Objectives¶

Extraction has no single objective. The same profile serves many buyers for different ends, and the end decides what a good profile even is: for one buyer it is a fresh location, for another a firm guess about income, for a third simply a name that matches a name it already holds. The objective everyone names is advertising. It is also the most benign of them, and it works as the alibi for the rest.

Targeting and attribution¶

The visible purpose: fit a message to a segment, then prove the message caused the sale. The proving is the larger appetite. A great deal of collection exists not to choose which advert to show but to measure whether an advert already shown did anything, which is why tracking follows a person well past the click and into the purchase. Targeting is the objective that gets talked about because it is the easiest to defend. It rarely costs the person anything on its own, and it justifies the infrastructure that the costlier objectives then use.

Sorting people by price¶

The same profile that personalises an offer can price it. Once a seller can estimate what a particular person will pay, or read their postcode, device, or past behaviour as a proxy for it, one price becomes many, and which one a person is shown stops being about the goods and starts being about them. Steering does the same job more quietly, surfacing the cheaper or the dearer options to different people and calling it relevance. The line between personalising and discriminating is a commercial decision, not a technical one, and this is the objective where being known first starts to cost money.

Scoring risk¶

Further along, the profile stops suggesting and starts deciding. Creditworthiness read off spending patterns, insurance priced from lifestyle and behavioural signals, tenants and job applicants screened against bought data: here the inferred attribute becomes a gate. A guess about a person, right or wrong, firm or shaky, turns into a loan refused, a premium raised, an application quietly dropped. The GDPR’s Article 22 restricts decisions taken solely by automation where they carry legal or similarly significant effects, which is a fair signal of how consequential this objective is understood to be. The boundary between commercial scoring and state scoring is thinner than it looks.

Building lookalikes¶

A profile of known customers has a second use: finding strangers who resemble them. Lookalike modelling takes the pattern of the people a business already has and reaches for others who share it, so a person’s data acts on people they have never met and will never know contributed to the decision. This is the objective where individual consent reaches its limit. Opting out protects the person who opts out; it does nothing for those the model locates through the pattern they left behind. The surveillance model meets the same problem at population scale, where a record is a building block in a model aimed at a group.

Onward sale¶

For much of the chain the data is not a means to sell something else. It is the thing being sold. Collection funded by licensing access to the profile, renting it to bidders, or matching it against a buyer’s own lists is a revenue line in its own right, independent of any product the collector advertises. This is what makes the reassurance that a company “does not sell your data” so slippery. Licensing is not selling, sharing is not selling, a matched-audience deal is not selling, and the data moves all the same.

Training the models¶

The newest objective folds the others’ raw material into something else again. Behavioural records, content, and interactions increasingly feed recommendation engines and, lately, the training of AI systems, where a person’s contribution cannot be pulled back out once it has been learned. Data gathered under one stated purpose becomes fuel for models whose outputs have nothing to do with it, and consent given to the first use rarely anticipated the second.

The friendly objective in front¶

Advertising sits at the front of the shop because it is the objective a person minds least. Behind it, the profile earns most where it sorts people into prices, reads them as risks, and decides on their behalf. Naming only the first objective is how the others keep working undisturbed.

Last reviewed: 2026-07-17.