How a Global Subscription Company Reduced CAC by Over 20% with Predictive LTV Signals
By the numbers
Overview
Company
Industry
Campaign Type
Snapshot
Many users start with a product purchase that includes a free trial. The team had been running tCPA campaigns optimized for purchases - a reliable proxy for volume, but one that couldn't distinguish long-term subscribers from one-and-done buyers.
The sooner the team could feed quality signals of long-term value to Google, the faster the algorithm would optimize toward subscribers and the less budget would be wasted. Unfortunately, that subscription data came in too late to inform campaign optimization.
Building this in-house would require dedicated data science resources, ongoing model maintenance, and platform-specific signal engineering. The company enlisted Voyantis - bringing cross-industry expertise in predictive LTV modeling and signal engineering - to build, engineer, and maintain predictive signals that train the algorithm to find future subscribers.
Impact at a glance
Reduced cost-per-subscriber by over 20%
while increasing trial-to-subscriber conversion rate 10-15%
Improved ROAS 10-15%
while scaling ad spend by 25%
Maintained ~90% model accuracy
through seasonality, pricing changes, and market shifts
Deployed custom models across multiple funnels,
tuned to their unique behaviors
The challenge
The patterns that separate subscribers from one-time users - how often they engage, which features they explore, how quickly they return - don't show up until weeks later.
That timing gap is expensive. While waiting for real subscription data, the algorithm optimizes for sign-ups - a proxy metric that treats every new user the same. Budget flows toward whoever converts fastest, not who brings in the most value for the business.
As ad platforms increasingly automate bidding and targeting, signal quality becomes one of the few levers growth teams control. The team needed to stop optimizing for more sign-ups and start optimizing for future subscribers.
The solve
Voyantis built predictive bidding signals of subscription likelihood, helping the team shift from optimizing for sign-ups to optimizing for subscriber value.
Predict subscription intent early
Voyantis helped the team define the right predictive target - likelihood to convert to a paid subscription by a defined window, the point where trial behavior has stabilized and most users have either converted or canceled.
Using the company's anonymized, first-party data, Voyantis built models that identify which attributes early in the user journey correlate with subscription intent. The models process thousands of features to surface patterns most humans miss: subtle combinations of engagement, timing, and context that distinguish future subscribers from one-time users.
With Voyantis, the team generates predictions within hours rather than waiting weeks.
Translate predictions into value-based bidding signals
Google doesn't know what to do with a probability, because it’s trained to ingest conversion events with dollar values. Voyantis translated subscription-intent scores into the format platforms need: conversion events with predicted values, delivered within hours.
Instead of sending a binary "signed up" event, the team now sends "this user is worth $X based on their likelihood to subscribe." The algorithm starts optimizing toward high-value users immediately, rather than spending weeks learning from incomplete data. As more data comes in during the trial period, Voyantis updates predictions and sends refined signals back.
Adapt as the market and business evolves
Businesses don't stand still, and prediction patterns don't either. Over the past year, the company navigated pricing experiments, product launches, seasonal demand spikes, and competitive shifts. Each changed the profile of who was signing up and why.
Without automated retraining, this would mean constant manual intervention, such as monitoring drift, rebuilding models, redeploying. A cycle that typically takes weeks or months. By the time new models are live, the market may have shifted again. Voyantis's always-on infrastructure monitors performance continuously and retrains models automatically - no manual intervention, no waiting for data science resources.
Customize strategies for each funnel
The initial funnel was just the starting point. The company has multiple paths to subscription, each with different user behaviors and conversion timelines. Rather than forcing a one-size-fits-all model, Voyantis built custom prediction infrastructure for each.
Voyantis’ impact
Over 20%
reduction in cost-per-subscriber
10-15%
improvement in ROAS
10-15%
increase in trial-to-subscriber conversion rate
A foundation for expansion
What started with one funnel now powers acquisition across the business. Multiple paths to subscription, each with different conversion timelines and user behaviors - now run on predictive infrastructure built specifically for them.
Next up: expanding to additional channels and campaign types. Same business, multiple optimization problems - each now running on signals tuned to what actually predicts value. With Voyantis operating as an extension of the team, each new funnel or channel builds on a foundation that's already proven.
