iamcici0424/marketing-mix-models

This repo contains the project details of the Applied Data Science Course, which is Marketing Mix Models regarding the advertising response measurement.

Marketing Mix/Multi-Touch Models: Advertising Response Measurement

School Term : Spring 2019

Group members:

• Chen(Cici) Chen [email protected]
• Zichen(Zoe) Huang [email protected]
• Jiaying(Claire) Wu [email protected]

Goal:

The main goal of this project is to evaluate the effectiveness of different advertising channels from four different perspectives from 500,000+ data. Meanwhile, we would learn from the advantages and disadvantages of the approaches and gain a more comprehensive understanding of mix modeling methods for continuous study.

Following suggestions by RICH FITZJOHN (@richfitz). This folder is orgarnized as follows.

proj/
├── data/ data used in this project
├── code/ codes run
└── doc/ instruction, pdf reports, and presentation slides if present

Introduction

• The ultimate goal of any marketing activities is to increase sales, either in short-term or long-term, and ideally each campaign or marketing channel should be evaluated based on the incremental profit, which is the additional sales we produce with advertising over what we would have sold without advertising, relative to its cost.

• By analyzing the correlations between marketing activities and the transactions, weakness and strengths of various channels, seasonality of consumer’s response, the marketing team could evaluate the cost efficiency and thus move forward to optimize the resources allocation for marketing spending, and strategically plan marketing events to improve the effectiveness of marketing efforts.

• Among diverse marketing activities, advertising is a major mean for branding and informing. Both online (emails, social media, displays, etc.) and offline (direct mails) advertisements play an indispensable role in marketing. As placing ads is expensive in terms of the cost of time, recourses, and budget, one most frequently asked question from the management team is that: how does my advertising work? With this question in mind, our team leveraged four methods to tackle the advertising effectiveness measurement problem: last-click attribution analysis, holdout testing (experimentation), marketing mix models and model-based attribution analysis.

Data

The dataset we used for this project is a synthetic one simulated by the Elea McDonnell Feit, Marketing Professor of Drexel University, and is organized from three perspectives: customer, impressions, and transactions. Data generation method and our inspiration for this project could be found here:

• Customer Dataset
• Impressions
• Transactions

Code

• Main Code
• R Shiny App
• Codes for Presentation Slides

Reports

• HTML Final Report
• HTML Presentation Slides
• PDF Final Report