Recommendation vs personalization

Recommendation is one way to personalize: it selects or ranks catalog items predicted to be relevant. Personalization is the broader goal of changing an experience for a person or context, and recommendation is a single mechanism inside it. The two words are used as if they mean the same thing, and that blurring hides what a system actually does.

The definitions

Recommendation is the ranking or selection of catalog items, such as products, articles, videos, or songs, predicted to be relevant. Its output is a ranked set of items. Personalization is the wider practice of adapting an experience to a person or context, which covers ranking but also layout, offers, messaging, timing, and channel. Recommendation answers one narrow question, “which items should we surface,” while personalization is the whole decision about what this person sees.

Distinguishing test: mechanism or goal? Recommendation is one mechanism of personalization, and it produces a ranked set of items. It is not the whole of it, because personalization also covers the layout, the offer, the message, the timing, and the channel. Recommendation can also exist with no personalization at all: a “most popular” list is a recommendation that is identical for everyone. Treat recommendation as a component that may sit inside a personalization system, not as a synonym for it.

Why the two get conflated

The confusion is commercial as much as conceptual. A recommender is the most visible and most sellable piece of personalization, the “you might also like” rail everyone recognizes, so vendors describe a recommendation product as if it delivered personalization in full. It does not. A site can run a strong recommender and still show every visitor the same homepage, the same offers, and the same emails, which is a lot of personalization left on the table.

Recommendation without personalization

Plenty of recommendation involves no personalization. A best-sellers rail, an editor’s staff picks, and a “most popular this week” shelf all rank items, but the ranking is the same for everyone. These are useful, cheap, and honest, and they set a baseline worth measuring against: a surprising share of the value attributed to personalized recommenders can be matched by popularity alone, so the gap between a personalized ranking and a popular one is the number that matters.

Personalization without recommendation

The reverse is just as common. A free-shipping banner shown to a returning visitor with a full cart, a reordered navigation, a send-time chosen for when a person usually opens email, and a price or offer varied by segment are all personalization, and none of them is a recommendation. They change the experience without producing a ranked list of catalog items.

Where recommendation sits in the loop

In the personalization loop, recommendation is one kind of action: surfacing items. The methods behind it, such as collaborative filtering, content-based filtering, and hybrid retrieval and ranking, are covered under Techniques. Seeing recommendation as a component rather than the whole keeps a team honest about what it has actually built, and about the surfaces where no recommender runs but personalization could still help.