How a Global Crypto Platform Built Predictive LTV from Anonymized Data
By the numbers
Overview
Company
Industry
Snapshot
The growth team at a leading crypto platform set out to build predictive intelligence into their decision-making. Without forward visibility into user value, they were allocating budget based on lagging indicators - unable to tell which users would drive long-term revenue until it was too late to act on it.
Building this in-house would require specialized ML infrastructure, months of experimentation, and ongoing maintenance across multiple platforms and geographies. The team partnered with Voyantis to move faster.
Impact at a glance
Enabled data-driven budget allocation
across platforms, channels, and geographies
Achieved ~90% model accuracy
using only anonymized, non-PII data
Deployed in weeks,
saving months of in-house development
Embedded predictions directly
into existing BI infrastructure
The challenge
The platform generates revenue from active traders, but at acquisition, a future power user looks identical to someone who will never return. That difference only becomes clear over time.
Predicting user value would solve this, but in crypto, the inputs are sparse. Users connect anonymously, move between platforms freely, and expect privacy by default. Most prediction approaches assume rich user profiles - data this platform didn't have and didn't want to collect.
The team needed accurate predictions within those constraints, across multiple platforms with different user behaviors and geographic segments.
The solve
Voyantis built predictive LTV models from fully anonymized data and integrated them directly into the platform's existing BI infrastructure.
Define the prediction target, based on the company’s unique growth context
Voyantis started by grounding the models in how value actually materializes—analyzing user cohorts and mapping which behaviors correlate with long-term revenue. They defined prediction targets at two time horizons: long enough to distinguish active users from one-time visitors, short enough to inform near-term decisions.
Build and calibrate models from sparse inputs
Most platforms assume they need rich user profiles to predict value. Voyantis proved otherwise. The models incorporated thousands of features—behavioral patterns, contextual signals like device language, browser, operating system, and geography. How users entered the platform and their early engagement patterns proved especially predictive.
Voyantis calibrated for outliers, ensuring predictions reflected realistic user distributions rather than being skewed by extreme cases.
Embed predictions into existing infrastructure
The growth team evaluates performance by platform, channel, and geography. Voyantis mapped predictions directly onto the views they were already using. Now the team compares predicted revenue per user against actual results to monitor performance in real time.
Voyantis maintains the models as the platform launches new products, enters new markets, and adapts to shifting user behavior. The infrastructure evolves alongside the business.
Voyantis’ impact
90%
model accuracy
6+ months
of development time saved
A foundation for data-driven growth
What would have taken months to build in-house was live in weeks. The platform now operates with a decisioning layer built on predicted long-term value, without compromising user privacy.
Predicted LTV now informs how the team allocates budget, evaluates channels, and scales across regions. What used to be guesswork is now grounded in data.
