Marginal cost optimization system for bidding in hyper-localized campaigns of Adwords

• Grant number: 17/00065-6
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: November 01, 2017
• End date: April 30, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computational Mathematics
• Principal Investigator: Samuel Martins Barbosa Neto
• Grantee: Samuel Martins Barbosa Neto
• Company: Getninjas Atividades de Internet Ltda
• Principal investigators: Lucas Fonseca Navarro
• Associated researchers: Bruno de Almeida Silveira ; Pedro Henrique de Oliveira Sugimoto

Abstract

The digital revolution has significantly affected the marketing market. US digital marketing spending reached U$59.6 billion in 2015 and in this segment, Google dominated 90.2% of the search market in 2012. For a tech startup, the digital channel for customer acquisition is the most efficient in terms of return for the investment. In the case of a digital marketplace, the marketing problem becomes to optimally balance supply and demand according to the category. More than that, the service market segmentation should also take into account the geolocation of professionals and clients, since interactions happen face-to-face. From an economic point of view, for this marketplace to be sustainable it's also necessary to ensure that the cost of acquiring a new user from this marketing channel always has a positive return on investment. Google AdWords is the leading digital marketing tool for online searchers. It works on an auction system where advertisers provide the maximum bid they are willing to pay for certain terms. Based on these bids, advertisers compete for positions on the search page and the cost of clicks are set according to the second general price model. Several bid optimization tools exist on the market. However, these tools were developed based on e-commerce problem, that simply maximize the volume of conversions with a limit on the average acquisition cost. The problem of a marketplace services is different: we seek to balance the supply and demand with geographical restrictions. This translates into minimizing the cost subject to city restrictions and minimum volume of conversions based on local demand of professionals. In this project a bidding optimization tool that meet the needs of a service marketplace will be developed. To achieve this, a structure in which each campaign is associated with a segment and a city is created. The problem of optimizing the bid is then divided into two parts: optimize each campaign to balance the bids of each term so that all operate in the same marginal cost and set the optimal marginal cost that balances the supply and demand for each campaign. To deal with this problem, a Bayesian approach is proposed, in which we use aggregations in larger groups (state or campaign, for example) that are considered priors and data of similar keywords and words are used to our likelihood function. This way you can balance all the terms in a campaign for a given marginal cost. It is believed that this approach will make it possible to accurately track AdWords campaigns in cases that geotargeting is critical, there is the need to satisfy the demand balance in each segment efficiently and considering the marginal cost of each purchase. (AU)