A/B test projection

A/B test projection uses neural networks to predict which subject lines perform best. Our model extracts linguistic features from winning A/B tests performed on Braze and uses those statistical language patterns to teach our AI what makes better subject lines.

Running a projection

In campaign composition, insert your message variants and their subject lines into the editor. When ready, go to the Target Audience step of the campaign creation flow. In the A/B Testing panel, select Run Projection.

A modal will open with the subject lines of any message variants you have already created. Optionally, you may insert additional subject lines (up to a maximum of ten) by manually entering one into the box and running the projection. Select Run Projection.

The subject line our AI predicts to be best will be highlighted with a Projected Winner label.

How accurate are the projections?

In testing, we found the projections to be about 70% accurate when picking between pairs of messages in real A/B tests. Consider this when interpreting the messages the model projects to win.

How do we use your data?

This feature learns from past A/B tests carried out on Braze. The actual copy of your or any Braze customers’ messages is never provided to the model. We first extract the high-level language patterns that predict winning messages in A/B tests. Then, we provide those patterns to our AI to teach it to discern which linguistic features constitute superior subject lines.