Definition
SKU (stock keeping unit)
An internal code that uniquely identifies a sellable product variation for inventory, catalog, fulfillment, and reporting.
A SKU, or stock keeping unit, is the internal identifier your store uses for one sellable variation: size, color, pack size, region edition, or bundle. It is not the marketing product title and it is not always the same as a UPC, EAN, or GTIN printed for retail scanners. Inventory, purchase orders, 3PL bins, product feeds, and support tools all hang off that code. When SKU design is sloppy.
Reused codes, free-text variants, missing barcodes. Every automation downstream invents stock or oversells. Clean SKUs are boring infrastructure. They are also the difference between agents that answer correctly and catalogs that lie.
What a SKU is (and is not)
A SKU identifies a stockable unit inside your business systems: Shopify variant, warehouse location, forecast row, and invoice line. Two black hoodies in size M and L are two SKUs even if they share a product page. A three-pack of the same bar is a different SKU from a single if you sell and count them differently.
The title customers see can change; the SKU should stay stable once inventory history attaches to it. SKU is not a universal product code. UPC, EAN, and GTIN are external identifiers for retail and shopping feeds; many DTC brands map one GTIN to one SKU, but the SKU remains yours. SKU is also not a style number alone. Style plus color plus size is the usual retail pattern.
Shopify's product model stores a variant-level SKU field for a reason: inventory and fulfillment run at the variant, not the product parent. If you only name products and leave variant SKUs blank, every integration will invent its own identity.
SKU vs UPC, barcode, and variant ID
Operators mix four identifiers. The platform variant ID is an opaque system key (Shopify GID or numeric ID). The SKU is the human- and ops-facing stock code you choose. The barcode (UPC/EAN) is what scanners and many marketplaces expect. The handle or product ID is for the parent listing. Confusing them breaks receiving: a warehouse that scans barcodes cannot put away stock if your SKU sheet only has internal nicknames.
Use SKUs that are unique across the entire catalog, including retired goods. Never reuse a SKU for a different physical item; archive and create a new code when the product changes enough that historical units-on-hand would lie. Put the barcode in the barcode field and the SKU in the SKU field.
Do not paste the same string everywhere "for convenience." Shopping feeds and product feed pipelines need both attributes clean so Google Merchant Center and Meta do not disapprove items for identifier mismatches.
Naming conventions that survive growth
A good SKU scheme is readable by humans and parseable by scripts: brand or category prefix, style, color code, size, and pack. Joined with a fixed delimiter and fixed width codes. Example pattern: `HDY-BLK-M` for hoodie black medium, or longer if you need season and region. Avoid spaces, lowercase chaos, and marketing adjectives that change every campaign.
Document the code table (BLK = black) so new merchandisers do not invent `BLACK` and `BLK` as two SKUs for the same fabric. Plan for variant explosion. Apparel and configurable goods can multiply SKUs faster than demand; every dead size still occupies a forecast row and a feed entry. Set rules for when a custom or made-to-order item gets a real SKU versus a non-inventory line.
Bundles need their own SKU plus a bill of materials that depletes component SKUs, or you will sell kits while components still show available. Revisit the convention when you add a second warehouse or 3PL. Location is not part of the SKU; location is inventory position for the same SKU.
Inventory, 3PLs, and oversell risk
Available-to-sell is a SKU-level number. If two listings share one physical pile but different SKUs, you double-sell. If one SKU maps to two different bins because someone duplicated the product, counts drift. Cycle counts, receiving, and pick confirmation all assume SKU uniqueness. A 3PL multiplies the cost of bad identity: they receive against your ASN SKUs, pick against order line SKUs, and push adjustments back to Shopify by those same codes.
Stockouts and oversells are often labeled as "forecasting problems" when the root is duplicate or missing SKUs after a migration. Before peak, audit: blank SKUs, duplicate SKUs across variants, SKUs that differ only by case, and products sold on multiple channels with different codes for the same unit. Align channel listings to a single internal SKU master.
Safety stock and reorder points only work if the unit you protect is the unit you sell. Inventory turnover metrics collapse when the denominator is garbage SKU master data.
AI agents and automation need SKU truth
Support and ops AI only help when tools return the right variant. An agent that checks inventory must call a tool with a SKU or variant ID, not a fuzzy product title. "Is the blue one in stock?" is ambiguous; `get_inventory(sku="HDY-BLU-M")` is not. The same applies to returns restock, exchange offers, and recommendation engines that attach "you might like" SKUs.
Hallucinated titles plus messy catalogs produce confident wrong stock answers and angry WISMO-adjacent tickets when the wrong item ships. Tool calling does not fix identity. If your Admin API returns three variants with empty SKUs and similar titles, the model will guess. Scope read tools to inventory and product fields you trust, and block write tools that restock or swap line items until SKU hygiene passes an audit. Catalog automation.
Bulk title rewrites, AI descriptions, feed enrichment. Should never renumber live SKUs without a migration plan. Operators on Shopify forums who try AI to "speed up catalog work" usually discover that automation amplifies whatever naming mess already exists.
Operational checklist for a clean SKU master
Run a quarterly SKU audit before any major catalog or AI project. Export all variants: flag blank SKUs, duplicates, and SKUs that differ only by capitalization or trailing spaces. Require barcode coverage on any unit that touches a warehouse scanner. Freeze a naming convention document in the same place as brand guidelines.
When launching a product, create the SKU before ads go live so feeds, 3PL, and the storefront share one code from day one. On migrations (new 3PL, new ERP, store rebuild), map old SKU to new SKU explicitly and keep a redirect table for open POs and historical orders. Never let marketplace channels invent permanent identifiers that never sync home. For multi-location inventory, keep one SKU identity and multiple location quantities.
Train support macros and AI tools to ask for order line + SKU when processing exchanges. The goal is boring consistency: every system that moves money or stock agrees on what one unit is.
Common questions
Frequently asked questions
What does SKU stand for in ecommerce?
SKU stands for stock keeping unit. It is your internal code for one sellable variation so inventory, fulfillment, and reporting can track that unit uniquely.
Is a SKU the same as a UPC or barcode?
No. A SKU is your internal identifier. UPC, EAN, or GTIN are external product codes used for retail scanning and many ad/shopping feeds. Map them carefully; do not treat the fields as interchangeable.
Should every product variant have its own SKU?
Yes if you stock, sell, or forecast it separately. Size and color variants that deplete different inventory need different SKUs. Shared inventory across listings should share one SKU, not duplicates.
Can I reuse an old SKU for a new product?
You should not. Historical inventory, orders, and analytics attach to the code. Reusing it for a different item corrupts counts and reports. Retire the old SKU and create a new one.
Why do AI support tools care about SKUs?
Agents look up live inventory and order lines by SKU or variant ID. Fuzzy titles and blank SKUs make tools return the wrong stock or force the model to guess, which creates oversells and bad answers.
Related terms
