How Upside Acquires More Revenue-Generating Users at Lower Cost with Voyantis
By the numbers
Overview
Company
Industry
Headquarters
Campaign Type
Snapshot
We used to run two campaigns that each solved half the problem, but neither optimized around a clear signal of value for our business. Voyantis gave us that signal, engineered specifically for how Google learns, with the infrastructure to evolve with our business, our customers, and the ad platforms. Now, we're running one campaign that drives better efficiency and brings in more revenue-generating users.
As one of the largest cashback marketplaces in the US, Upside reaches over 35 million consumers. Their business thrives when app users claim offers, submit receipts, and transact week after week. But those repeat customers often reveal themselves over weeks and months, while Google's algorithm stops learning after 7 days.
Upside was running two separate campaigns optimized for early actions like app downloads and offer activations, but neither separated casual users from future loyalists.
Upside needed a way to bridge that gap. The growth team enlisted Voyantis to predict 60-day LTV at sign-up and engineer signals that Google's algorithm could act on, replacing two campaigns with one predictive approach that delivers both efficiency and scale.
Impact at a glance
Validated a predictive LTV approach
with +44% ROAS, -15% CAC, and +15% in most valuable users
Proved efficiency holds at scale
with uplift sustained even at significantly higher spend
Implemented reliable infrastructure
with fallback models and real-time monitoring
Built to scale across channels
with platform-specific signal engineering
The challenge
Upside's Google acquisition strategy relied on two separate campaigns. One optimized for installs, an early action that gave Google plenty of volume but no indication of user quality. The other optimized for offer activations, a stronger proxy for value, but one that typically happened 10 days after sign-up - or three days outside Google’s conversion window. Neither identified the repeat customers who would drive long-term revenue.
Upside had built predictive models in-house but couldn't solve two problems: getting predictions to fire within those first 7 days, and engineering those predictions into signals the algorithm could act on. The team turned to Voyantis to prove that optimizing for predicted value would not only beat their current setup, but actually drive deeper business impact at the same spend and scale.
The solve
Voyantis and Upside designed a phased approach, structured for clean measurement and rapid iteration.
Test clean, scale fast
Upside started focused with 10 of 210 available geographic markets. Early results like lower CAC, lower cost per sign-up, and healthy pROAS came quickly.
Within weeks, Upside expanded to a full 50/50 geographic split. The test was straightforward. Could predictive optimization beat the combined performance of both legacy campaigns at the same spend level?
Build a model trained on Upside's anonymized, first-party data
Voyantis built a model trained on Upside's proprietary data to predict 60-day LTV for each user. The model processed thousands of behavioral and contextual signals - transaction patterns, engagement frequency, purchase amounts, gas type, card type - weighing positive and negative indicators together.
The 60-day prediction window was deliberate: long enough to capture true user value but short enough to keep the model trained on recent behavior. Voyantis’s predictions proved 91% accurate on average, giving Upside confidence that the model was capturing real user value.
Engineer signals Google can act on
Having a model that predicts user value is only half the battle. The other half is delivering that prediction to Google in a way the algorithm can digest.
Google was built to learn from real conversions like sign-ups, purchases, and installs, not probability scores. Voyantis translated Upside's predictions into a format, timing, and frequency that trains the algorithm and delivered them within one hour of sign-up, updating them as needed. This was fast enough to influence bidding while Google's learning window was open.
Scale with confidence
Upside wanted to ensure that their early wins held as they pushed more spend. They increased the budget significantly on the Voyantis-powered campaigns.
Even at higher spend, the Voyantis campaigns maintained lower CAC and continued to deliver ROAS uplift. Compared to the aggregated control group, the Voyantis campaign drove:
- 44% increase in ROAS
- 15% reduction in CAC
- 15% increase in most valuable users
Voyantis proved that efficiency and scale are compatible.
Infrastructure built to keep running
Reliability matters just as much as results. Upside needed a system they could trust long-term, one that could evolve with their business, their customers, and the platforms themselves.
Voyantis built in fallback models, real-time monitoring, and pipeline health checks that catch issues before they become problems. When something breaks, resolution happens within 24–48 hours with minimal lift from Upside's team.
Voyantis’ impact
44%
increase in ROAS
15%
reduction in CAC
15%
increase in most valuable users
A foundation for scale
With full migration complete, Upside's acquisition engine looks different than it did a year ago. The team no longer manages two separate campaigns trying to solve the same problem, and they don’t have to wait on predictions that arrive too late to make a difference.
Most importantly, the team has the data, the infrastructure, and the team to expand impact across campaigns and additional channels like Meta.
Google was the proof point, but we know Meta has its own learning logic and signal requirements. What gives us confidence is that Voyantis understands those differences. They engineered platform-specific signals specifically for how Google learns, and we trust them to do the same for Meta.
