Reactive inventory decisions — It is rarely a lack of data. It is that the data lives in four systems that were never designed to answer a question together.
Reactive inventory decisions: bu ne anlama geliyor
Ask a plant manager why a material ran short and you will usually get a precise answer. Ask them the morning before, and you will usually get a shrug.
The information needed to see it coming existed. It was simply distributed across an ERP, a production system, a supplier portal and someone’s inbox.
The three questions that are hard to answer together
What will demand be? What do we actually hold, including in transit? And what can the line realistically produce in that window?
Each system answers one of these well. None answers all three, and the join between them is done by a human under time pressure.
What good looks like
A decision platform holds the join. Demand signals, stock positions and capacity constraints are read together, the risk is surfaced before it becomes a shortage, and the recommended action arrives with the numbers that justify it.
What “reactively” actually looks like
Not carelessness. A planner watching a coverage report, noticing a line is short, and raising an order — competently, quickly, and after the shortage already exists.
The reaction is to a state rather than to a signal that predicted it. By the time coverage is visibly low, the lead time has already started running, and the expediting that follows is the cost of having found out late rather than of having decided wrongly.
The three inputs that would have made it proactive
Demand at the level the decision is made, not the level the report aggregates to. A forecast accurate at category-month says nothing useful about one component in one week.
Lead time as a distribution. The contractual number describes an agreement; the receipt history describes what happens, and the gap between them is where the shortage was born.
The cost asymmetry, written down. What a line stop costs against what the extra stock costs — a ratio that is decided implicitly a hundred times a year and stated almost nowhere.
Why it persists
Because reacting works, most of the time, and the times it does not are absorbed as separate operational problems: freight, scrap, an unhappy customer. None of them are attributed back to the decision that produced them, so the decision never looks like the thing to fix.
An organisation that starts recording the decision — what was ordered, against what forecast, under what lead time assumption — can make that attribution within a quarter. Most find one or two suppliers and one or two components account for a surprising share of it.
The first change worth making
Not a model. The review cycle.
A weekly review on a component with a four-day lead time is holding safety stock to cover a delay the process itself creates. Shortening the cycle costs nothing but scheduling, and it frequently produces a larger improvement than any forecasting work would.
What to measure afterwards
Service level against average inventory held, by supplier. Both numbers already exist in most systems and almost nobody plots them together.
If inventory falls while service holds, the decision improved. If service rises while inventory holds, the reorder points were wrong rather than the forecast. Either finding is progress, and the pair distinguishes them — which a single availability figure cannot.
The organisational part
None of this survives if the planner is measured on stockouts alone. A person accountable for one side of an asymmetry will manage that side, correctly and expensively, and no model changes that incentive.
Measuring both sides is the change that makes the rest usable.
Where to start if none of this is recorded today
One decision type, one quarter. Replenishment on a single category is enough: what was ordered, against what forecast, under what lead time assumption, and what happened afterwards.
Three months of that produces something most operations have never had — a record showing which shortages were forecast failures, which were lead time failures, and which were review-cycle failures. Those three call for entirely different fixes, and without the record they are indistinguishable.