I think the term is A-B testing. When a company wants to see what effect a change will have, they don't force it on everyone at once, just on a certain number of people (A), and then see what happens compared to the rest (B).

This is why you'll always get people saying, "Huh, I haven't seen that. It's not doing it for me on [browser]." They're in the (B) group for now.

The data the company wants is to know if the test people like the change, or are at least willing to put up with it? Or do they spend less time on the site? If so, how much? If the results are within their predictions, they'll expand the testing until everyone is in (A).