How Semrush Reduced CAC by 17% by Identifying High-Value Users Earlier
By the numbers
Overview
Company
Industry
Headquarters
Campaign Type
Snapshot
Semrush had built a sophisticated acquisition strategy, running value-based bidding on Google with conversion values at each funnel stage. It was a strong foundation, but treated all users at the same stage equally.
The team saw room to push further with Voyantis. They wanted to give Google higher-fidelity signals earlier to identify valuable users at registration, not weeks later. The team knew their data could reveal which sign-ups would become valuable long-term subscribers. What they needed was the infrastructure to predict it early and deliver it to Google in real time.
Voyantis’s predictive growth platform helped them make that shift, transforming first-party data into real-time signals that guided Google toward Semrush's most valuable users.
Impact at a glance
Increased ROAS by 21%
by surfacing high-value users that their existing approach couldn't differentiate
Reduced CAC by 17%
while maintaining conversion volume
Scaled predictive signals to 99% of global Google Search campaigns
with platform-specific signal engineering
Unlocked programmatic model retraining and recalibration
as the business and product evolves
The challenge
Semrush operates on a freemium model. Users can register for limited free access or start a 7-day trial of paid features across three tiers: Pro for small teams, Guru for agencies, and Business for enterprises.
The earlier Google receives value signals, the better it can optimize. But Semrush users convert after a 7-day trial, and retention patterns take months to emerge. The later the signal, the less influence it has on targeting. Waiting for actual outcomes means spending budget on trial and error.
The team had built a 5-year LTV model in-house that worked well for strategic planning and budget allocation, but not for real-time activation. It ran monthly batch analyses on user cohorts who had already converted - valuable for understanding past performance but not for optimizing acquisition spend in real time.
Semrush needed predictions at the user level, generated immediately at registration, for users who hadn't converted yet.
The solve
Semrush integrated their first-party data with Voyantis's predictive growth platform to replace static funnel values with dynamic, real-time predictive signals, and guide Google’s algorithms toward their most valuable users.
Predict value at registration
Working with Voyantis, Semrush identified 120-day LTV as the North Star metric for the models - the point by which user value has largely materialized and retention patterns become clear.
Semrush's previous approach assigned the same value to every user at each funnel stage, but with Voyantis they were able to use first party data to model individual user value. The model processed thousands of features like page visits, sign-up source, onboarding quiz responses, subscription type, country, early engagement patterns, and more.
Voyantis’s platform generates predictions within hours of registration and refines them as users engage. A day-one estimate based on registration data becomes a day-eight estimate informed by trial behavior, with each update sending more precise signals to Google.
Predicting behavior 120 days into the future is complex. Voyantis' models achieved 87% accuracy by analyzing early behavioral signals - from registration data through initial engagement patterns - that held predictive power over the long term.
Differentiate across subscription tiers
A top-tier enterprise account generates significantly more revenue than a mid-tier Pro subscription, and retains differently over time. Voyantis calibrated the model to account for these differences, predicting not just whether a user would convert, but which tier they were likely to choose and how long they were likely to stay. This lets Google's algorithm distinguish between two sign-ups who look identical at registration but have very different long-term value.
Translate raw predictions into platform-native signals
Predictions alone are not enough. Google needs signals purpose-built to train the algorithms.
Voyantis's platform translates LTV predictions into conversion values and events delivered in the timing, format, and value Google expects. Instead of static values tied to funnel events, Google receives dynamic predicted values for each user. The algorithm can bid more aggressively for high-LTV registrations and more conservatively for users less likely to convert or retain.
This signal engineering runs automatically, integrating directly with Google Ads without manual intervention.
Maintain accuracy as the business evolves
Semrush operates in a fast-moving market. As AI reshapes the search landscape, the product evolves constantly with new packages, new pricing, new toolkits, new user segments, and more. A static model would quickly fall out of sync.
Voyantis handles the ongoing calibration. As Semrush introduces product changes, Voyantis retrains the models to reflect new data structures, updated user segments, and shifting conversion patterns. This includes migrating to updated infrastructure, incorporating new signals like add-on products and user type distinctions, and continuously monitoring model accuracy against actual outcomes.
For most teams, this kind of maintenance would require dedicated data science resources and constant attention. By partnering with Voyantis, Semrush gets a system that stays accurate as the business shifts without pulling their team away from the work of market differentiation.
Voyantis’ impact
21%
increase in pROAS
15%
reduction in CAC
15%
increase in LTV:CAC
Scale across global campaigns
Semrush deployed Voyantis's dynamic pLTV signals on a portion of US search campaigns first, comparing results against their existing approach of static values tied to funnel events. The difference was clear with campaigns bringing in customers at lower cost and with higher ROAS.
Based on these results, Semrush expanded Voyantis signals from 8 test campaigns to over 170 campaigns globally, growing their share of Google spend running on predictive signals from 30% to 80% across Search and Performance Max.
The rollout coincided with significant market pressure. CPCs in Semrush's category increased by over 80% during the measurement period - the kind of shift that typically erodes efficiency gains. Despite this, the predictive signals held their advantage. A causal impact analysis confirmed the uplift held steady at 21% post-rollout, even as market conditions shifted.
The results reflect a fundamental shift in how Semrush acquires customers. Instead of treating every sign-up the same, Google's algorithm now distinguishes between users likely to convert and retain versus those who won't make it past the free trial and bids accordingly. With predictive signals driving the majority of their Google spend, Semrush has built acquisition infrastructure that scales with the business.
