Choosing a Planning Tool for Process Plants in the AI Era

Introduction

You've sat through three planning software demos this quarter. Every one opened with AI on the first slide, showed a forecast with a tidy confidence band, and ended on a Gantt chart that looked great. Afterward, nobody on your team could say whether any of them would have known that one grade can't follow another without a full washout.

The AI era hasn't changed what a process plant needs from planning software. It has made it harder to tell from a demo whether a tool actually has it. The good news is that you don't need to evaluate the AI. You need to test the tool against your own plant, and you can start doing that in the first meeting.

Forecasting and planning answer different questions

A forecast answers how much you'll sell. A plan answers what to run, on which line, in what order, out of which tank, and shipped to whom. Most of what gets sold as AI in planning lives on the forecast side, and it can be genuinely useful there.

But a better forecast fed into a plan that ignores your changeover matrix still produces a schedule the floor can't run. The decisions a process plant cares about are an optimization problem with hard constraints: sequence-dependent changeovers, tank limits, yields that vary by line, shelf life. Solving that takes mathematical optimization, and it should be deterministic, so the same data gives the same plan and every decision traces back to a rule or a cost.

A forecast tells you how much. A plan tells you what to run, where, and in what order. In a process plant the second is an optimization problem with hard constraints, and that's the part to evaluate.

Bring one of your own rules to the demo

Here's the test. Pick a rule that causes real pain on your floor and hand it to the vendor. A changeover that takes an hour going light to dark and a full washout going dark to light. A tank that has to keep a minimum heel. Two products that can never run back to back.

Then ask to see that rule inside the plan, not on a slide about flexibility. Is it a hard constraint the plan won't break, or a setting that produces a warning someone has to catch? Does the plan change when the rule changes? A vendor who can bring it back in a follow-up session, modeled on your data, is showing you something real. A vendor who keeps steering back to the forecast is showing you where their product stops.

Hand the vendor one real rule from your plant and ask to see it as a constraint in the plan. Whether it holds, or turns into a warning or a workaround, tells you more than any feature list.

Ask where the plan stops

The second test is scope. Plenty of tools plan production well and stop there, handing a schedule to a spreadsheet that then has to reconcile it with purchasing, transfers between plants, inventory targets, and customer shipments. Every handoff is a place where the plan stops being optimal and starts being negotiated.

So ask plainly: does one model cover purchasing, production, transfers, inventory, and shipments together, or does each step get planned separately? And then the more revealing question: after go-live, which of our rules will still live in a spreadsheet? Every honest vendor has an answer to that one. The length of the answer is the size of the gap you'll be managing by hand.

Ask whether one model plans the whole chain or hands pieces to spreadsheets, and which of your rules will stay in a spreadsheet after go-live. That answer is the gap you'll be managing by hand.

Toolkit or finished model

The last question is who builds the math. Some vendors sell a toolkit: a solver, a framework, and a user interface, with your team writing and maintaining the constraints. That's a sound choice if you have an operations research group and want to own the model long term.

Most mid-market plants don't, and they shouldn't have to hire one to plan a tank farm. For them the better fit is a finished model, built around their rules and maintained by the people who built it. That's how WonForge works: a custom-built optimization model of your plant and supply chain, hosted and supported for you, starting with a pilot on your own data so the model proves itself before any production commitment.

Toolkits suit companies with their own operations research team. Plants without one are better served by a finished model, built around their rules and maintained by the vendor.

Frequently Asked Questions

Is AI useless for supply chain planning in process manufacturing?

No. Machine learning can be genuinely useful for demand forecasting, spotting anomalies, and cleaning messy data, and a better forecast makes for a better plan. What it doesn't replace is the decision layer: which line, which sequence, which tank, which customer. That part has to respect hard physical rules every time, which is why it's solved with optimization, with forecasts feeding in as inputs.

What should we bring to a planning software demo?

Bring one changeover rule that causes real trouble, one tank or storage constraint, and a recent week that went wrong, along with what the plan said versus what actually ran. Ask the vendor to show those rules working as constraints, ideally in a follow-up session on your own data. How they handle a week you already know the answer to is the fastest way to judge the tool.

Do we need an operations research team to use an optimization-based planning tool?

It depends on the route. With a toolkit, yes, because your team writes and maintains the model. With a finished model, no. Your planners work with the plan, run scenarios, and compare them, while the vendor builds the constraints and changes them when your plant changes, such as a new grade, a new line, or a new tank. The math stays with people who know how it's built.

Conclusion

The AI label on a planning tool tells you almost nothing about whether it can plan a process plant. What tells you is how it handles your rules: whether a directional changeover or a tank heel becomes a real constraint, whether one model covers the whole chain, and who keeps the math current when the plant changes. Test those three things and the right tool usually picks itself. If you want to see your own rules inside the math, Check Your Fit is a short, no-commitment call to find out.

Which Supply Chain Planning Tool is Right for Process Manufacturing in the AI Era?