Data is at the core of improving accuracy in Predict and enhancing value. This article covers everything you need to know about improving the accuracy of your predictions--and it all starts with data:

  1. Upload historical data into Predict so you can make predictions against your unique, historical data. 
  2. Segment your data based on different email and audience types. 
  3. Keep your data up-to-date. 

1. If you haven't already, upload a dataset

It's true. Your audience is unique.

As fruitful as industry trends can be, there's nothing more valuable than understanding how your audience specifically responds to different words and phrases. This is where you can take advantage of the full potential of Predict. 

If you haven't already, upload a dataset in Data & Analytics.  Then, each time you generate a prediction, you can select your own dataset. As a result, you can use your brand's historical data to estimate how a subject line is likely to perform. This method is generally more accurate than comparing to only industry data. 

Is more truly merrier?

Technically, you can upload as many as 20,000 subject lines at one time, so feel free to upload as many as you like. The more linguistic variety in the tool, the more Predict has to work with.

However, quantity is not always quality. If you are aiming for accuracy, here are a number of tips to improve the accuracy of predictions and optimize value from our dataset insights.

2. Upload datasets for specific email types or themes

Not all emails types are created equal.

Separate your emails based on engagement

Trigger campaigns like cart abandonment or welcome emails have significantly higher open rates than promotional emails. Grouping them together in a dataset will skew your data.

The example below illustrates exactly this. This user combined order confirmation emails--with open rates above 50%--with promotional emails. Nearly all the emotions in this dataset perform below the average open rate because the average open rate of their trigger emails is very high.

In reality, several of these emotions perform well! Take out those trigger email subject lines from your dataset of ad hoc email subject lines, and you'll create a more accurate picture.

Separate your emails by theme

On that note, in addition to improving the accuracy of your dataset, segmenting by email types can reveal unique insights and trends. For example, one user uploaded subject lines for Black Friday emails from the last 10 years. They discovered a few trends, such as

  • Urgency was a top performer
  • Mentioning Black Friday in the subject line was key, 
  • "Free Shipping" didn't resonate, and 
  • Subject line length did matter.

Other ideas for segmenting your data by email type:

  1. Different touches of an email sequence--like touch 1, touch 2, touch 3
  2. Holidays
  3. Regular newsletters
  4. Regular promotions
  5. Time of year

Here's an example of a dataset for an email type, BOGO subject lines.

3. Segment by audience.

Not all audiences behave the same.

So we shouldn't bucket them together, right? Upload datasets for separate audiences so you target them more effectively. For example, subject lines sent to your active audience can be your "active" dataset. Subscribers who haven't purchased in 12 months can be in a separate dataset.

Here's a list of potential audience-specific datasets:

  • Actives
  • Inactives
  • Purchased in last 30 days
  • Female audience
  • Male audience
  • Specific brand follower or subscriber
  • Geographic location
  • Parents
  • Millennials

4. Update your datasets regularly.

What works today may not work tomorrow.

Time has an impact on the efficacy of certain emotions. Audiences change, seasons change, consumer confidence increases or decreases--there are a number of different factors which impact audience behavior.

Thus, in order to maintain accurate predictions, we recommend uploading fresh datasets every 3 months.


Data is at the core of improving accuracy in Predict and enhancing value.

The first step is uploading data into Predict so you can make predictions against your unique, historical data. The second and third step is segmenting your data based on different email and audience types. Finally, keep your data up-to-date. This will allow you to accurately improve subject lines which will resonate with your audiences.

Still need help?

If you aren't sure how to upload a dataset, please navigate to the Data Upload Guides section of the Help Center or chat with us by clicking on the icon in the bottom right.

Did this answer your question?