PREDICTIVE LTV SIGNAL OPTIMIZATION
Case Study: This represents a composite analysis based on industry data and AdZeta's methodology. Results are illustrative of potential outcomes.

DTC Supplement Brand Lifts ROAS 53% and Day-90 Retention 41% with AdZeta's ValueBid™

A direct-to-consumer supplement brand selling multi-SKU wellness stacks on subscription replaced first-order conversion bidding with predicted Day-90 retention value using AdZeta's ValueBid™. The result: a 52% ROAS lift, a 41% increase in Day-90 retention, and a 2.1x increase in the share of new customers who built a multi-SKU stack within their first 60 days.

DTC Supplement Brand Lifts ROAS 53% and Day-90 Retention 41% with AdZeta's ValueBid™
PROJECT HIGHLIGHTS
01

Client Profile: DTC Supplement Brand with Multi-SKU Wellness Stacks

A growth-stage DTC supplement brand selling protein, greens, sleep, and recovery formulations on subscription. The economics live in two milestones: customers reaching Day 90 (the cycle-3 retention floor), and customers building a multi-SKU stack of three or more products. Both milestones predict 12-month LTV roughly 5x baseline.

02

The Challenge: Day-1 Conversion Hides Day-90 Reality

Roughly 47% of new supplement customers churn before their second auto-ship. The auction has no visibility into this. The bidder optimizes toward Day-1 conversion, where churners and stayers look identical. The brand was paying full retail CAC to acquire customers half of whom would never reach cycle 2.

03

Previous Marketing Strategy & Limitations

Campaigns ran on Meta Advantage+ Shopping with first-purchase value optimization, and on Google Performance Max with tROAS configured against AOV. The team had invested in creative that emphasized stack benefits, but the bidder kept skewing toward single-bottle trial buyers because they checked out fastest. Audience tweaks did not solve the underlying signal problem.

04

The AdZeta Solution: Day-90 Retention Probability via ValueBid™

AdZeta deployed ValueBid™ to predict each visitor's Day-90 retention probability and 12-month stack revenue from non-PII first-party signals (formulation interest depth, browsing path, stack-builder usage, regimen survey responses). Predictions flowed into Google via OCI and Meta via CAPI as the conversion value the bidder optimized against.

05

Strategic Implementation & Execution

Integration ran from Recharge, Shopify, and the brand's data warehouse into the AdZeta pipeline in Week 1. Model training completed in Week 2 with AUC 0.88 on Day-90 retention prediction. ValueBid™ went live in Week 3, with a six-week paired control on 50% of media spend. Following the relearning period, ValueBid™ became the default conversion signal across both platforms.

06

Quantifiable Outcomes & Business Impact

At Week 16 post-relearn, blended ROAS was up 53%, Day-90 retention had climbed 41%, and the share of new customers building a multi-SKU stack within 60 days had increased 2.1x. The brand reallocated $180K of monthly spend from single-product Meta campaigns into stack-builder Google Performance Max, where ValueBid™ identified the highest pLTV impressions.

Understanding the Problem

The Core Challenge: The Auction Cannot See the Cliff at Day 90

In supplements, customers fall off a cliff between auto-ship 1 and auto-ship 2. The brand's economics depend on customers crossing that cliff. The auction has no idea the cliff exists. It optimizes for the same Day-1 conversion regardless of whether that customer will be active in 90 days or gone in 30. Without a forward-looking retention signal in the bidding objective, the bidder will always favor the cheapest cliff-faller over a more expensive long-term subscriber.

Supplement Brand funnel
without
AdZeta
Broad Wellness Shoppers
Day-1 Conversions (Mixed Retention Profile)
~47% Churn Before Auto-Ship 2
Single-Bottle Buyer Concentration
Optimization Goal
First-Order Conversion Volume
Supplement Brand funnel
with
AdZeta
pLTV-Targeted Stack Builders
High-Retention-Probability Conversions
Day-90 Retention Above 65%
Multi-SKU Stack Concentration
Optimization Goal
Predicted Day-90 Retention + 12-Month Stack Revenue
01
Day-1 Conversion Volume Does Not Predict Profit

Supplement unit economics break down at the cycle-1 churn rate. A 47% churn rate before auto-ship 2 means roughly half of every CAC dollar gets recovered, then evaporates. Optimizing the bidder for more Day-1 conversions just buys more of the same problem at higher volume.

