The familiar phrase says that if you are not paying, you are the product. It is close enough to be useful and imprecise enough to obscure how the money actually moves.
You are not sold. What is sold is access to your attention, priced according to how confidently your behaviour can be predicted. The distinction matters because it explains which data is worth collecting and which is not.
The mechanisms, separated
Free is not one business model. Several distinct ones share the label, and they differ enormously in what they require from the user.
| Model | Revenue source | Data appetite |
|---|---|---|
| Targeted advertising | Advertisers bidding for attention | High and continuous |
| Data licensing | Selling or sharing datasets | High, and the point |
| Freemium | A minority upgrading to paid | Low, usage metrics only |
| Marketplace fees | Commission on transactions | Moderate, transactional |
| Subsidised hardware | Device margin or ecosystem lock in | Varies widely |
Only the first two require extensive behavioural profiling. A freemium tool has no commercial reason to know anything beyond how the product is used, which is why the model rather than the price is the useful signal.
Why prediction is the product
An advertiser is not buying a page view. They are buying a probability that a particular person will act, and that probability is what the profile produces.
This explains the collection pattern. Interests, purchases, location history, and social connections all improve the estimate, so all of them are worth gathering. It also explains why data is retained long after any obvious use, because behaviour over time predicts better than behaviour today.
The auction nobody sees
When a page loads, a request describing the available advertising slot is broadcast to many potential buyers, along with information about the person viewing it. Bids return, one wins, and the advertisement appears. The whole exchange takes well under a second.
The consequence is underappreciated. The description is sent to every participant, not only to the winner, so a single page view can distribute information about a person to dozens of companies that never displayed anything and have no relationship with them.
Those participants may retain what they received. Losing the auction does not oblige anyone to discard the data that came with the invitation to bid.
Attention is the scarce input
Revenue scales with time spent, which makes engagement the metric every advertising funded product optimises toward, whether or not anyone states it as a goal.
That objective is not neutral with respect to content. Material provoking a strong reaction holds attention better than material that does not, so systems ranking by engagement systematically favour it. This is a consequence of the revenue model rather than an editorial decision, and it appears wherever the model appears.
Collection exceeds requirement
A useful test when evaluating a service is whether the data gathered is necessary for the feature it supports.
A mapping application needs location to give directions. It does not need location while it is closed. A torch application needs no data at all. Where collection exceeds the functional requirement, the surplus exists for a reason, and that reason is usually the revenue model rather than the product.
Free has a shutdown risk
A product with no direct revenue from its users has no obligation to them and no income to defend, which makes discontinuation a normal outcome rather than a failure.
The practical consequence is data rather than inconvenience. When a service closes or is acquired, the dataset is an asset that transfers, and the terms binding the new owner are the ones in force at the time of transfer rather than the ones accepted originally.
What to ask before adopting one
Three questions separate the models quickly. How does this make money, which is often answered plainly in a filing or a help page. Would the feature work with less data than it requests. And what happens if the company is acquired, since datasets transfer with the business and the acquirer is bound by the terms in force at the time rather than by the original intent.
A service with a clear revenue model that does not depend on you is structurally more stable, because there is no incentive pressing it to collect more over time.
Note: paying for a service removes the incentive to profile but does not remove the capability. Several paid products collect as thoroughly as free ones, so the payment model is evidence of intent rather than proof of behaviour.