A distributor's inventory problem is not "how much stock should I hold". It is "what should I order, from whom, and when, so that the customers I have promised to serve are served, without tying up more working capital than the business can carry". Inventory optimization is the discipline of answering that question with the operator's own data rather than with a rule of thumb.
What it decides
At minimum, an inventory optimization system decides three things for every SKU at every stocking location: the timing of the next purchase order, its quantity, and the source it goes to. In a distribution business that also holds stock at more than one site, it decides transfers between sites. Where a customer has approved a substitute product, it decides whether to serve a shortage with the substitute. Every one of those decisions competes for the same cash and the same supplier and warehouse capacity, which is why they are optimized together rather than SKU by SKU.
How it differs from forecasting
A forecast estimates demand. It is an input, and an uncertain one. Inventory optimization takes the forecast, its uncertainty, the supplier's lead-time and fill behaviour, the customer's service commitment, and the cost of holding, expediting and missing a sale, and turns them into a decision. Two distributors with the same forecast should place different orders if one has a reliable local supplier and the other imports through a single port. Treating a point forecast as truth is the most common way to end up lean at exactly the wrong time.
How it differs from reorder points
Reorder-point systems (order Q when stock falls to R) are inventory optimization in a narrow form: R and Q are chosen once from average demand, average lead time and a service target. The textbook treatment (Silver, Pyke and Thomas) is still the right place to learn the mechanics. What a static reorder point cannot do is respond when the supplier's lead-time distribution widens, when a strategic customer's demand shifts, or when cash is constrained this month. Modern optimization re-solves the whole plan at each review with the latest evidence.
What evidence a distributor should demand
Before changing how a business orders, an operator should insist on a replay: take the last year of purchase orders, receipts, sales and stock counts, reconstruct what the current policy did, and run the candidate policy through the same history under many plausible futures. The comparison that matters is not the average saving but the conservative lower bound and the service achieved on held-out weeks. A candidate that only wins on average, or wins only on the weeks it was tuned on, has not earned the right to change anything.
Two warning signs in any vendor conversation: a promised saving before the data has been seen, and a service level that is quoted without saying which metric (unit fill, line fill, order fill or on-time-in-full) it refers to. These are different numbers and they move differently.
Where it pays and where it does not
Optimization pays most where uncertainty is large and expensive: long import lanes, suppliers that short-ship, customers with hard service commitments and high-margin lines. It pays least on SKUs with short, reliable replenishment and low margin, where a simple rule is close to optimal and the cost of being wrong is small. A good system should say so, SKU by SKU, and stay lean where lean is safe.