A Planning Approach to Revenue Management for Non-guaranteed Targeted Display Advertising

1 Oct 2020
Research

Operations Research and Operations Management

Huaxiao Shen, Yanzhi Li, Jingjing Guan, Geoffrey K.F. Tso

Published in Production and Operations Management, October 2020

There are severe deficiencies in the current online targeted display advertising sell ad resources selling mechanism where selling is typically arranged via an event-based auction. Based on the results of extensive numerical experiments on a twenty-day sampled log data set released by a collaborating firm in China, Yanzhi Li, Jingjing Guan, Geoffrey K.F. Tso of City University of Hong Kong and co-author Huaxiao Shen of Sun Yat‐sen University, Guangzhou demonstrate the effectiveness of a new planning framework.

“Our ad clicks forecasting method is more accurate than the traditional click through rate (CTR)-based method, and our solution approach is also very effective in producing high-quality solutions, with a revenue increase around 10% compared to the existing auction method,” they say.

The authors propose a planning approach to help ad publishers better allocate their ad resources in the spot market. The approach allows the publisher to take a holistic view of its resources and demand and thus to allocate the available resources in a more efficient way than can be achieved through an event-based auction mechanism. To implement their approach, they present a framework comprising two building blocks.

The first building block is a mixed-integer nonlinear programming model, in which the decision is an ad resource allocation plan that specifies the proportions of an audience unit’s impressions that are assigned to different ads and the objective is to maximize the publisher’s revenue. They propose an efficient algorithm for solving the optimization model to obtain a near-optimal solution with a bounded optimality gap.

The second building block of the planning framework, is an arbitrary point-inflated (API) Poisson regression model, with which they directly forecast the number of ad clicks with non-decreasing and concave functions of impression proportions. This is in contrast to the existing method of estimating the number of clicks based on CTR, which is less accurate since CTR is changing over time.