Skip to Content

The plan was monthly. The plant was hourly. Agents closed the gap.

Not an AI programme. A scheduling problem that AI happened to be the right tool for.
October 6, 2026 by
The plan was monthly. The plant was hourly. Agents closed the gap.
Al Jawad Bot
Manufacturing · Technical
The brief

A building-materials manufacturer, two plants, one country. Schedule adherence below 70%; expediting costs rising. Time horizon: five months.

The diagnosis


Planning ran monthly against a demand signal that moved weekly, and the plant absorbed the difference through informal expediting. The cost of that absorption had never been measured.

Every deviation escalated to the plant manager because no rule existed for deciding at the edge.

The transformation


We wrote the decision rules with the schedulers first, on paper, then deployed bounded agents to apply them: reschedule within tolerance, escalate outside it, log every decision.

The agents own a named scope and can be replaced in under a week. Every decision they make is readable by an auditor.

The numbers
−64%
Exceptions escalated
+14 pts
Schedule adherence
−620 hrs
Manual handling per quarter
100%
Agent decisions logged

They wrote the rules with my schedulers before they wrote any code. That is why the schedulers trust it.

Chief Operating Officer · Building-materials manufacturer

What we learned


The paper rules were the deliverable. The agents were an implementation detail — and saying so out loud is what got the plant to adopt them.

We would set the replacement test earlier. Knowing an agent can be switched off in a week is what let the client approve it at all.

A hospital group was writing off claims it had already earned.
Rejection rates were treated as a billing problem. The leak was upstream, in admission and coding.