKUs by contribution margin. These get the most conservative gates.
Days 3–6: classify into lifecycle states
Assign every SKU a lifecycle state. Don't agonize — the agent will surface corrections as review rows once it's running. The point is to start with a defensible baseline.
- AUDIT — awaiting initial classification (where every existing SKU starts).
- PUBLISHED / ACTIVE — live: accumulating data, or classified with enough of it.
- OPTIMIZING / ON_DISCOUNT — a rewrite cycle or last-chance discount test in progress.
- DRAFTED / SEASONAL_VAULT — pulled from Shopify, kept in the database (or held for the season).
- ARCHIVED — permanently dead, retained for history.
Days 7–8: stamp cost-confidence tiers
Tier A means verified cost — Shopify cost_per_item or a manual entry; Tier B means estimated, from an AliExpress reverse-image match; Tier C means no cost data at all. The tier gates the one action where margin is at stake: discount_test only unlocks at Tier A.
{ "sku": "4471", "state": "ACTIVE", "cost_tier": "A", "margin": 0.52 }Days 9–10: dry-run and read the rows
Connect read-only and let the Catalog Specialist propose. The first dry-run will tell you where your classification was wrong faster than any spreadsheet review. Tune ALLOWED_TRANSITIONS where it complains, and you're ready to flip.