Catalog raises $3M to build the product data layer for AI commerce. Read the announcement.
All posts
Product Data

Data mapping for ecommerce: Product fields need clear rules

Map product fields from suppliers and business systems into storefronts, marketplaces, and feeds. Set rules for variants, identifiers, conversions, and tests.

Your catalog can show the right product on your storefront and still send the wrong variant, price, or availability to a shopping channel. The failures often begin where one system's product fields are translated into another's. You need clear rules for what each field means, which source owns it, and how every destination should receive it.

What data mapping means for ecommerce product data

Data mapping defines how a source field and its values become a destination field and its values. A useful rule records the field's meaning, scope, owner, transformation, and failure behavior. Mapping can rename a field, convert its type or units, standardize a value, combine sources, or apply only to a particular market or product category.

That is one job within product data integration. Integration moves records between systems. Mapping defines how a destination should interpret each record. Normalization gives equivalent values a controlled representation, while validation checks whether the mapped output is fit for use. For the wider connection and sync workflow, see our product data integration guide.

One source record can feed a storefront, marketplace, shopping feed, or AI shopping surface. Each destination can use its own field names, accepted values, item relationships, and presentation rules. A field label alone does not tell you whether the values mean the same thing.

Catalog provides a product data layer for AI shopping: we structure product data, publish it to AI shopping surfaces, keep it synced, and measure outcomes. The mapping contract still determines which source owns each fact and what each field means.

Record each mapping rule as a contract

A bare pair of field names does not explain what the destination will receive. Each mapping rule should tell both the data producer and consumer which source wins, what the output becomes, and how a bad record is handled.

Contract fieldExample rule
Source and ownerThe enterprise resource planning (ERP) system's price_minor field is owned by the pricing team.
ScopeColor and size belong to the sellable variant; price and inventory belong to its offer.
Canonical meaning and typeprice.amount is the amount in major currency units; currency is stored separately.
DestinationUse the documented price field for the selected channel and schema version.
TransformationDivide an integer price in cents by 100, then format the amount with its currency.
Missing or conflicting valueA missing amount blocks the offer. Resolve conflicts through a documented source priority.
ValidationRequire a positive amount, a recognized currency for the market, and a value that matches the product page.
Owner and versionRecord the rule owner, approval, source timestamp, and mapping version.

Use a shared product record to stabilize meaning across systems, with category-specific fields for details such as ingredients, fit, or compatibility. Each destination can still use its own output layout. Preserve the source value and provenance alongside a derived value so your team can trace and correct a bad transformation.

Build a product-data map in six steps

1. Define the destination first

Name the channel, market, locale, schema, and business use. Mark each destination field as required, conditional, or optional. Keep separate mappings for fields whose meaning or formatting differs by channel.

Shopify's channel mapping, for example, can use product fields, variant fields, or metafields as sources. A mapping applies to the selected channel, and a blank chosen source can fall back to that channel's default source. Its tool requires a compatible source type and does not accept money fields such as price as channel-mapping sources. Document this behavior deliberately so a fallback never hides a missing value that should block the listing. See Shopify's channel-mapping documentation.

2. Profile source fields and assign owners

Inventory the systems that provide product information, including supplier files and your product information management (PIM) system, then inspect real values in representative records. Record the source path, data type, allowed values, unit, locale, update cadence, entity level, and owner for each important attribute.

A field named size might mean garment size, product dimensions, or shipping-package size. Map its meaning and scope. Assign an authoritative source for each field and market. If the ERP owns the base price but an approved promotion overrides it, write down when that override applies. Keep the original product and variant identifiers so a new sync updates the right item instead of creating a duplicate.

3. Separate products, variants, and offers

Decide whether a field belongs to the product family, a sellable variant, or an offer. A jacket's material may be shared across its variants, while its size, SKU, price, stock, image, and selected-variant URL may differ. Products sold by multiple merchants may also need separate offer records.

Keep stable identifiers for the product family and each sellable variant. A stock-keeping unit (SKU) can serve as a variant ID only if it is unique, stable, and never reused; otherwise, assign a separate item ID and preserve the SKU as another field. Do not make the parent identifier stand in for every variant. When a destination groups variants, retain the parent relationship and test both keys.

4. Define transformations, precedence, and empty-value behavior

Write conversions and lookups as explicit rules. For example:

  • Map supplier color labels to a controlled color value, while retaining the original label for traceability.
  • Convert dimensions from a known unit to the agreed canonical unit. Keep product dimensions separate from package dimensions.
  • Combine an integer price in cents with its currency to produce a correctly formatted price.
  • Map your internal category to each destination's taxonomy without replacing the internal category.
  • Build a channel-specific title from approved product and variant fields while preserving the storefront title.
  • Derive availability from sellable quantity and preorder or backorder state instead of passing a warehouse code through as customer-facing status.

