E-Commerce Bench: Evaluating LLM Agents on Long-Horizon Autonomous Business Operation

Abstract

E-Commerce Bench evaluates LLM agents over a 365-day autonomous business operation. Agents research markets, negotiate with suppliers, manage inventory and promotions, fulfill orders, handle returns, and maintain cash flow under changing demand and supply shocks. The benchmark evaluates 18 frontier models across seven dimensions.

Publication
arXiv preprint

E-Commerce Bench goes beyond chaining short tasks for more turns. The market evolves over a simulated year, and early choices affect later negotiation leverage, inventory, demand, and solvency.

Xinjie Shen 沈鑫杰
Xinjie Shen 沈鑫杰
PhD Student @ Georgia Tech

I study how to train capable agents and make them safe and reliable in open-ended environments.