Demand Forecasting for Inventory Control, Worked on Real Numbers

Why These Two Things Are Really One Thing

A forecast is a prediction. Inventory control is the set of rules that turns that prediction into an action: how many units to hold as a buffer, and when to place the next order. Treated separately, both fail. A demand forecast that sits in a spreadsheet and never feeds a reorder point is an academic exercise. A reorder point set once and never touched by fresh demand data drifts wrong the moment the season changes.

The book most people find when they search this phrase — Thomopoulos's Demand Forecasting for Inventory Control — treats them as one discipline for exactly this reason. It's a solid academic reference: regression, moving averages, smoothing, safety stock under normal and truncated-normal distributions. What it doesn't do, and doesn't claim to, is tell you what to do on a Tuesday when your actual sell-through diverges from the curve. That gap is where most inventory problems actually live.

The Formula: From Forecast to Reorder Point

Two numbers matter. Safety stock covers the uncertainty in demand during lead time. Reorder point is the stock level that triggers the next order.

• Safety stock = z × σ × √(lead time)
• Reorder point = (average demand × lead time) + safety stock

Worked example: a SKU sells an average of 120 units a week, with a standard deviation of 30 units a week (some weeks run hot, some cold — this is normal, not a red flag). Lead time from your supplier is 4 weeks. You want a 95% service level, which corresponds to a z-score of 1.65.

Safety stock = 1.65 × 30 × √4 = 1.65 × 30 × 2 = 99 units. Average demand over the lead time = 120 × 4 = 480 units. Reorder point = 480 + 99 = 579 units. When stock on hand plus stock on order drops to 579, you place the next order. That's the whole mechanism — everything else is refinement around getting the average and the standard deviation right for the demand pattern you actually have.

From Forecast to Purchase Order: The Stages

Every method — regression, exponential smoothing, a two-stage forecast for a promo — feeds the same pipeline. The output of one stage is the input to the next, and an error early on compounds by the time it reaches a purchase order.

The Mistakes That Break This, Including Ones We've Made

The most common error isn't a bad formula, it's applying one formula everywhere. A fast-turning SKU with steady weekly sales fits the normal-distribution safety stock math cleanly. A slow, lumpy SKU that sells three units one week and zero for the next four does not — the standard deviation assumption breaks down, and the formula will oversize the buffer for a product that doesn't need it. Across more than $500M in managed revenue and 100+ brands, this is the single most common forecasting mistake we see, and we've made it ourselves: applying one global safety-stock factor across a whole catalog and getting it visibly wrong on the slow movers, where a count-based method fits the demand pattern better than a normal-distribution one.

Other recurring errors: forecasting at the aggregate level and ordering at the SKU level (the totals can look fine while individual SKUs stock out); never re-forecasting after a promotion, so the post-promo demand spike gets baked into the baseline as if it's permanent; and confusing forecast accuracy with forecast bias — a forecast can be wrong by the same amount every week (fixable with one adjustment) or wrong randomly in both directions (a different problem entirely).

When the Number Is Wrong: What to Check Next

If the reorder point isn't preventing stockouts or isn't freeing up cash, don't start by changing the formula. Start by checking, in order: has lead time actually changed since the safety stock was set (carriers and suppliers drift); is the forecast error systemic (the same direction every period, meaning bias) or random (meaning noise you can't fully eliminate); and has the demand pattern itself shifted — a seasonal product like a gardening kit or a gifting item behaves nothing like a year-round staple, and the same z-score and lead-time math shouldn't be applied to both without adjustment.

If the inventory settings are correct and sales are still falling, the leak usually isn't in the warehouse at all — it's in the ad account. A perfectly timed reorder doesn't help if bids collapsed, a listing lost the buy box, or a campaign got throttled by budget. That's a different diagnosis and a different fix; that's what Dr. PPC looks at.

Where the Textbook Stops and the Live Account Starts

If you're looking for the demand forecasting and inventory control PDF because you want the formulas, the Thomopoulos book (and others like it — Vandeput, Siemsen, Waters cover similar ground) will get you the math: safety stock under different distributions, forecast error measurement, top-down versus bottom-up methods. That's genuinely useful groundwork and worth having.

What none of those books do is watch your actual Amazon sell-through daily, catch when a lead time silently changed, or flag that a SKU crossed into aged-inventory territory before the surcharge hits. That's an operational job, not a reading one — and it's the gap between knowing the formula and having someone check it against your numbers every week.

Dr. Stock, from Full Circle, is built for that operational half: watching reorder timing against actual demand, catching stockouts before they hit rank, and flagging cash trapped in slow SKUs — with purchasing decisions always going to a human, regardless of how much autonomy the account runs on elsewhere. It doesn't replace the formula. It's the thing that keeps checking it against reality.