How can I increase email open rates with machine learning?

Question

For example, I have email sending logs like this:

day_of_week| time  |   cta    | is_opened |is_clicked
1          |10:00AM|CLICK HERE|True       |False
7          |07:30PM|BUY NOW   |False      |False
...

I want to write a program to see "best performing day and time to send emails".

This example is only for sending day/time. I want I can add extra parameters (like CTA, sender name etc.) when I need it.

Is the machine learning best way to do it? (I have no experience in ML) I'm experienced with Python and I think I can use the TensorFlow to do it.

ps: These are marketing emails that we send our members, not spam or malware.

Are you asking us to help you scam people or make them click on potentially dangerous links that would fill their PCs with ads or malware? This is what pops in my head when I see such e-mail titles. — ForceBru, Sep 19 '17 at 15:02
If I understood your question correctly, I would suggest to look at linear regression algorithms. They are one of the most popular approaches in machine learning. — Egor4eg, Sep 19 '17 at 15:05
@Egor4eg I'm going to learn about linear regression, thank you! :) — Umut Çağdaş Coşkun, Sep 20 '17 at 06:17

score 1 · Accepted Answer · answered Sep 19 '17 at 18:25

There are two views for your case :

given day, time, etc. predict will be opened/clicked or not.
given cta, etc. predict the best day-time to send the email

For the first case, you can use Neural Net, or any classifier to predict it will be opened/clicked or not

For the second case, I assume this is your case, You may look at Multivariate Regression, because two variables you need to predict (day_of_week, time) may not be handled separately (e.g. by creating two models then predict day_of_week and time separately ). You need to predict two variables simultaneously. But you need to clean your data first, so it only contain opened/clicked email.

And of course, you can implement it using Tensorflow.

How can I increase email open rates with machine learning?

1 Answers1