ProductionNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.

AI rewrites the production schedule in minutes when a line slows or an order changes

Added 2026-10-15 · Last updated 2026-10-15

Role

Production planning & scheduling · Typical users: production scheduler, planner, plant manager, operations and continuous-improvement engineer

The problem

The schedule is wrong by mid-morning: a line slows, a machine goes down, a rush order arrives or a material is late, and the scheduler pulls plant managers, engineering and operations into a room to rebuild the plan. The sequence matters too, because the wrong product order adds cleaning and changeover time. In Microsoft's customer story, a major beverage bottler was replanning production schedules 10 to 15 times per week through manual meetings, using averages and operator experience, while the experienced engineers who held the scheduling know-how were approaching retirement.

Step by step

1. Measure the baseline for one line or area: replanning events per week, hours spent per replan, changeover and CIP downtime per week, and schedule adherence.
2. Collect the real data, not the standards: actual line speed per SKU, actual changeover time for each product-to-product pair, ramp-up time, CIP rules, and staffing limits. Real changeover times are often where the gains are.
3. Write down the scheduling rules that live in people's heads (allergen order, color or flavor sequence, minimum run sizes, due-date priorities) in plain sentences and have the senior scheduler confirm them.
4. Pilot an optimizer on one line: give it the data and rules, and compare its sequence with the human schedule for the same week on changeover time and on-time orders.
5. Set a trigger for re-optimization (line down more than X minutes, rush order, material delay) and keep the scheduler as the person who releases the new plan.
6. Use what-if runs before planned events: maintenance stops, a big order, a shift cut.
7. Track the same baseline numbers monthly and expand line by line only when the gains hold.

Tools

• Sight Machine (industrial data platform) with Microsoft OptiMind in Microsoft Foundry: used at the beverage bottler
• Other scheduling tools to evaluate: advanced planning and scheduling (APS) systems such as Siemens Opcenter APS, DELMIA Ortems, PlanetTogether, Asprova, and solvers such as Google OR-Tools or Gurobi for in-house models
• Any LLM (ChatGPT, Claude, Copilot) with the prompt below to write down your scheduling rules and find the changeover data you are missing

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

• The figures are published by Microsoft and Sight Machine about an unnamed bottler and are not independently audited; measure your own replanning hours and changeover downtime first.
• An optimizer is only as good as its data: standard changeover times instead of actual ones give a wrong 'optimal' plan.
• Food-safety, allergen and quality rules are hard constraints: a person confirms they are encoded correctly before any AI schedule is released.
• Keep the scheduler in charge of releasing plans; operators need to know who decided a change and why.
• Capture the senior scheduler's rules in writing now, whatever tool you choose; that knowledge is at risk when they retire.

You are an experienced production scheduler and industrial engineer.

I will paste: (1) this week's orders [order ID, SKU, quantity, due date, priority]; (2) the lines [line, SKUs it can run, rate per SKU as recorded]; (3) our changeover data [from SKU, to SKU, recorded changeover or cleaning time]; (4) our rules [allergen or color sequence, minimum run size, CIP rules, shifts and staffing]; (5) what changed [for example: line 2 down until 14:00, rush order 4512 added].

Do this:
1. Restate our scheduling rules as a numbered list of constraints and ask me about any rule that is ambiguous or missing.
2. List the product-to-product changeovers that appear in this week's orders but have no recorded time in my data.
3. Propose a run sequence per line that respects the rules and puts products with short recorded changeovers next to each other, and mark each choice with the rule or data it is based on.
4. List the orders at risk of missing their due date after the change, and the options (move to another line, split, overtime, ask the customer), for me to decide.
5. Write the questions I should answer before trusting an optimization tool with this line.

Do not compute run times, totals or dates yourself: give the sequence and I will calculate times in our scheduling sheet or tool. Do not release or change any schedule.
Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Sources

• https://www.microsoft.com/en/customers/story/26648-sight-machine-microsoft-foundry