Facebook has done enough feature launches that they train their teams to look for and expect a huge initial engagement peak as people explore the product. It's Deltoid 101.
I don't work there anymore, but gods I miss Deltoid. It was a fantastic tool. Google's experimentation UX is like a decade behind.
The best thing about (maybe this didn't make it to deltoid) was that they counted degrees of freedom for the t stats based on the number of days the experiment was running, which was incredibly effective in stopping people launching broken crap.