State what an empty value means. The destination may need an omitted field, a fallback, a held prior value, or a rejected record. Empty is not automatically zero, false, or in stock. For example, ChatGPT's product-feed conventions treat an empty optional value as no value and disallow generic placeholders such as n/a. Your mapping should follow each destination's documented behavior.

Define conflict and update precedence per field. A simple "last update wins" rule can let a delayed inventory file overwrite newer stock or let a blank feed remove a valid price. Make precedence explicit for each source and use case.

5. Test the output with representative records

Test a small sample before activating a full catalog. Include products from different categories and locales, plus records that exercise the edge cases in your rules:

  • a product with several sizes or colors;
  • a missing or duplicate identifier;
  • zero stock, preorder, or backorder;
  • a missing image or a variant-specific image;
  • an unknown category or value;
  • a price with a different currency or decimal format;
  • conflicting values from two sources.

Validate both structure and meaning. Structure checks confirm field names, types, requiredness, and accepted values. Meaning checks confirm the right variant, image, URL, price, currency, availability, and category reach the destination. A valid record must still agree with the live product experience. Google Merchant Center's product-data specification identifies incorrect identifiers and variant attributes, poor images, and conflicts with a product page as common listing problems.

Use destination previews to inspect values, then check the destination's own errors and approval state. Shopify's channel preview shows the values a channel receives, but does not tell you whether the channel approves or displays the listing.

6. Version the mapping and monitor for drift

Treat a mapping change like a data-contract release. Record the version, owner, date, affected destinations, and products that need to be reprocessed. Before a field or identifier change, compare old and new outputs so that a schema update does not silently split product families or overwrite approved values.

Monitor more than whether a feed job completed. Track required-field coverage, products rejected or omitted, mapping errors by field, variant-group failures, and the freshness of price, inventory, and content. Reconcile the destination against the source on a schedule to catch missed events, stale values, and drift. Assign an owner to each error class so the same exception does not become a permanent manual fix.

For a practical review framework, use our product data quality guide.

Worked example: map one color variant to two feeds

Assume a 750 mL bottle has a product-family ID of ARC-750, a variant SKU of ARC-750-CLAY, a supplier color label of Canyon Clay, a price of 3,200 cents in USD, and a quantity available to sell. For the Google output, assume a free listing. The values below are illustrative. The product record retains the source values and the canonical rules produce destination-ready outputs.

Source valueCanonical ruleDestination output
Family ID ARC-750 and SKU ARC-750-CLAYKeep both as strings. Use the family ID for grouping and the SKU as the stable variant ID.Google Merchant Center: item_group_id=ARC-750, item_group_title=Arc Insulated Bottle, id=ARC-750-CLAY. ChatGPT: group_id=ARC-750, item_id=ARC-750-CLAY, listing_has_variations=true.
Supplier color Canyon ClayMap to the approved value Clay; keep the source label for traceability.Google Merchant Center: color=Clay, variant_option=color:Clay. ChatGPT: color=Clay, variant_dict={"color":"Clay"}.
3,200 cents and USDDivide by 100 and combine with the currency code.Both feeds: price as 32.00 USD.
Sellable quantity and preorder stateApply the stock rule, then map to an accepted destination value.Both feeds: availability=in_stock.
Approved image and product URL for the variantResolve media by variant and use a URL that selects the same color when available.Google Merchant Center: image_link and link. ChatGPT: image_url and url.

The key is the separation between a unique variant and its shared product group. Google Merchant Center requires a unique id for each product. For variants, item_group_id is required in free listings and for Shopping Ads targeting Brazil, France, Germany, Japan, the United Kingdom, or the United States; Google lists it as optional for variants in other target countries. When you provide it, assign one stable, unique item_group_id to the group and repeat it on each variant. Pair it with the same item_group_title across the group and variant_option entries that name every variant-identifying property. Also include standard attributes such as color, size, material, pattern, age_group, or gender when they distinguish variants. ChatGPT's product-feed schema uses a unique item_id, shared group_id, listing_has_variations=true, and variant_dict; the group ID must differ from each item ID. See Google's item-group guidance and the ChatGPT product-feed schema. The source field can remain stable while each adapter supplies the destination's own field names and variant structure.

If AI shopping is a sales destination for your catalog, explore our product data layer.