02
Stack Adoption Is the Real LTV Signal, and It Is Invisible to the Bidder

Customers who build a 3+ SKU stack have a 12-month LTV roughly 5x higher than single-product buyers. The bidder cannot see stack intent at the moment of the first conversion. Without a predictive signal, the auction treats a single-bottle trial buyer and a stack-curious shopper identically.

03
Discount-Driven Acquisition Pollutes Retention Models

Heavy first-order discounting (40 to 50% off cycle 1 is standard in supplements) pulls in price-shoppers who fail to convert at full cycle-2 pricing. ValueBid™ models trained without discount-aware features systematically over-predict LTV for these cohorts. Discount-aware feature engineering was a prerequisite, not an afterthought.

ADZETA'S SOLUTION

The Strategic Approach: How AdZeta Delivered for the Brand

ValueBid™ replaces the first-order conversion signal with a real-time score predicting Day-90 retention probability multiplied by 12-month stack revenue. That score becomes the conversion value the bidder optimizes against. The auction relearns on the new objective and starts winning impressions for the customer profile that actually drives profit.

  • Day-90 Retention + Stack Revenue Modeling

    AdZeta's brand-specific model jointly predicts Day-90 retention probability and 12-month stack revenue using formulation interest depth, browsing path, stack-builder usage, and quiz response patterns. The composite pLTV score is the conversion value sent to the auction.

  • Discount-Aware Feature Engineering

    The model includes explicit features for first-order discount tier, promo source, and price sensitivity. This prevented systematic LTV over-prediction on heavily discounted cohorts that historically churn at twice the rate of full-price acquisitions.

  • Real-Time Signal Delivery via Google OCI and Meta CAPI

    Predictions are delivered as conversion values to Google's Offline Conversion Imports and Meta's Conversions API in real time. The buying team kept Google Performance Max and Meta Advantage+ Shopping campaigns untouched.

  • Cohort-Level Performance Reporting

    AdZeta's reporting layer breaks ROAS, retention, and LTV out by acquisition cohort, model version, and platform. The brand's finance team uses the cohort report to forecast quarterly subscription revenue with materially tighter bounds than they had pre-rollout.

A/B Test
Control Group: First-Order Conversion Bidding (50% Hold-Out for Six Weeks)

Google Performance Max with tROAS configured against AOV. Meta Advantage+ Shopping with first-purchase value optimization. The default configuration the brand had been running for 18 months.

Experiment Group: ValueBid™ Day-90 Retention Signal

Same campaign structure, same audiences, same creative. ValueBid™ Day-90 retention probability multiplied by 12-month stack revenue replaced AOV as the conversion value sent to Google OCI and Meta CAPI.

Google Ads
Meta Ads

53%

ROAS Improvement

Blended Google + Meta ROAS at Week 16 post-relearn

41%

Day-90 Retention Lift

Share of new customers still active subscribers at Day 90

2.1X

Multi-SKU Stack Adoption

New customers building a 3+ SKU stack within 60 days

36%

Decrease in Discount-Driven Churn

Cycle-1 to cycle-2 churn on heavily discounted acquisitions

MEASURABLE IMPACT

Stop Buying the Same Churning Customer at a Cheaper CAC

The brand was running a sophisticated stack of audiences and creative. The auction was still pulling in single-bottle trial buyers because that was what AOV bidding rewards. ValueBid™ flipped the objective from "cheapest converter" to "highest predicted Day-90 stack revenue," and the customer mix shifted accordingly. The auction is the highest-leverage point in the whole stack. Retention modeling that lives outside it leaves money on the table by design.

We knew our cycle-1 churn was the biggest profit lever in the business. We had a great churn model. It just lived in the warehouse and never made it into the bidder. ValueBid™ was the first thing that put our retention prediction directly in front of Google and Meta as the conversion value. Within a quarter, our auto-ship 2 retention had moved 41 points and our finance team could finally forecast subscription revenue without quarterly heartburn.
Marcus A. Head of Growth

Ready to Bid on Day-90 Retention Instead of Day-1 Conversion?

If your supplement brand's profit lives at cycle 2, 3, and beyond, but your auction only sees cycle 1, AdZeta's ValueBid™ closes that gap. Predicted retention and stack revenue delivered to Google OCI and Meta CAPI in real time.

Lets's Talk
Scroll