Case study
Predicting the Future: How Advanced Probability Models Rewrote SaaS Economics
The Opportunity
Refining CLV calculations for optimized marketing
The client, a major SaaS provider with 12 million subscribers on monthly billing, aimed to maximize the value of its customer base. While their software products were widely adopted, marketing decisions were hampered by unreliable customer lifetime value (CLV) estimates, capping their ability to scale efficiently and hit aggressive growth goals.
Accuracy was undermined on multiple fronts: simplistic averages that ignored customer variability; overfitted machine learning models prone to errors in dynamic environments; and a lack of robust prediction for churn, upsells, and long-term revenues amid shifting market conditions. The imperative was evident: develop a precise, generalizable CLV framework that could inform strategic allocation of their $100 million annual marketing budget, focusing spend on high-potential segments while minimizing waste on low-value ones.
Initially, we lacked refined models tailored to their data. Subscriber behaviors varied widely, with heterogeneity in usage, demographics, and channels complicating forecasts, and traditional methods inflated risks of overfitting or underperformance. The revamped approach was crafted to leverage these differences as strengths, incorporating probabilistic rigor to deliver actionable insights and drive sustainable, profitable expansion without escalating inefficiencies.
The Solution
Implementing advanced probabilistic modeling
To address these limitations, Deimos-One reframed CLV from simplistic averages to a robust, heterogeneity-aware system, beginning with its foundational definition. Customer Lifetime Value or CLV is the summation of all customer revenues over a time horizon and discounted to present value. This is a textbook example and is inadequate for real business applications. Predicting churn, upsell revenues, and variable revenues across long time horizons are the cruxes of CLV.
We worked to solve this problem by using a mixed probability model called the shifted beta-geometric distribution. This model is superior to linear regression and averages. This model is superior to machine learning models. Why? The answer is generality. Probability distribution models provide a high degree of generality that strikes the perfect balance between accuracy and avoiding overfitting.
Customers don’t fit a normal distribution as large differences exist between customers. Let’s explain and investigate what this means.
Graphed below is a comparison of four CLV model approaches: linear regression, geometric distribution, machine learning, and shifted beta-geometric distribution. Also included is the actual reality (dotted line).
The shifted beta-geometric distribution accounts for heterogeneity or differences among customer segments. This is in contrast to CLV averaging models and linear regressions. Machine learning models can achieve greater accuracy but suffer overfitting/loss of generality and are black box at times. Shifted beta-geometric is not prone to overfitting and accounts for differences in customer churn, usage minutes/tokens, upsells, marketing channels, and demographics.
This precise modeling allowed the client to reallocate their $100 million annual marketing budget—directing investment toward high-value customers while limiting spend on lower-value segments. As a result, revenues grew 7% on an annualized basis and overall customer churn fell by 2%.
The graph below shows the accuracy of the beta-geometric CLV model. The model is within 0.25% to 1% variance to actual data.
The Impact
Precision CLV Drives Marketing Efficiency
Annualized revenue growth from targeted marketing reallocation, focusing on high-CLV segments for compounding returns.
Reduction in overall churn, as precise predictions enabled retention strategies tailored to at-risk groups.
Marketing budget optimized, shifting spend from low-value to high-potential customers without increasing total outlay.
The client prioritized what counts—accurate forecasting, efficient resource allocation, and sustained subscriber loyalty—and tracked progress rigorously. By integrating the shifted beta-geometric model into dashboards linked to real-time churn, upsell, and revenue data, decisions accelerated from months to days.
This precision streamlined marketing workflows, routing budgets to channels and demographics with proven high CLV while curtailing ineffective campaigns. Outcome: minimized waste, elevated retention rates, stabilized acquisition costs, and empowered teams with data-driven confidence.
It enhanced strategic agility and long-term profitability. Subscriber satisfaction rose, with improved personalization boosting engagement metrics. Clearer CLV insights allowed marketing and product teams to innovate faster, achieving record-high retention and positioning the company for ongoing market leadership.
Lessons Learned
Strategic Roadmap
Model for heterogeneity in CLV outcomes
Rebuild CLV calculations end-to-end—incorporating churn probabilities, upsell potentials, variable revenues, and demographic variances—so every factor informs predictions and feeds back into marketing optimization. Standardize data inputs, lock in time horizons + discount rates, and embed segment-specific metadata so models learn from diverse behaviors, not simplistic averages.
Operating Model
Involve stakeholders in model design
Craft the framework with marketing, data science, finance, and product experts from the start. Encode real-world variables like upsells, demographics, and retention rules into probabilistic structures. Stakeholders validate that modeling aligns with operational realities, ensuring deployments are segment-ready, minimize iterations, and maintain precision as subscriber bases evolve.
More Enrollments
Prioritize generality over complexity
Optimize for balanced accuracy, not hyper-specific fits. Leverage shifted beta-geometric distributions to handle customer differences, outperforming averages or linear regressions while avoiding machine learning's overfitting pitfalls. Integrate usage tokens, channel effects, and churn drivers cohesively—yielding durable gains in forecast reliability, budget efficiency, and decision speed across teams.
Adoption & Scaling
Success demands proven revenue & churn shifts
Deliver models + a metrics hierarchy, not isolated equations. Dashboards connect CLV outputs to marketing spend → acquisition channels → retention outcomes → revenue uplift, monitoring heterogeneity, overfitting risks, and segment yields. If churn rises or growth stalls, refine distributions, variables, or integrations—and validate via 7% revenue boosts, 2% churn drops, and optimized $100M budgets.
Related Case Studies
Helping a debt relief company achieve 5,900% lead lift with stable CAC
Reframing consumer finance storytelling with Unreal Engine
Answering the Age-Old Question of “Lift” Using Synthetic Control
Are you ready to scale your business?
Complete the form below and tell us about your project or describe a challenge your company needs help solving. We'll schedule a free 20-minute strategy session with one of our experts to discuss how we can work together and help your business grow!