Theory: - recommendations are stickier with higher adoption if they follow ux patterns (urlbar) - contextual recommendations are awesome Unknowns: - balancing pushiness to value Branches: - no recommendations at all (count 'would offer') - push open *once* (under some conditions) - passive only - NOT DOING: push always, b/c that would be really annoying to users Measured outcomes by ARM: - By site: - offer vs accept vs 'not for this site' - Global - clicks on never show again - took an offer - Uninstall addon - General Firefox retention (total days) Prior art: There is a concurrent study around heuristics for recommendations. This is more about understanding pushiness and placement of recommendations. Rollout Plan: Channel - Release Deploy date - 08/11/2016 Close enrollment - 08/25/2016 Sample size: 1k participants per arm 3k total participants
Requesting r+ for study validation.
Resolved over email. Jgaunt will be doing analysis here as well.