You don't :/

Again at some point there is a part of subjectivity in how you analyze the dataset.

It is still better than having no idea what your users are doing with your product.

A/B testing with groups of random people put together in a room, using unreleased versions of the application.