Leveraging Data Analytics to Evaluate Proactive Interventions to Prevent Inventory Defects

Abstract

At an automated fulfillment center typically used in the retail industry, products fallen from a robot-driven shelving pod could cause inventory quality issues and obstructions on the floor, reducing throughput. Leading indicators of fallen products are limited, resulting in a lack of targeted and proactive actions. This project aims to evaluate potential interventions to reduce fallen products based on computer vision signals, accounting for the cost, complexity, and effectiveness of the interventions.

This project developed a framework to perform cost-benefit analyses for the potential interventions that could prevent inventory defects. Characteristics of multiple potential proactive interventions combined with multiple potential vision-based predictive signals form a complex solution space. We start by formulating a common basis of comparison for the options, focusing on how to measure, validate and quantify the effectiveness of the interventions. Experimental data will be derived from a hypothetical pilot that can be used to test hypotheses and evaluate intervention cost and benefit in the context of input signal characteristics and operational complexity. Quantifying the trade-offs and break-even points between use cases ultimately determines the project NPV or ROI hence helping to guide the optimal decision making.

This thesis provides insights into how to leverage analytical tools to evaluate options through the case of preventing inventory defects. This framework could be generalized and applied to any system, be it in logistics or manufacturing, where there are potentially multiple predictive signals and multiple proactive interventions to improve operations.