Product hierarchies: brand, category, subcategory, and item

Where does a can of sparkling lemon water belong? A product hierarchy answers that question for every item in a dataset. It's the set of nested groupings, usually category, subcategory and item, with brand kept as a separate attribute, that decides which products get added together when someone reports a total.

The hierarchy sounds like housekeeping, and in a sense it is. But every category number you read depends on it. If two analysts file that lemon water in different places, they'll report different sizes for two categories, and both will be arithmetically correct.

What are the levels, and what does each one mean?

Most retail hierarchies run from broad to narrow. A department is the widest grouping, such as beverages. A category sits inside it, such as bottled water. A subcategory splits the category further, into still and sparkling water, say. At the bottom is the item, which in scan data usually means a single UPC (Universal Product Code), the barcode number for one product in one size.

Brand runs across this ladder rather than inside it. A single brand can own items in several subcategories and even several categories. That's why many analysts treat brand as an attribute attached to each item, alongside size, pack count and flavor, rather than as a rung of its own.

A small example helps. Imagine three items: a 16.9-ounce still water from Brand A, a 12-ounce sparkling water from Brand A, and a 12-ounce sparkling water from Brand B. Group by subcategory, and the two sparkling waters sit together. Group by brand, and the two Brand A items sit together. Both views come from the same three rows.

Why do hierarchies differ between companies?

Because each company builds its hierarchy for its own decisions. A manufacturer may group items by how its sales team is organized. A retailer may group them by where they sit in the store. A data provider needs something neutral enough to compare across many retailers and many brands.

None of these is wrong. They answer different questions. The trouble starts when numbers built on one hierarchy get compared with numbers built on another and nobody notices. A juice total that includes juice drinks and one that excludes them can differ by a wide margin, and the difference is purely definitional.

Teams that practice category management, running each category as its own business unit, usually keep a mapping between their internal groupings and whatever an outside dataset uses. That mapping is worth more than most people think.

Where do the awkward items go?

Every hierarchy has edge cases. A sparkling juice could be juice or soft drink. A granola bar could sit in snacks or in breakfast. A variety pack holding three flavors can't sit under any one flavor.

A few rules keep these from turning into arguments:

1. Classify by what the product is, and ignore who makes it.
2. Write down the rule for each edge case the first time it comes up, so the next analyst doesn't decide differently.
3. When an item truly straddles two groups, pick one and note it rather than splitting its sales.
4. Review the placement of new items when they first appear in the data.

The fourth point matters more than it looks. New UPCs arrive constantly, and an unclassified item quietly falls out of every total until someone catches it.

How should you check a hierarchy before trusting a total?

Ask for the list of subcategories inside the category you're reading, and scan the largest items in each. If something looks misplaced, it probably is. Then check whether the hierarchy changed during the period you're comparing. A reclassification halfway through a year can look exactly like growth or decline.

NRS Insights groups scan data into categories so trends can be compared month to month, and each release is collected on the news page.