Choose a stable unit of information
A row can represent a product concept, a sellable variation, or a package containing several items. Those meanings should be separated before a spreadsheet becomes a system. Otherwise, an export may treat a family description as a purchasable item or repeat shared information inconsistently.
The model begins with examples from the intended range. We ask what makes one entry distinct, which reference remains stable, and which relationships the receiving systems need to understand.
Place attributes at the right level
Some information belongs to a whole product family, while other details vary by color, size, or package. A common description may be inherited, but a variant-specific measurement should remain attached to the correct variation. Exceptions need an explicit rule.
The attribute dictionary records a field’s meaning, expected format, and source. It should also explain whether the value is required for all items, only for a family, or only when a particular output is prepared.
Keep classification separate from presentation
Internal families help organize the information. A buyer’s category list or a website navigation menu serves a different purpose. Treating them as the same structure can make one external change unexpectedly reorganize the entire catalog.
We describe the relationship between internal classifications and destination categories. A mapping may be simple, but ambiguous cases should be visible for review instead of silently placed in an approximate category.
Review a representative sample
A model should be tested with an ordinary item, a variant family, a package, and an item with missing information. Ask the team to identify where each value belongs and how it would reach an intended output.
The delivery can include sample records and a record of unresolved decisions. The aim is a structure that people can explain and maintain. Broad migration or platform implementation should follow a reviewed model rather than precede it.