Two dimensions of an assortment

An assortment is described by two independent quantities.

Width is how many different product groups the store carries. It answers the question of how
many different shopper needs the retailer covers.

Depth is how many variants exist inside a single group: models, brands, sizes, volumes,
colourways. It answers how precisely the store covers one specific need.

Width = number of product groups (categories) in the matrix
Depth = Total SKUs / Number of product groups   (average depth per group)

The average depth is useful only as a first bearing: it hides the skew. Real catalogs usually
contain a few categories absorbing most of the SKUs and a long list of categories holding a dozen
items each. So the distribution is read alongside the average — how many categories hold fewer than
10 items, and what share of SKUs sits in the five largest groups.

Working with both dimensions is the job of category management,
but the consequences land on the site: it is the interface that decides whether a shopper can
actually use the choice on offer.

Standard strategies

Strategy Width Depth Example What it means for the site
Niche store Narrow Medium–high A single-sport store, a mono-brand Simple navigation, the weight sits on the product page and expert content
Category killer Narrow Maximum A large chain in one category (electronics, DIY) Facets and comparison are critical, search carries most of the traffic
General retailer Wide Medium A hypermarket, a multi-category store The homepage has to route rather than sell
Marketplace Maximum Maximum A platform with third-party sellers Duplicates and near-duplicates, relevance is decided by algorithms
Discounter Medium Minimum A hard discounter with a limited matrix Choice is narrowed deliberately, the interface is simple, filters matter little

No strategy is inherently better than another — they distribute the load differently. A narrow, deep
assortment shifts the load onto the instruments that narrow choice; a wide, shallow one onto routing
the shopper between categories; and both dimensions at maximum demand each at once.

The paradox of choice

Growing depth brings an obvious benefit: a higher probability that the variant a shopper needs
exists in the catalog at all. And an equally non-obvious cost: every new variant increases the work
a shopper has to do in order to choose.

The mechanism is simple. People look at the first 10–20 items of a listing. If the assortment
doubles in depth and the product order does not change, the shopper sees the same head — only now
with twice as many unseen items behind it. Nominally choice has grown; in fact it has not.

The instruments that turn depth into genuinely accessible choice:

  • Faceted search. Filters by brand, size, price and attributes are
    the basic way to narrow hundreds of items down to a dozen. The requirement for deep catalogs is
    complete and reliable attributes, otherwise the filters cut out relevant products.
  • Meaningful sorting. Sort by popularity on a deep catalog preserves the head: the popular is
    shown because it has been shown.
  • Listing personalization. Reordering products around one
    shopper’s preferences pays off in proportion to the depth of the category: in a category of 20
    items the order barely matters, in a category of 2,000 it decides everything. In fashion cases
    such personalization delivers +12.3% AOV, in grocery +7.5% CR.
  • A clear category structure. If the depth has accumulated inside one flat category, no sorting
    will rescue it: the category has to be split.
💡

Before extending depth, check how many items in the current category receive at least one impression
in listings. If a substantial part of the catalog is already invisible to shoppers, adding new SKUs
increases the warehouse, not the choice.

What depth means for algorithms

A deep assortment changes what recommendation mechanics have to do.

First: the long tail grows — the share of products with few interactions.
Those items carry almost no behavioural statistics, so algorithms trained on clicks and purchases
systematically underrate them. It is a self-sustaining loop: the product is not shown because there
is no data, and there is no data because it is not shown.

Second: catalog coverage falls — the share of products that make it into at
least one recommendation. A metric of little consequence for a 500-item store becomes the key one at
tens of thousands.

Third: the cost of uniform blocks rises. When a category holds 40 near-identical items, a block of
five options from one brand in one price band does not help anyone choose. This is where
recommendation diversity works — deliberately widening the block by brand,
price and attributes.

Category depth What becomes critical
Up to ~50 items A clear structure, basic sorting
50–500 items Faceted filters, attribute quality
500–5,000 items Listing personalization, recommendation coverage
Over 5,000 items Long-tail work, block diversity, duplicate control

The bands are indicative — the specific thresholds depend on the category and on how distinguishable
the products inside it are to a shopper. The general pattern is stable: with every order of
magnitude, responsibility for the choice shifts from the shopper to the algorithms.

How to assess an assortment from the site’s side

  1. Count width and average depth, then the distribution of SKUs across categories — an average
    without a distribution is useless.
  2. Measure the share of items that received at least one listing impression in a month, and the
    share with sales over a quarter.
  3. Check attribute completeness in the deep categories: incomplete attributes switch the filters off
    for part of the range.
  4. Assess recommendation coverage of the catalog — how many products make it into blocks at all.
  5. Compare conversion across categories of different depth: a slump in the deepest ones almost
    always points at navigation rather than demand.
  6. Before extending the matrix, answer which instrument the shopper will use to narrow the choice —
    and deploy it before the new batch of SKUs arrives.
  7. After the extension, re-measure the same metrics: a rising item count with a flat share of
    displayed products means only the warehouse has grown.