AI-Driven Supply Chain Resilience in Nigerian E-Commerce: A Simulation-Based Framework for Demand Sensing, Inventory Optimization, and Last-Mile Logistics

Publication Details

Publication Date

10/11/2026

Volume / Issue

Vol 1, Issue 6 (2026)

Article No.

081

Journal

British Journal of Contemporary Research

DOI

https://doi.org/10.67693/BJCR-WXES8AZA

Received

30 Oct 2026

Views

31

Downloads

11

Abstract

African e-commerce faces a paradox: rapid growth in digital commerce alongside infrastructure deficits, volatile demand, and fragmented logistics networks. AI is widely promoted as a solution for building supply chain resilience, yet firm-level evidence from African contexts remains scarce. This paper develops and tests a simulation-based framework integrating three AI-driven modules - demand sensing, inventory optimisation, and last-mile logistics - using publicly available, literature-parameterised synthetic datasets for Nigerian retail and e-commerce. Drawing on Dynamic Capabilities Theory and Contingency Theory, the framework shows how AI builds resilience under imperfect data, infrastructure constraints, and demand volatility. Machine-learning demand sensing reduced forecast error by 28.8% relative to a seasonal naive baseline (MAPE 1.80% vs. 2.53%; MASE 0.70 vs. 0.99). AI-driven inventory policies outperformed fixed reorder points in three of four disruption scenarios, reducing stockouts by up to 100% under supplier delay and cutting total cost by 8.9–29.6%; however, under a 40% demand spike, the AI policy underperformed, and the forecast was found to underestimate demand by 28.1%, revealing that forecast-driven seizing fails when the sensing model is not recalibrated to structural demand breaks. Route optimisation reduced simulated delivery distance by 30.0%, 41.3%, and 43.9% across 50-, 100-, and 200-customer benchmarks - a reduction that scales with network size. The paper contributes an integrated, reproducible framework, a contingency-sensitive extension of Dynamic Capabilities Theory, and a methodological template for simulation-based resilience research where primary data access is constrained. An ETA prediction model was attempted but achieved no better than chance on the synthetic data, identifying a boundary of the framework.

License

This article is published under the Creative Commons Attribution 4.0 International License . Free to read, share, and adapt with attribution.