Overview

Follow this guide to learn how Comeback Score helps merchants improve customer retention by predicting which first-time customers are most likely to make a repeat purchase within 90 days.

Understanding Comeback Score

What Is Comeback Score?

Comeback Score is a predictive Boosted Event that estimates the likelihood a first-time customer will return to make another purchase within 90 days. Using machine learning, each eligible customer is assigned a binary score:

• 1 – Likely to make a repeat purchase within 90 days
• 0 – Unlikely to make a repeat purchase within 90 days

This predictive signal is delivered as a custom event through Elevar and can be activated in platforms like Meta, GA4, and Klaviyo to support smarter audience targeting, retention campaigns, and customer segmentation. Rather than treating every first-time customer the same, Comeback Score helps marketers make more informed decisions based on a customer's predicted likelihood to return.

Improving Customer Retention:

Most ecommerce brands lose a significant portion of first-time customers after their initial purchase. Comeback Score helps identify which customers are most likely to return, allowing brands to focus retention efforts where they can have the greatest impact. By surfacing predictive retention signals early, marketers can build more effective post-purchase strategies that encourage repeat business and long-term customer growth.

Comeback Score vs Traditional Retention Campaigns:

Traditional retention campaigns treat all first-time customers similarly. Comeback Score helps marketers prioritize customers based on their predicted likelihood to return.

Traditional Retention Campaigns:

• Targeting: Broad customer segments
• Decision making: Based on historical behavior or manual rules
• Campaign strategy: Similar messaging for all first-time customers

Comeback Score:

• Targeting: Customers predicted to return within 90 days
• Decision making: Machine learning-powered predictive signal
• Campaign strategy: Personalized retention, suppression, and win-back workflows based on return propensity