At the core of Persado Predict is--you guessed it--Predictions. These will give you a gut-check of how a subject line is likely to perform. Let's dive in a bit deeper and look at how we calculate uplift and various factors which impact your prediction.

Ready? ­čĹŹ

Understand how uplift is calculated

Let's start with the data. The predicted uplift is dependent on which industry and dataset you choose to compare to. Each time you generate a prediction, select which industry and dataset you'd like to compare against.

This is when the "automagic" happens:

  • Our Natural Language Processing (NLP) algorithm scans your subject line, detects and tags emotional phrases, and identifies key words in your subject line.
  • Then, it scans the performance of these emotions and similar phrases for the selected industry and dataset.
  • Finally, it calculates an open rate for the subject line based on the historical performance of the language in your subject line.

Based on how the language performed historically for the industry or dataset, we calculate an estimated response rate. Uplift is then calculated using the estimated response rate and the average response rate for that industry or dataset. Here's the uplift formula:

in which U is uplift, Rp is Response predicted, and Rc is the average response rate. Multiply U by 100 to get a percentage.

What does my prediction mean?

Keep in mind that each prediction is an estimate. It's calculated based on the average response rate of the industry or selected dataset and the predicted response rate of the subject line. We also look at how those particular emotions and/or words and phrases historically performed.

 Therefore, keep in mind that the accuracy of predictions depend on a few factors:

  • The volume of data in your dataset. The more the better.
  • The frequency the emotions, words, and phrases in your subject line have been tested in your dataset/industry.
  • How 'clean' your dataset is. If you mix an active and inactive audience together, or mix trigger emails with promotional emails, your dataset will be noisy. Segment your datasets into different audiences or email types so you are comparing apples to apples.

Ultimately, Predict won't have perfect accuracy, but rather use it as a brainstorming tool which will suggest great alternatives and provide different ways of conveying the same message.


If you have further questions, please chat with us or email Thanks!

Did this answer your question?