Probabilistic Customer Lifetime Value for a Robo-Investment Platform

VisualVest

Finance & Insurance
PyMCStreamlit

The Challenge

VisualVest needed to understand how much their investor customers were worth over the long run. That number drives some of the most consequential decisions a business makes — how much to spend acquiring new customers, which segments deserve the most attention, and what the future revenue base actually looks like.

The trouble was that nothing off the shelf fit. VisualVest operates as a subscription service with monthly billing, but revenue isn't a flat fee. It's a percentage of assets under management, which fluctuates with markets and customer behavior. The classic models used for customer lifetime value assume either a retail shopping pattern or a fixed subscription price — neither applies here. On top of that, European data privacy requirements meant the entire analysis had to stay within compliant infrastructure.

Our Approach

We started with the standard statistical framework for subscription businesses — modeling the probability that any given customer churns in a given month versus staying active. But we extended it substantially to reflect VisualVest's reality.

Rather than treating all customers as identical, we estimated individual-level churn tendencies informed by each customer's characteristics. Customers who hadn't yet left were handled correctly as ongoing observations, not thrown out of the analysis. And because revenue depends on assets under management rather than a fixed fee, we built that variable payment structure directly into the value calculations.

We also built an internal dashboard so VisualVest's team could explore the results themselves — projected lifetimes, expected revenue, and total customer value, all broken down by cohort and acquisition channel. Every estimate comes with honest uncertainty bands, so decisions are informed by what the model knows and what it doesn't.

Results

The model gave VisualVest something they hadn't had before: a principled, uncertainty-aware view of their customer base's long-term value. The team could finally segment customers not just by observable behavior but by predicted future worth, and they could do it with a clear sense of how confident those predictions were.

The engagement expanded into further work, and VisualVest's team noted that the willingness to keep refining and challenging the model — even after it was "done" — was something they hadn't experienced from other consulting partners.

PyMC Labs Team

  • Christian
  • Tomi
  • Ben Vincent
  • Thomas Wiecki
  • Larry
  • Ricardo

More Work in Finance & Insurance

Everysk

Uncovered private equity and venture capital performance from cash flow data alone, producing a return index that tracks established industry benchmarks.

Time-Varying Factor ModelsBayesian InferenceMCMC

Alva Labs

Built psychometric and validity models that substantially outperformed Alva Labs' existing assessment system, giving them both better trait estimates and hard evidence that their assessments predict job performance.

Item Response TheoryPsychometric ModelingHierarchical Bayesian ModelingOrdinal Regression

HelloFresh

Rebuilt HelloFresh's marketing measurement and testing infrastructure, cutting model runtimes from 20 minutes to 2 and overnight test pipelines from 6 hours to 6 minutes across 15+ global markets.

MMMGaussian ProcessesHSGPA/B Testing

Let's Chat, We Respond Fast

Tell us about your problem. We typically respond within 24 hours.

Schedule a Consultation