Customer Lifetime Value (CLV) Estimation in Subscription-Based SaaS Business Models

Accurate estimation of Customer Lifetime Value (CLV) is critical for strategic decision
making, capital allocation, and customer relationship management in subscription-based
Software-as-a-Service (SaaS) business models. Traditional probabilistic models, such as the
Beta-Geometric/NBD (BG/NBD) and Gamma-Gamma frameworks, often struggle to
capture complex non-linear feature interactions, time-varying covariates, and dynamic
churn hazards characteristic of modern cloud enterprise platforms. Conversely, standard
machine learning regressions frequently fail to account for right-censored event data and
dynamic contract renewals. This paper presents a novel hybrid estimation framework that
integrates non-linear predictive architectures, temporal attention networks, and deep
survival analysis with classical probabilistic discounting. Utilizing an empirical B2B SaaS
dataset spanning 2015 to 2021 comprising 142,000 enterprise accounts, our model achieves
superior performance over legacy models, reducing Root Mean Squared Error (RMSE) by
28.4% and Mean Absolute Percentage Error (MAPE) by 19.2% in long-term CLV
projections. We demonstrate how combining high-dimensional behavioral usage telemetry
with contract expansion vectors enhances early churn detection and revenue yield
forecasting, leveraging advanced big data paradigms to drive enterprise business
intelligence. Furthermore, we establish the theoretical and empirical underpinnings of
dynamic cohort valuation under varying macroeconomic discount regimes.