Life-cycle model and dynamic prediction method for digital service platform payment obligations

Abstract

This paper addresses the scientific and applied problem of modelling and dynamically predicting the lifecycle of customer payment obligations on digital service platforms. Predicting only a final “paid” or “unpaid” status does not represent intermediate states, their duration, or changes in payment probability following new events. The study aimed to develop a multi-state lifecycle model of a payment obligation and a method for constructing its dynamic predictive profile. A payment obligation was represented as transitions among creation, initial processing, repeated contacts, payment, and completion of the internal cycle without payment. The model incorporates obligation characteristics, observed event history, duration in the current state, position relative to the due date, admissible transitions, and conditional transition intensities. The proposed method constructs a predictive profile containing probabilities of subsequent transitions, payment probabilities over specified horizons, estimated time to payment, and probability of cycle completion without payment. The profile is updated following a new event or changes in temporal characteristics. Right censoring was applied to incomplete observations; therefore, absence of payment by observation end was not treated as cycle completion without payment. Scientific novelty consists in combining an event-based lifecycle representation with regular updating of individual predictions. The experiment used digital platform data containing 19,033 events, from which 15,819 payment obligations were reconstructed. Paid obligations accounted for 19.8 %, while payment probability varied by obligation type, current state, and prediction time. The findings support replacing static final-status classification with dynamic multi-state prediction. The results provide an information basis for adaptive selection of payment-obligation processing strategies.