Adaptive User Journeys in Pharma E-Commerce with Reinforcement Learning: Insights from SwipeRx

TL;DR

The paper tackles the need for adaptive, personalized user journeys in healthcare digital tools, especially in LMIC settings. It introduces an end-to-end reinforcement learning platform integrated with SwipeRx, detailing architecture (SDK, backend, algorithmic engine, frontend), predictive modeling, and bandit-based recommendations. Through XP1–XP3 experiments on SwipeRx, it demonstrates significant uplift in basket size and engagement, supported by rigorous impact analyses (t-tests, LMMs, bandit sensitivity analyses) and qualitative interviews. The findings suggest a scalable framework for adaptive interventions that can improve pharmaceutical supply chains, health worker capacity, and patient care across digital health tools.

Abstract

This paper introduces a reinforcement learning (RL) platform that enhances end-to-end user journeys in healthcare digital tools through personalization. We explore a case study with SwipeRx, the most popular all-in-one app for pharmacists in Southeast Asia, demonstrating how the platform can be used to personalize and adapt user experiences. Our RL framework is tested through a series of experiments with product recommendations tailored to each pharmacy based on real-time information on their purchasing history and in-app engagement, showing a significant increase in basket size. By integrating adaptive interventions into existing mobile health solutions and enriching user journeys, our platform offers a scalable solution to improve pharmaceutical supply chain management, health worker capacity building, and clinical decision and patient care, ultimately contributing to better healthcare outcomes.

Figures (24)

• Figure 1: Schematic representation of the RL platform's architecture.
• Figure 2: Screenshots of the SwipeRx application.
• Figure 3: T-sne visualization in the contextual trait space for XP1. The figure shows the best arm for each participant at the end, with the size of the point proportional to its confidence.
• Figure 4: Difference between mean accumulated daily expenditure of users in the adaptive group vs. pure control for XP2. Confidence intervals (90 %) associated to the t-test are shaded. The black horizontal line is drawn across 0.
• Figure 5: Breakdown by interaction (opened, closed or ignored) of messages that lead to purchase of the less frequently bought item in XP2 (22.9% of all messages), including the percentage of each interaction these represent.