Answers are grounded in this paper and cite their sources.
The study examines automated forecasting and evaluation of digital marketing performance using statistical models from different methodological classes. The paper aims to experimentally compare deterministic and probabilistic models in three groups of digital marketing analytics tasks: advertising channel attribution, media-mix modelling, and customer retention forecasting. To achieve this aim, a unified experimental data-processing pipeline for a SaaS enterprise was developed, evaluation metrics were defined for each analytical task, rolling-origin cross-validation was implemented, and the results were interpreted from a managerial decision-making perspective. The study employs multi-touch attribution methods, including Last-Click, Shapley values, and Markov models; regression and Bayesian structural time-series models for estimating the impact of media spending on revenue; and discrete probabilistic retention models, particularly BdW, for forecasting subscriber activity and customer lifetime value. The results show that algorithmic attribution models reveal the hidden contribution of upper-funnel channels, while still confirming the limited predictive role of attribution models. In media-mix modelling, Bayesian structural time-series models produced lower forecasting error and more stable forecast bias than ordinary least squares regression. In retention modelling, the BdW model reduced the overestimation of the active customer base and the risk of inflated LTV estimates. The scientific value of the paper lies in the formalisation of an experimental comparison of different statistical model classes within a unified analytical pipeline. The practical significance lies in supporting the selection of appropriate models depending on the business scenario, data structure, and uncertainty level.
No takes yet. Share an insight, caveat, or question.
Kyrychenko et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: