Personalisation tailors content, products or messages to each user. Done well it saves users time; done badly it feels intrusive or narrow.
Types of Personalisation
- Recommendations: products, articles, courses.
- Ranking: ordering search results or feeds.
- Messaging: choosing which offer or email to send.
- Interfaces: adapting layout or defaults to behaviour.
Data
Explicit signals (ratings, preferences, likes) and implicit ones (views, clicks, purchases, time spent). Implicit data is plentiful but noisy: a click isn't always interest.
The Cold-Start Problem
New users and new items have little history. Use popular items, content similarity, onboarding questions and context such as location or referral source until behaviour data accumulates.
Evaluate With Experiments
Offline metrics guide development, but A/B tests measuring real outcomes — engagement, conversion, retention, satisfaction — are the true test.
Avoid Common Traps
- Filter bubbles: add diversity and exploration.
- Feedback loops: recommending only what was previously recommended.
- Popularity bias: a few items dominating.
- Creepiness: using sensitive inferences users didn't expect.
Respect Users
Explain why items are recommended, let users adjust or reset preferences, comply with privacy law, and avoid personalising on sensitive attributes without clear consent.