Advertising marketplace needed to balance exploration and exploitation in assessing effectiveness of ads quickly to not waste money on poorly performing ones.
WHAT WE DID
Built an adaptive learning and optimization system to this multi-arm bandit problem to efficiently identify best performing ads. Pioneered Thompson Sampling to multivariate testing.
Improved margin and revenue. Approach was validated by Google also adopting the same Thompson Sampling technique.[/vc_column_text][/vc_column][/vc_row]