Lookalikes, attribution & programmatic
How Audience Data Is Used in Advertising
What platforms actually infer about people, where the signals come from, and why the phrase personalised is doing a great deal of work in most ad copy.
Evergreen No expiry
Modern advertising works by predicting what an unidentified person is likely to do next. This is worth understanding plainly, because the industry's own language tends to blur the line between what is known and what is inferred.
Signals, not identities
Most advertising measurement and targeting no longer relies on knowing who somebody is. It relies on observations: what was watched, what was searched, what was bought, what was skipped. A single observation identifies nobody. Enough of them, attached to a device or an account, produce a profile that is accurate in aggregate and often wrong in detail.
This is why the industry moved to aggregated and modelled approaches: individual identifiers are being restricted in several jurisdictions, and the behavioural signal turned out to be more useful than the identifier anyway.
What a platform can usually infer
Broad category interest, likely intent, recency of engagement, and some approximation of price sensitivity. What it usually cannot infer is anything specific about a person's identity, and the difference matters when evaluating an advert that claims to reach a particular kind of person.
Why the language is slippery
Personalised usually means content assembled from inferred interest. Relevant usually means selected by predicted behaviour. Targeted usually means a segment the advertiser selected, which the platform then matched to individuals. None of these means the advertiser knows who is receiving it.
The privacy consequences are real
The signals that make this effective are the same ones that make it sensitive. Behavioural data reveals interests a person did not choose to disclose and would not expect to be inferred. Regulation in several jurisdictions now restricts certain uses, and the direction of travel is toward more restriction rather than less.
The practical conclusion
Treat inferred targeting as useful and imprecise. It is genuinely good at finding people likely to respond and genuinely bad at identifying which specific person. Campaigns that depend on precision are fragile in a way that campaigns depending on scale are not.