A Leading US Fintech Platform Acquires More Profitable Users at 20% Lower Cost
By the numbers
Overview
Company
Industry
Campaign Type
Snapshot
One of the largest consumer fintech platforms in the US, the company serves tens of millions of users across a suite of financial products through a mobile-first ecosystem.
Powering that ecosystem is a diversified acquisition strategy with hundreds of channel partners, with Google Search among the largest. The growth team had already iterated through multiple optimization strategies, including CAC-based bidding to drive signups and static value-based bidding to get closer to revenue. Both drove results, but the team saw room to go further by acquiring more users who actually request funds and return for repeat advances. They partnered with Voyantis to build predictive infrastructure purpose-built for fintech's complexity and their scale.
Impact at a glance
Reduced CAC 20%
while more than doubling average revenue per user
Doubled ROAS
within the first 8 weeks
Automated signal strategy
that adapts to Google's algorithms and recalibrates as the business evolves
Freed internal data science to focus on product
with fully managed infrastructure maintenance
Established a foundation for cross-channel expansion
and broader use of predictive data across the business
The challenge
The team knew they needed to acquire more users who request funds early and keep coming back - the behaviors that generated revenue for the business. The challenge was that the algorithm needs signals fast, and it's on the advertiser who knows their business and goals best to provide them. Requests for cash often happened weeks or months later, well outside Google's attribution window.
CAC-based bidding scaled the business, and static value-based signals got closer to revenue. But in both cases the team was relying on proxy events like account sign-ups and predetermined values by funnel stage, which didn't always correlate with who's still transacting at day ninety. The more Google learned from that data, the further it drifted from the users who actually generate value.
The only way to close the gap was to deliver dynamic, user-level predictions to Google within hours of signup, giving the algorithm a forward-looking view of each user's value before their behavior had a chance to prove it.
The solve
Figure out what to predict (and when)
Not every user behaves the same. Some request funds within days of signing up, others take months. Some transact once and never return, others become repeat users who drive compounding value. The prediction target had to be precise enough to separate those groups and delivered early enough to be useful.
Voyantis analyzed the company's first-party data and identified that 90-day LTV was the sweet spot in differentiating converters from those that would actually request cash and generate revenue for the business.
Model each user's long-term value from the moment they sign up
Fintech companies sit on some of the richest behavioral data of any industry, including transaction patterns, credit behavior, product engagement. Few industries can observe financial behavior this directly.
Based on the company’s unique growth context, Voyantis trained models on thousands of features across the company's non-PII data, identifying which combinations of behaviors correlate with long-term value. Unlike a standard LTV model that optimizes for average accuracy, Voyantis's models are built to maximize the spread between predicted user values and create enough separation to drive bidding decisions.
The models achieve 95% accuracy, surfacing patterns no manual analysis could uncover and generating user-level predictions within hours of signup.
Translate predictions into bidding signals
Unlike a conversion event that either happened or didn't, predictions are estimates that evolve as more data comes in. Getting a prediction into Google's system in a way the algorithm can learn from comes down to timing, formatting, and cadence. A mistimed signal or misformatted value sends Google off to chase the wrong users. Those small errors multiply across millions of bid decisions.
Voyantis manages this autonomously, generating predictions at signup, stripping hidden correlations before they reach the platform, and updating signal values as engagement data accumulates during the attribution window. Google receives signals in the format and cadence it can actually learn from, and a more accurate picture of user value as confidence grows.
Close the loop with continuous learning
Optimizing on predicted value creates a feedback loop. Predictions shape how Google bids. Bidding shapes which users come in. Those users shape the next round of predictions. When the loop works, performance compounds. When it drifts, the whole system degrades.
Voyantis monitors every layer autonomously, from data pipelines to model predictions to signal delivery to campaign performance. When models retrain, new versions are validated against existing ones before deployment so Google's learned bidding patterns aren't reset, and when signals degrade, fallback models activate automatically. The system stays calibrated as the business evolves, so no manual monitoring is necessary.
Voyantis’ impact
2X
increase in ROAS
20%
reduction in CAC
2X
increase in revenue per user
Acquiring more revenue-generating users at a lower cost
Within eight weeks of deploying Voyantis signals against their existing tCPA strategy, they company saw:
- 20% reduction in CAC
- 2X increase in ROAS
- 2X increase in revenue per user
The team was acquiring fundamentally better users at lower cost, and the gap between the two kept widening as Google learned from more accurate signals. The efficiency freed up budget to reinvest, giving the business room to scale further without sacrificing the quality of users coming in.
Adaptive infrastructure, built for scale and complexity
Voyantis delivered predictions to expose future value, signal engineering to turn it into performance, and closed-loop adaptation to keep it aligned as the business evolved. With the operational layer handled, the team's data science resources stayed focused on product, while the infrastructure continued learning in the background.
The impact extended beyond paid media. A more accurate view of user value across the business informed forecasting, planning, and how the team thought about growth beyond paid acquisition, giving them a foundation that gets more valuable as the business scales.
