Back to Home
Industry

Can AI Run a Vending Operation? New Study Raises Doubts

Researchers benchmarked five AI agents running a simulated 20-machine vending network for 180 days; only GPT-6.1 Sol beat conventional inventory software, and experts say human oversight and spending limits remain essential.

V
Vending News Staff
•Nov 14, 2026 · 12:00 AM
Can AI Run a Vending Operation? New Study Raises Doubts

AI-GENERATED INSIGHT — This article was drafted using AI and reviewed by editorial staff. Source: Vending Times

Four of five artificial intelligence agents managing a simulated vending network generated lower financial returns than a conventional inventory management system, according to a new study that raises questions about how much operational control businesses should hand to autonomous AI. The study, "AutoVend: Benchmarking LLM Agents on Long-Horizon Autonomous Vending Operations," was conducted by researchers affiliated with StepFun, Zhejiang University, Xi'an Jiaotong-Liverpool University and other institutions. They evaluated five AI agents managing 20 vending machines across factories, college campuses, residential properties, commercial establishments and offices over 180 simulated days, each starting with roughly $6,000 in operating capital and responsible for product selection, pricing, inventory purchasing and replenishment scheduling. The environment included a shared warehouse, delivery delays, expiration, storage costs and labor, and was calibrated with six months of transaction records from 99 machines operated by an unidentified commercial smart vending company, covering 220,128 products sold from a catalog of 1,372 items. Only GPT-6.1 Sol consistently beat the conventional system, finishing with average net assets of about $46,400 versus $31,600 for the rule-based benchmark, roughly 47% more. The other four models finished below the benchmark, though all five posted positive net asset growth. Some agents minimized waste but left shelves empty; Claude Sonnet 5.5 recorded a 0.16% waste rate but completely empty shelves on more than 25% of operating site-days, while a DeepSeek model scheduled back-to-back campus visits earning about 77 cents in gross profit against a $5.25 service cost. "If a vending machine location has too many unknown variables, the AI will struggle to make decisions on that machine," Manuel Mojica, founder of Vending Circle and owner of MM Healthy Vending, told Vending Times, adding there should be "a break point in the system that forces the human to take action if it reaches a certain price or amount of stock." Chris Selland of Differential Factor said the findings show large language models are "the wrong tool for the day-to-day job of running the operation," and recommended deploying AI where there is more variance and uncertainty. Christopher Graney-Ward of ToonSoft said the leading model's win proves autonomous systems can outperform conventional software "if the setup is right," and recommended purchasing limits, price boundaries and human approvals. Source: Vending Times — https://www.vendingtimes.com/news/can-ai-run-a-vending-operation-new-study-raises-doubts/
#AI#vending operations#inventory management#research#automation

Source: Vending Times

We use cookies and similar technologies on our website to enhance your experience and personalize content and ads. By continuing to use our website/app, you consent to the use of these technologies and the processing of your personal data for personalized and non-personalized advertising. By clicking 'Accept', you consent to the use of cookies and the processing of your data as described above
Powered by Acceptrics