A PIM data model groups products into families that share an attribute set, defines each attribute with a type, scope and validation rule, and organises variants as a parent with child products that differ along defined axes such as size or colour. A good model mirrors how customers choose products and how channels require data.
A product data model defines how a PIM (product information management) system stores product information. It rests on three building blocks: families, attributes and variants. Families group products that share an attribute set. Attributes define each piece of data with a type, a scope and validation rules. Variants organise sellable items under a parent along axes such as colour or size. When the model follows how customers choose products, enrichment, quality scoring and channel exports all become simpler.
What are families, attributes and variants?
A family is a set of products described by the same attributes, such as "cordless drills" or "dining chairs". The family decides which attributes appear and which are required for each channel.
An attribute is one property of a product, for example voltage, material or marketing description. Each attribute has a type (text, number with unit, select list, boolean, date, asset, reference), and it may be localisable, channel-specific or both.
A variant is a sellable item that shares most content with siblings and differs along one or more variant axes. In Akeneo's documented model, a "family variant" distributes a family's attributes across levels, a "product model" holds the shared values, and variant products hold the values that differ. Other platforms use different terms, but the logic is the same.
What does a worked data model look like?
The table below shows an illustrative model for three families. The attribute names and splits are examples only.
| Family | Shared attributes (parent) | Variant axes | Variant-level attributes |
|---|---|---|---|
| T-shirts | Brand, fit, fabric composition, care instructions, description | Level 1: colour; level 2: size | Colour images (level 1); GTIN, SKU, stock unit (level 2) |
| Cordless drills | Brand, chuck size (mm), max torque (Nm), battery platform | Kit contents | GTIN, SKU, number of batteries, case included |
| LED panels | Dimensions (mm), IP rating, mounting type | Colour temperature (K) | Luminous flux (lm), GTIN, ETIM class EC number |
The rule of thumb: put an attribute at the highest level where it is always identical. Images often vary by colour and almost never by size, so they belong at colour level. Identifiers such as the GTIN (Global Trade Item Number) always sit at the sellable level.
How should attribute types and rules be designed?
Strong typing is the cheapest data quality control available. A free-text "voltage" field collects "18V", "18 V" and "18 volt". A numeric attribute with a unit stores one comparable value.
Good practice for attribute design:
- Use select lists for any value set that can be enumerated, such as material, colour family or certification.
- Store numbers with units and convert on export, never in the source value.
- Separate marketing text from technical data, so translation and approval workflows can differ.
- Decide scope explicitly: global, per locale, per channel, or per locale and channel.
- Add validation (regular expressions, minimum and maximum values, allowed file types) at attribute level.
- Name attributes consistently, with a code that never changes and a label that can be translated.
B2B standards follow the same principle. In the ETIM model, every feature is alphanumeric (a fixed value list), logical, numeric or range, and numeric features carry a unit. ETIM 11.0 is planned for final release on 1 December 2026, so teams aligning with ETIM should plan a mapping update. Our guide to ETIM, BMEcat and ECLASS explains the standards.
How do families relate to the product taxonomy?
Families describe data structure, while categories describe navigation and classification. A single family such as "chairs" may appear in several web categories ("office", "dining", "outdoor"). Keeping the two separate lets merchandisers reshape navigation without restructuring data. The companion article on product taxonomy design covers category trees and standard classifications.
Which data modelling mistakes are hardest to undo?
Some decisions become expensive once products and integrations depend on them:
- One giant family with hundreds of optional attributes, which breaks completeness scoring.
- Variant axes chosen for convenience, such as using "pack size" as an axis when packs have different GTINs and content.
- Unit and language baked into values, which blocks conversion and translation.
- Attribute codes that mirror one system, such as ERP field names, which tie the model to that system's limits.
- No reference entities for brands, suppliers or materials, which leads to duplicated text.
Review the model with category owners and channel managers before migration starts, as described in the PIM implementation roadmap. The model also sets the baseline for product data quality metrics.
Key takeaways
- A PIM data model rests on families, typed attributes and variants organised along explicit axes.
- Place each attribute at the highest level where its value is always identical, and keep identifiers at the sellable level.
- Strong typing, units and select lists prevent most quality issues before they occur.
- Keep families (data structure) separate from categories (navigation and classification).
- Align with ETIM or ECLASS early where B2B partners require them, and plan for the ETIM 11.0 release.
Frequently asked questions
What is a product family in a PIM?
A product family is a group of products that share the same set of attributes, such as all drills or all T-shirts. The family defines which attributes appear, which are required for each channel and how completeness is calculated. Families are a data-structure concept and usually differ from the navigation categories shown to shoppers on a website.
How many variant levels should a product model have?
Most catalogues need one or two variant levels. A T-shirt might use colour at the first level and size at the second, so that images and descriptions sit at colour level and stock-keeping data sits at size level. Additional levels add maintenance effort, so each level should match a real difference in content or channel requirements.
Should a PIM data model follow ETIM or ECLASS?
Where trading partners require ETIM or ECLASS, align families and attributes to those classes from the start, or map to them in a dedicated classification. ETIM features have defined types and units, which makes mapping straightforward. Companies that sell only through consumer channels can design their own model and map to standards where channels need them.
For guidance on structuring a PIM selection, see the independent resource PIMselection.com.