Over the past two decades, emails have been essential in the field of marketing as a way to establish branding and raise awareness of a topic.
To raise the efficiency of these email marketing campaigns, it is necessary to tailor subject lines, emails as well as other factors that are sent.
This tailoring is most often accomplished by performing A/B testing done by marketing specialists which involves several trial and error stages where the feedback is used to generate the next split test. However, it is possible to reduce the amount of time spent during split testing by using a data-driven approach to instantly predict the response to potential changes.
With enough data, this may allow hundreds of split tests to be “simulated” in a much shorter period of time.
The purpose of this project is for the company to position themselves as the Canadian market leader in online marketing and has identified that this such feature can provide a significant stimulus to their business growth.
The basics of the strategy implemented is to receive the email recipient’s information as well as the email information in order to give back a prediction success of both open and click through rate data for each recipient. The data is then parsed through and fed into Google’s BERT. From there, Euclidean Distance formula and Cosine Similarity is used to describe the distance between texts.
A proof-of-concept was first created with a prediction algorithm for subject lines and email text.
The strategy of this project is to use the recipient information with email information to predict the success of an email (the open rate and click-through rate). The figure below outlines the basics of this strategy.
The strategy to calculate the success rate stems from the assumption that the closer the match between the recipient and the email, the more likely it will be for the recipient to open the email and/or click a link.
Specifically, the main data is the textual information in the email, matched with a description of the recipient’s employer (assumed by the domain of their e-mail address).