When a user volunteers granular data about their interests, you can use pre-existing correlates between those interests and purchasing probabilities. Those correlates can be derived from exogenous data generated from a subset of users with the same interests (who voluntarily accept more tracking for free premium features), experiments, educated estimates from experts, etc...

The above approach might not be perfect, but I would compare it to the recent trend towards organic food. Organic food is simply costlier and less efficient to produce. Nonetheless, a growing number of people wouldn't do without it. They only visit stores with organic meat and produce. Websites that completely eschew tracking and third party tracker data can be the "Whole Foods" of websites.