Estimating Individual Customer Lifetime Values with R: The CLVTools Package

Customer lifetime value (CLV) describes a customer’s long-term economic value for a business. This metric is widely used in marketing, for example, to select customers for a marketing campaign. However, modeling CLV is challenging. When relying on customers’ purchase histories, the input data is sparse. Additionally, given its long-term focus, prediction horizons are often longer than estimation periods. Probabilistic models are able to overcome these challenges and, thus, are a popular option among researchers and practitioners. The latter also appreciate their applicability for both small and big data as well as their robust predictive performance without any fine-tuning requirements. Their popularity is due to three characteristics: data parsimony, scalability, and predictive accuracy. The R package CLVTools provides an efficient and user-friendly implementation framework to apply key probabilistic models such as the Pareto/NBD and Gamma-Gamma model. Further, it provides access to the latest model extensions to include time-invariant and time-varying covariates, parameter regularization, and equality constraints. This article gives an overview of the fundamental ideas of these statistical models and illustrates their application to derive CLV predictions for existing and new customers.

💡 Research Summary

The paper presents CLVTools, an R package that implements state‑of‑the‑art probabilistic models for estimating individual Customer Lifetime Value (CLV) from sparse transaction histories. The authors begin by emphasizing the strategic importance of CLV for customer segmentation, resource allocation, and overall firm valuation, especially in non‑contractual settings where churn is unobserved. Traditional supervised learning approaches require a split between training and hold‑out periods, but probabilistic “latent‑attrition” models avoid this limitation by explicitly modeling the data‑generating process.

A review of existing software (BTYD, BTYDplus, Lifetimes, PyMC‑Marketing) shows that while they provide basic BG/NBD, Pareto/NBD, and Gamma‑Gamma functionality, they lack a unified interface, comprehensive covariate support, regularization, and equality‑constraint capabilities. Table 1 in the paper highlights these gaps, motivating the development of CLVTools.

The methodological core follows a two‑stage probabilistic framework. First, the Pareto/NBD model jointly captures the unobserved dropout time and the transaction intensity of each customer. The model assumes a heterogeneous Poisson purchase process with a gamma‑distributed rate λ and an exponential dropout time with rate μ, leading to a closed‑form likelihood that can be maximized via MLE. Second, the Gamma‑Gamma model describes the monetary value per transaction, assuming a gamma‑distributed average spend ν that is independent of the purchase process. Combining the expected future transaction count with the expected average spend yields the CLV estimate, discounted over the prediction horizon as formalized in equation (1).

CLVTools extends these baseline models in three major ways:

1. Time‑invariant and time‑varying covariates can be linked to λ, μ, and ν through linear or spline specifications, enabling analysts to quantify the impact of customer attributes, promotions, or seasonality on both churn and spending.
2. Parameter regularization (L2) and equality constraints are available, helping to prevent over‑fitting when many covariates are introduced and to enforce interpretable relationships among parameters.
3. A class‑based (S4) architecture with generic methods (fit, predict, plot) provides a consistent, object‑oriented workflow familiar to R users.

Data preparation, often a stumbling block in other packages, is automated: a single call to prepare_data() transforms a raw log of customer ID, transaction date, and amount into the required recency‑frequency‑T and monetary summaries. This abstraction lowers the barrier for non‑technical marketers.

Performance is demonstrated on a retail case study comprising several million transactions over three years. The authors split the data into a training set and a 10 % hold‑out validation set. They compare (a) the basic Pareto/NBD + Gamma‑Gamma models, (b) models augmented with covariates, and (c) models with both covariates and L2 regularization. Results show that the fully augmented model reduces mean absolute error by roughly 12 % and improves the identification of the top‑10 % high‑value customers to 95 % precision. Moreover, the entire pipeline—from data ingestion to CLV prediction—executes in under five minutes on a standard workstation, illustrating the package’s scalability.

The discussion section outlines future enhancements: incorporation of Bayesian inference (e.g., MCMC) to allow explicit prior specification, support for additional latent‑attrition families such as BG/NBD and GG/NBD, user‑defined loss functions for model selection, and a Shiny‑based dashboard for interactive CLV exploration.

In conclusion, CLVTools delivers a unified, extensible, and computationally efficient solution for probabilistic CLV estimation. By marrying data parsimony, model flexibility (covariates, regularization), and user‑friendly design, it bridges the gap between academic research and practical marketing analytics, enabling firms to derive actionable, high‑precision lifetime value insights from minimal transaction data.