Reducing shopping cart abandonment and increasing purchase conversion rates.
A paper by Amy Wenxuan Ding, Shibo Li, and Patrali Chatterjee (2015).
Online retailers are plagued by two common problems, low conversion and high shopping cart abandonment rates. Key to improving both, is an increased understanding of a website user’s intent. When the user’s intent is known, a website can offer content that is optimized for that particular user. In doing so the website will serve that individual user, and thus all users, better. Users that are served better lead to improved conversion and shopping cart abandonment rates. However, user intent isn’t known by a website a priori thus making the generation of intent-based webpages impossible.
“if I have 3 million customers on the Web, I should have 3 million stores on the Web”
Jeff Bezos, CEO of Amazon.com
Ding et al. demonstrate a model that can determine individual user intent in real-time, enabling a website to offer webpages that are optimized for the individual user’s intent. The model does so by analyzing user navigation behavior as shown in shopping cart decisions combined with click-stream data. To the best of my and the author’s knowledge, this paper is the first to propose a model that can learn a user’s unobserved intent and use it to increase website performance.
Ding et al. conducted research on data from Barnes and Noble’s website, BN.com as well as a lab experiment. In both cases users were classified as either low- or high purchase intent and several intent based interventions followed as soon as a user was classified.
The system proposed by the paper significantly reduces shopping cart abandonment rates by 32.4% and improves purchase conversion rates by 6.9%.
A practical recommendation from the simulation lab experiment is that online retailers should identify intent as early as possible to maximize the effects. However, a fundamental trade-off exists between the time that the model has for determining user intent and the timing of the intervention using intent-based webpages. The effects are strongest if they happen as early as possible, yet the accuracy of the intent prediction increases with each click. The paper proposes no standard optimum balance. In the case of BN.com the optimal balance was after three page views.
Similarly to the balance between accuracy and timing, the paper offers no universally effective interventions. The following table offers examples of intent-based interventions that did increase performance on BN.com
|Low Purchase Intent Shoppers||High Purchase Intent Shoppers|
|Banner ads||Banner ads|
|Email solicitations||Optimal combination of hypertext links and pictures|
|Optimal number of hypertext links|
As the improved conversion and shopping cart abandonment rates represent large monetary gains, it is advisable from a managerial point of view to implement similar systems in all web stores. However, as each website is different both the balance between accuracy and timing and the content of intent-based interventions will have to be determined on a case-by-case basis.
Learning User Real-Time Intent for Optimal Dynamic Web Page Transformation – Amy Wenxuan Ding, Shibo Li, and Patrali Chatterjee (2015)