What is personalization?

What genuinely works

Personalisation earns its cost when it reduces effort. Showing someone the thing they were looking at, in their language, at their local price, with their saved preferences applied — all of that removes work the user would otherwise do.

It stops working when it becomes a demonstration of what you know. Inserting a first name into a subject line, or referencing behaviour the user did not realise was recorded, produces the opposite of the intended effect: it signals surveillance rather than service.

The line

• Reducing effort reads as helpful — resuming where they left off, defaulting to their usual choice.
• Demonstrating knowledge reads as invasive — referencing behaviour from another context or another site.
• Data they gave you knowingly is safe to use visibly.
• Data inferred without their awareness is not, even where it is legal.
• The useful test: would the user be comfortable if you explained how you knew? If not, use it to improve the experience without naming it.

Personalization without identifiers

Cross-site behavioural personalisation has largely gone with the third-party identifier. What remains is first-party — your own logged-in data — and contextual, based on what the user is doing right now rather than what they did elsewhere.

On conversational surfaces the contextual version is unusually strong, because the user has explicitly stated their situation. Tailoring to what someone just said they need is more precise than any inference from browsing history, requires no stored profile, and does not feel invasive because the user supplied the information deliberately — see synthetic audience.