How Alibaba’s OCPC Algorithm Boosts ROI and Platform Revenue in Taobao Ads

Recently, a paper titled “Optimized Cost per Click in Taobao Display Advertising” from Alibaba’s Precision Technology team was accepted at the KDD conference.

Research problem : Traditional fixed‑bid ad systems perform coarse‑grained traffic matching, leading to inefficiency and short‑term revenue focus.

Research scenario : The study focuses on Taobao’s mobile CPC display ads, where multiple metrics such as ROI, GMV, CVR, CTR, and RPM must be balanced.

Proposed OCPC smart‑bidding algorithm : A two‑level optimization framework that first sorts ads by estimated revenue (eCPM) and then adjusts bids to maximize a composite ecosystem metric while respecting advertiser ROI constraints.

ROI optimization

Expected ROI per click is defined as the expected GMV divided by the cost per click:

Further derivations lead to the linear relationship between ROI and average conversion rate (Equation 2) and the bid adjustment bounds (Equation 4).

Comprehensive metric optimization

Within the bid adjustment bounds, the algorithm solves a constrained optimization problem (Equation 5) that maximizes a composite metric f() under the eCPM ranking. Example objective functions include maximizing total GMV or a weighted combination of GMV and platform revenue (Equation 6).

Experimental validation

Offline simulations and online production tests on single‑item ads show percentage improvements in RPM, GPM, CTR, CVR, and PPC (Table 1, Table 2). Online results indicate that 67 % of campaigns achieved simultaneous gains in GPM and ROI, while 24 % experienced a quantity‑quality trade‑off with overall conversion increase (Table 3). Similar gains were observed for banner ads on the mobile homepage (Tables 4 and 5).

Future work

The team plans to incorporate deep learning and reinforcement learning techniques to further improve real‑time traffic value estimation and bidding decisions.