A Multivariate Hidden Semi-Markov Model of Customer-Multichannel Engagement
56 Pages Posted: 5 Feb 2020 Last revised: 11 Aug 2022
Date Written: February 13, 2019
Abstract
In multichannel retailing, customers often dynamically vary their use of a firm’s online and physical stores for their search and buy activities. We posit that in a multichannel customer journey, the customer’s motivation to engage with the firm’s channels is latent and dynamically transitions across several states. Using customer level data from a firm with online and offline presence, we measure the dynamic transitions in customer’s motivation for channel engagement with a discrete time multivariate hidden semi-Markovian model (HSMM). We predict the effect of these transitions on customer activities like website visits and online and in-store purchase. An implication of the HSMM is that we can explicitly estimate the state duration properties, e.g., mean length of stay in a state, with any distribution that better fits the underlying data. This flexibility on one hand improves predictive ability of state-dependent customer activities and on the other, provides managers the ability to time interventions based on state duration estimates. In our model comparisons, we find that HSMM with Poisson state duration predicts the online visits and purchases better than HSMM with geometric duration and HMM. Using the proposed HSMM, we uncover four engagement motivation states of varying duration. We demonstrate the model’s ability to assist managers in forecasting and discuss alternative targeting and timing of customer-channel intervention strategies.
Keywords: Multichannel customer behavior; Customer Purchase Journeys; Customer engagement; Duration modeling; hidden semi-Markov models; multivariate HMMs; multichannel marketing.
JEL Classification: C11, C13, C14, C15, C18, C41, C52, C53
Suggested Citation: Suggested Citation