Also called: recommendation engine · Last checked: 7 November 2026
In short: A recommendation system is software that suggests products, films, songs or posts each person is likely to want, based on what they and similar people have viewed, bought or rated before.
Table of Contents
How it works
- Collect signals. The system records what people click, watch, buy, rate or skip, along with details about each item.
- Find patterns. A machine learning model learns which items tend to appeal to which kinds of people.
- Rank and show. For each visitor it scores thousands of items and shows the top few, then learns from what they pick.
Where you’ll meet it
Netflix rows, Amazon’s “customers also bought” panels, Spotify playlists and the YouTube and TikTok feeds all use recommendation systems. Shopify and many email tools offer product recommendations as a built-in feature.
What it means for your business
- More sales per visitor. Suggesting related items on product pages and in emails raises basket size without extra traffic.
- Buy before you build. Most online shops can switch on recommendations in their ecommerce or email platform. Custom models make sense only with large catalogues and plenty of data.
- Test it. Compare sales with and without recommendations for a few weeks before paying for an upgrade.
Watch out for
- Cold start. New products and customers have no history, so early suggestions are weak.
- Narrow loops. Showing people only what they already like can hide the rest of your range.
- Privacy rules. Using personal data for recommendations needs a lawful basis and clear notice under UK and EU law.
Example
Netflix. In March 2025 Netflix described on its tech blog a single foundation model for recommendations, trained on members’ viewing histories in the way large language models are trained on text. It wanted one model to centralise what many separate models, such as those behind “Continue Watching”, learn about members. It reported steady gains as the model and its data grew.
Recommendation system vs predictive AI
| Recommendation system | Predictive AI | |
|---|---|---|
| What it does | Suggests items a person may want | Forecasts an outcome, such as demand or churn |
| Output | A ranked list for each person | A number, score or category |
| How they relate | One kind of predictive AI | The wider family |
Frequently asked questions
Can a small online shop use a recommendation system?
Yes. Most ecommerce and email platforms include product recommendations as a setting or low-cost add-on, and they work best once you have a few hundred orders of history.
Why do recommendations sometimes feel wrong?
The system only sees past behaviour. A one-off gift purchase or a shared account can skew it, and new customers have too little history for good suggestions.
Does a recommendation system need personal data?
Usually, yes. It works from purchase and browsing history. Some systems use only what is in the current basket or session, which lowers privacy risk but gives weaker suggestions.