Performance Prediction scores are included alongside your self-service content to help you assess potential performance in market, and ultimately help you decide what to deploy to your audience. This article will walk you through everything you need to know about using these scores.
Want a quick summary? Check out this video.
Performance Prediction Score Basics
Performance Prediction scores are calculated using Persado’s Knowledge Base, which aggregates data across multiple brands, industries, campaigns, and languages. These scores tell you how a message is predicted to perform based on the grand scheme of messages in your industry in our Knowledge Base. Scores are presented between a range of 0 and 100, with 0 being the lowest and 100 being the highest score.
What data is used to calculate Performance Prediction scores?
We use engagement data (e.g, clicks) from your particular industry to calculate these scores. Our knowledgebase contains over 10 years of engagement data (200B interactions annually) attributed to tagged words and phrases. We measure how individual words and phrases impact performance, which enables us to predict how well any given message, composed of various words and phrases, will perform in market.
One caveat to keep in mind: The more any given word or phrase has been tested, the more certain we can be in the performance score. If your brand is using language Persado has never tested before, it might take some time to build up high confidence scores.
Performance Prediction Scores in Action
Performance Prediction scores are included to help you rank Variants as you review. We recommend using them as a relative indicator of performance between the Variants of a specific campaign.
Here are some best practices to keep in mind when considering Performance Prediction scores:
A small difference between scores isn’t something to fixate on, and you don’t need to pick the top score simply because it’s the highest. We recommend picking one of the best scores available to you for your campaign.
If you’re improving upon a control, as a benchmark, choose a Variant with a higher Predicted Performance score than the control. Aim for at least 10 points better.
You should use Performance Prediction scores to compare anticipated performance of Variants within a single campaign, as opposed to comparing scores across different campaigns, where the product or offer itself could have an impact on the score.
High performance scores do not correlate with uplift. The score is generated by a regression model that finds the probability that a specific phrase will perform better than any other phrase. Higher scores means a greater likelihood that the message will outperform other messages.
These scores are not guaranteed; the best way to ensure performance is actually to test and learn what works for your audience. Any content you choose will be validated again in market with metrics like clicks and orders that will inform our Knowledge Base’s performance over time. This is why it’s so important to upload your results to Persado Portal!
Please keep in mind that large language models (LLMs) are constantly learning and improving over time. In rare instances our model may produce a typo or "hallucinate" an odd phrase. Our NLP and Content teams monitor the outputs of our model to reduce these occurrences as much as possible!