RESEARCH RESULTS OF M32’S R&D PROGRAM IN ARTIFICIAL INTELLIGENCE
Conducted in partnership with ÉTS, this study provides new knowledge to better support publishers.
Montreal, March 17, 2022 – M32 would like to congratulate Huu-My NGUYEN and Pascal Giard for the recent publication of their study, as part of its research and development program in artificial intelligence, in partnership with the École de technologie supérieure de Montréal (ÉTS).
This partnership focuses on research and innovation, aiming to deepen the understanding of the complex digital advertising industry and develop cutting-edge tools. One key point that M32 wants to bring forward is the growing role of artificial intelligence in offering new solutions to publishers.
Indeed, to continue to be a leader in the field of monetization, M32 is focusing on the development of artificial intelligence (AI) and partnerships with researchers and universities. Such advances allow them to expand their knowledge on an ongoing basis and to create tomorrow’s products and services.
The purpose of this study is to be able to forecast eCPM trends — the cost per thousand impressions of online ads — using AI. This data is very important information for publishers who must set their floor prices as accurately as possible to maximize their sales and revenues.
M32 is pleased to share the full results of this research here for the benefit of all members of the industry:
Title: Forecasting the Time Series of eCPMs in Online Advertising
Online advertising has become the main channel of revenue for many web publishers. With the development of real-time bidding (RTB), publishers now are able to sell their advertising space in real-time, where the price is determined by the demand of the market at a time. In this thesis, we made an attempt to help publishers forecast the expected effective cost per mille (eCPM) of their ads in an RTB market in the next 30 days using the historical data of eCPM in the past 2 years. First, we explore the use of an Auto Regressive Integrated Moving Average (ARIMA) model to fit the time series of historical eCPM and make the forecast. Second, we examine the distribution of eCPM over a period and develop a confidence indicator which suggests the market volatility. The training and forecasting process is then integrated into our industrial partner data pipeline for evaluation in a production environment.