Lukáš Kakalejčík, Jozef Bucko, Paulo A. A. Resende
Multichannel Marketing Attribution Using Markov Chains for E-Commerce
Číslo: 2/2021
Periodikum: Statistika
Klíčová slova: Attribution modeling, multichannel attribution, e-commerce, marketing, order of Markov chain
Pro získání musíte mít účet v Citace PRO.
Anotace:
There are plenty of online media that can be used by ecommerce companies in order to drive the revenue.
However, use of this media is usually connected to investment into the selected media. From the company
perspective, it is wise to evaluate the outcome of these investments in order to choose the best media mix
possible. As customers do not usually buy during their first website visit, it is important to monitor their
customer journey and assess the value to particular interactions. The objective of this paper is to analyze
the data of selected companies using the Markov chains. The data about online customer journeys were analyzed.
We found that the Markov model decreases the credit assigned to the channels favored by last-touch heuristic
models and assigns more credit to the channels favored by first-touch or linear heuristic models. By using Markov
order estimator GDL (Global Dependency Level), we also found that 4th and 5th order was the most suitable.
Zobrazit více »
However, use of this media is usually connected to investment into the selected media. From the company
perspective, it is wise to evaluate the outcome of these investments in order to choose the best media mix
possible. As customers do not usually buy during their first website visit, it is important to monitor their
customer journey and assess the value to particular interactions. The objective of this paper is to analyze
the data of selected companies using the Markov chains. The data about online customer journeys were analyzed.
We found that the Markov model decreases the credit assigned to the channels favored by last-touch heuristic
models and assigns more credit to the channels favored by first-touch or linear heuristic models. By using Markov
order estimator GDL (Global Dependency Level), we also found that 4th and 5th order was the most suitable.