New AI 'Team' Can Answer Supply Chain Questions in Plain English

Supply chain planning is a job that quietly needs four different brains at once. Someone has to pull data out of databases, read the performance charts, forecast next month's demand, and figure out what went wrong when numbers slide. Most companies don't have one person who can do all of that, so questions sit in a queue waiting for a specialist — while shelves sit empty or warehouses sit overstuffed.

This paper proposes splitting the work the way a good office does. A 'coordinator' AI reads the planner's question in everyday language, then delegates to smaller specialist AIs (helper programs that each handle one job well). The planner can ask something loose and exploratory, like 'why did our West Coast orders dip?' or run a fixed, routine report — and move back and forth between the two without switching tools. All the industry know-how lives inside each specialist, so adding a new skill means rewriting instructions, not rebuilding the software.

In a test environment mimicking multi-level inventory management — where products flow from factories to warehouses to stores — the multi-AI setup answered correctly about 90% of the time, matching a single do-everything AI while using roughly four times fewer 'tokens.' Tokens are the word-pieces AI processes, and they're the main thing you pay for. The researchers also showed the system catching a bad inventory plan and automatically improving a demand forecast, which are exactly the tasks that normally eat a planner's afternoon.

The appeal is cost and flexibility: cheaper to run and easier to extend. The honest catch is that this was proven in a simulated test bed, not in a real company with messy data, surprise shipping delays, and suppliers who change terms. Still, it points to a future where the person ordering inventory doesn't need a data science degree to get an answer.