Hydrant drives 2.6x higher conversion and 3.1x higher revenue per customer with Pecan AI churn prediction built in two weeks

Background

Hydrant relied heavily on email marketing but had no reliable way to prioritize which customers to target with retention offers. Sending undifferentiated discount offers wasted budget and trained customers to expect discounts rather than rewarding loyalty.

What Was Implemented

• Pecan AI churn prediction model analyzing 180-day purchase history per customer
• Numerical churn probability per customer (not fixed-bucket segmentation)
• Winback model for customers with 60–180 days of no purchase
• Pecan integrated into existing stack: Snowflake (data source) → Pecan (modeling) → Klaviyo (campaign execution)
• Model built and predictions generated in 2 weeks
• A/B testing of targeted winback offers vs. control groups

Results

Pecan's models predicted actual churn with high accuracy: 83%+ of customers flagged as most likely to churn did churn. In winback campaigns, customers targeted using Pecan predictions achieved 2.6x higher conversion rates and 3.1x higher revenue per customer than control groups. The model was built in 2 weeks . The book's claim of "35% churn reduction" is not supported by the Pecan case study (unverified — not found in sources fetched).

Lessons

• Numerical churn probability scores (vs. fixed segments) allow marketers to set flexible, contextual thresholds for who receives retention offers
• Winback campaigns that target customers with the *lowest* re-purchase probability—and pair them with offers—can yield outsize conversion lifts
• Two-week model build times show that predictive analytics need not require months of data science investment
• Integrating predictions directly into the marketing automation platform (Klaviyo) closes the loop from model to campaign execution