You don't have a forecasting problem.

Every planning function produces a forecast accuracy number, and most produce it monthly. It is the one measure of planning that is easy to compute, easy to compare between periods and easy to put in a pack. It is also, on its own, close to uninformative about whether the plan will hold.

Ask instead what the working week actually looks like. Six things recur across manufacturing businesses of very different sizes and in very different sectors, and not one of them is a model.

None of that appears in a forecast accuracy report. All of it decides whether the plan survives the month.

Put those symptoms beside each other and they stop looking like six separate problems. Each one is a decision that was never assigned, a process that exists on paper but not in the working week, or a handoff between two functions that are measured against each other.

That is a different category of problem from a model that is not accurate enough. A model can be improved with better data, a better method or more compute. None of those things assigns ownership of a number. None of them creates a forum where a trade-off is actually settled. None of them makes a plan and its execution agree.

They compound, which is why fixing one in isolation rarely holds. Undefined ownership makes governance impossible, because there is nobody to hold the cadence. Weak governance makes silos permanent, because there is no forum in which a conflict is resolved rather than escalated. Silos guarantee firefighting, because every trade-off becomes an argument. And firefighting consumes exactly the time that would have prevented the next fire.

Consider what actually happens when accuracy improves. The forecast is closer to what was eventually sold. That is a real gain and it is worth having. But the plan is not the forecast. The plan is what the business commits to make, buy, hold and promise, and it is produced by people applying judgement to the forecast under constraints the forecast does not describe.

So the improvement lands on the input and stops there. If the buffer policy is stale, a better forecast is absorbed by cover sized for a different business. If nobody owns the number, a better forecast is overridden by whoever escalates loudest. If the monthly meeting reports rather than decides, a better forecast changes the slide and not the commitment.

The distance between what was planned and what actually happened is where the working capital and the missed promises live, and it is not made of forecast error. It is made of decisions that were slow, unowned, or never taken at all.

If the problem is organisational, the fix is a structure rather than a tool. Seven capabilities decide whether a planning function works, and they are not independent of each other: each one stands on the one below it.

The order is the argument. Governance without clear ownership is a meeting with no authority. A decision framework without governance is a policy nobody applies. Instrumentation applied to a process nobody follows digitises the dysfunction rather than removing it, and automation applied to a decision nobody owns makes the wrong call faster and more consistently than a person could.

This is why automation sits last, and why last does not mean least. Automation multiplies whatever it is pointed at. That is an argument for building the thing worth multiplying first, not an argument against building it.

The same argument can be read as a position rather than as a list. We call it the WhiteBox SCM Planning Maturity Ladder, and it exists to answer a different question from the one the accuracy report answers: not how good is the forecast, but how much of the structure is actually in place.

A planning function that spends its week expediting has not failed at forecasting. It has not yet built the capability that would make a forecast useful, because a forecast is only useful to somebody in a position to act on it. A function running integrated business planning is not better at statistics than one that is not; it has more of the structure underneath it. The difference between reporting on the business every month and planning it every month is not analytical sophistication. It is whether the meeting has the authority to change what happens next.

Each level contains the ones beneath it, which is the part a maturity model drawn as six bars tends to hide. Nothing is skipped and no level can be bought. Scenario planning laid on top of a process nobody follows produces scenarios nobody acts on, and a business that has not yet made its planning visible cannot make it autonomous, because there is nothing there to automate that anyone would trust.

The useful first question is not which model to use, or which platform to buy. It is which of the seven capabilities is load-bearing for this business right now, because that is the one where effort spent anywhere else gets absorbed.

Finding it does not require a programme. It requires an honest account of how the last bad month actually went: what was decided, by whom, on what information, and how late. That account tends to locate the constraint faster than any amount of analysis of forecast error, because it asks about decisions rather than about numbers.

It can be an uncomfortable answer, because an organisational answer is slower than buying something and harder to put in a business case. It is also the only starting point that does not waste the effort that follows it. And if the conclusion is that the sequencing already in place is the right one, that is a real answer and it is worth having.