🤖 AI / ML

Catalog Coverage

Coverage · Item Coverage

Definition

Catalog coverage is a recommender metric measuring the share of items that appeared in at least one recommendation over a period, relative to the full catalogue.

What catalogue coverage is

Catalogue coverage is the share of items that received at least one recommendation over a period:

Coverage = unique recommended SKUs / total SKUs in catalogue × 100%

If a catalogue holds 50,000 SKUs and the recommender showed only 8,000 of them in a month, coverage
is 16%. The other 84% of the assortment is effectively invisible through the recommendation channel.

Why coverage matters commercially

Low coverage means the recommender concentrates traffic on products that are already popular.
That creates a feedback loop: popular items get more clicks, accumulate more signal and become more
popular still. The long tail of the catalogue never monetises.

In retail the problem is sharpest in:

Niche categories with specialised inventory
Seasonal new arrivals, on sale but with no purchase data yet
Secondary brands with less marketing support

Coverage versus precision

Priority	What you get	What you lose
Maximum precision	Highly relevant recommendations	Low coverage, a dead long tail
Maximum coverage	The whole assortment gets exposure	Some recommendations are irrelevant
Balance (hybrid)	Acceptable precision plus long-tail reach	Algorithmic complexity

The optimum depends on catalogue structure and margin. If long-tail margin exceeds bestseller
margin, raising coverage moves profit directly.

Tip: track coverage by category rather than only in aggregate. A new category can sit at zero
coverage while the headline number looks healthy.

Frequently asked questions

What counts as good coverage?

It depends on catalogue size and the business goal. On a 1,000-SKU catalogue, 80% is normal. On a million SKUs, 20–30% is already meaningful. What matters more is the trend and the comparison with business metrics: if the top 100 items produce 80% of revenue, aggressively raising coverage may not pay back.

Why does collaborative filtering have low coverage?

Because it builds recommendations from interactions. Items with no purchase or view history fall into a cold zone and never enter the output. That is the cold start problem. Content-based filtering and hybrid methods give better coverage.

How do you raise catalogue coverage?

Hybrid algorithms combining content and collaborative signals, an explicit boost rule for new items, diversity mechanics to reduce concentration on hits, and periodic monitoring of never-recommended SKUs with manual inclusion through merchandising blocks.

Gravity Field