Why LTV Beats Conversion Value
Bidding to immediate purchase value optimises for the first transaction. Bidding to predicted LTV optimises for total customer worth, which often correlates poorly with first-purchase value. Bidding to predicted LTV typically improves blended payback by 20 to 40 percent in repeat-purchase categories.
Building a Predicted LTV Model
Most teams start with cohort-based proxies: first-purchase product, first-purchase channel, and discount code used at acquisition. Even simple proxies outperform raw purchase value when the cohort signal is strong.
Sending Predicted LTV to the Platform
Pass predicted LTV as the conversion value through the Conversions API or Enhanced Conversions for Leads. Platforms then optimise towards higher predicted-LTV customers within your bid cap.
Common Pitfalls
Sending zero or negative LTV values for any cohort breaks bidding. Using model predictions with low confidence intervals introduces noise that actually hurts performance. Start with conservative cohort proxies before deploying full machine learning models.
Frequently Asked Questions
Not for cohort-based approaches. Full ML models benefit from data science but aren't required to get started.
Most accounts see directional lift in 30 to 60 days, with stable lift in 90 days.
Subscription, ecommerce with repeat purchases, and lead generation businesses with downstream qualification.
Yes. Use a geo split or campaign-level holdout to measure incremental lift.
Yes. Use Maximize Conversion Value with optional target ROAS and pass enhanced conversion values.