A practical guide to demand forecasting, production planning and order promising, before you evaluate any tool.
Most manufacturers don't have a forecasting problem. They have a data problem that shows up as a forecasting problem. Before you compare AI planning tools, it helps to know what those tools actually do and what they need from you.
Forecasting, planning and promising are three different jobs
Vendor demos tend to blend these together. Pulling them apart makes every evaluation easier.
- Forecasting answers "what will we sell?" This is where machine learning earns its place. A model learns from your posted sales history and predicts demand by item and period.
- Planning answers "what do we make or buy, and when?" MRP and master production scheduling are rules-based. They apply your BOMs, routings, lead times and reorder policies to the forecast and open orders.
- Promising answers "what date can we commit to this customer?" Available to promise checks stock and scheduled supply. Capable to promise works out when new supply could cover a shortfall. If an order needs 10 and 6 are on hand or scheduled, capable to promise looks at when the other 4 can be made or bought.
Rules-based planning is not a weakness. It's predictable and you can audit it. The AI part is usually the forecast that feeds it and the assistants that sit on top of it.
Why forecasting matters more in a volatile year
A forecast is only useful if it keeps up with reality. In the , 70.6% of manufacturers named trade uncertainty as their top business challenge. When input costs and lead times move, a forecast you refresh once a quarter falls behind fast.
That's the practical case for AI forecasting. It's not about predicting the future perfectly. It's about updating the picture faster than a planner can by hand.
Readiness, not budget, is the real gap
Almost every manufacturer is looking at this. Research has shown nearly 98% are exploring AI, but only 20% feel fully prepared. In our experience, the gap comes down to four things:
- Clean master data. Items, BOMs, routings and vendor lead times you trust. AI repeats your errors at speed.
- History in the ERP. Posted sales history the model can learn from, not a spreadsheet kept on the side.
- Shop floor data in the system. Output, consumption and capacity recorded when they happen.
- Planning parameters set up on purpose. Reorder policies, safety stock and lead times that match how you really run.
None of these are exotic. They're the same habits that make MRP trustworthy, which is why a readiness check tends to pay off before any AI is switched on.
It also takes pressure off your planners
Planning talent is hard to find. The found 46% of manufacturers struggle to fill planning and scheduling roles. A system that handles the first draft of the forecast gives your planners time for the judgment calls only they can make.
Five questions to ask any ERP vendor about AI planning
- Is this feature generally available, in preview, or something we'd build?
- How much sales history does the forecast need, and at what level (item, location, variant)?
- Which parts of planning are AI, and which are rules-based?
- Does a person review what an agent drafts before it reaches a customer?
- Can we build our own agents on our own data, or are we locked into what ships?
If a vendor can't answer the first question clearly, treat the rest of the demo with care. Preview features can be useful. You just want to know when you're signing up for one.
Seeing it in one system
Answers to those questions vary by product. For a worked example, our team at Western Computer is walking through them in Dynamics 365 Business Central on Thursday, October 8, in a 30-minute webinar: . We've spent nearly 40 years and 1,750+ implementations helping manufacturers plan on Microsoft ERP, and we'll be straight about what's ready and what isn't.
