Forecasting for Inventory: How to Get the Number Right

What inventory forecasting actually is

Inventory forecasting is the calculation that turns a demand prediction into an actual stock number: how many units to hold, and when to trigger the next order. Demand forecasting asks how many units customers will want next month. Inventory forecasting asks a narrower, more useful question: given that demand, your lead time, and how wrong you might be, what do you need sitting in stock right now, and on what date do you place the next purchase order.

The two terms get used interchangeably, which causes real damage. A demand forecast that's off by 10% is a rounding error you absorb. An inventory forecast that's off by 10% is either a stockout that kills rank mid-campaign, or cash sitting in a warehouse for six months. The stock number is where the forecast has to hold up, not just the demand curve underneath it.

The four ways to build the number

Quantitative forecasting uses your own sales history — at least a year, ideally more, so seasonality shows up rather than getting mistaken for a trend. It's the most reliable method available, and it's also useless for a SKU that launched eight weeks ago.

Qualitative forecasting fills that gap: market research, competitor signals, category knowledge, used when there's no sales history to lean on. It's slower and more subjective, but it's the only real option for a new launch.

Trend forecasting looks at the direction of travel — is this category growing or shrinking — and adjusts the quantitative baseline accordingly. Seasonal forecasting does the same for recurring calendar patterns: Q4, back-to-school, a weather-driven spike.

Graphical forecasting is not a fifth method, despite showing up as one in most breakdowns of this topic. It's a way of looking at the other four — a chart that makes a trend or an outlier visible. Useful, but it doesn't generate a stock number on its own.

The math, worked on real numbers

Here's the calculation that actually produces the stock number, with figures instead of variables:

• Average daily sales: 40 units
• Supplier lead time: 30 days
• Lead time demand: 40 × 30 = 1,200 units — what you'll sell before a new order arrives
• Demand on a bad week: 65 units/day (a competitor stockout, a review spike, a promo)
• Safety stock: (65 − 40) × 30 = 750 units — the buffer against that bad week
• Reorder point: 1,200 + 750 = 1,950 units — the stock level that triggers the next PO

That's the formula underneath every forecasting tool on the market, dressed up differently. The number that actually moves is the safety stock multiplier — how far above average you plan for — and that's a category judgment, not a math problem. A commodity item with steady demand needs less padding than a seasonal or trend-driven one. We've run this math across brands inside the $500M+ in managed revenue Full Circle manages across 100+ brands, and the formula never changes; the multiplier is where most forecasts quietly go wrong.

When the forecast is wrong (or the fix isn't)

If the reorder point turns out wrong, check the lead time first — it's the number most forecasts hard-code and never revisit, and a supplier delay of even five days changes the lead time demand by hundreds of units on a fast-mover. Second, check whether the "bad week" figure used for safety stock was actually the worst case, or just the most recent one.

If the setting is already correct — safety stock is sized right, the reorder point is triggering on time, and you're still stocking out — the leak usually isn't in the forecast at all. It's in execution: a PO placed late, a shipment stuck, a warehouse that didn't check it in on time. Fix the process, not the formula.

And if you've tightened the forecast and fixed execution and sales still fell off, check the ad account before you touch inventory again. A paused campaign or a lost buy box will crater sell-through in a way that looks exactly like a demand shift. That's a Dr. PPC problem, not an inventory one.

The mistakes that wreck accuracy — including ones we've made

The most common mistake is forecasting off gross units sold instead of net, ignoring returns, which on some categories run high enough to quietly inflate every reorder point downstream. The second is treating a promotion-driven spike as a new baseline and reordering as if it will repeat.

We've made this one ourselves: building a forecast on a trailing 90-day average that missed a demand spike caused by a competitor's stockout, because the model had no way to see a buy box or BSR move happening in real time. The fix wasn't a better formula — it was adding a live check against buy box and BSR movement, so the forecast gets a manual override before the reorder point fires on stale assumptions, not after.

The third mistake is sizing safety stock once and never touching it again. Lead times drift, suppliers change, categories mature. A forecast built in January on that quarter's lead time is wrong by August if nobody's checked it since.

Where this fits with a managed service

Dr. Stock, from Full Circle, is built around this exact math — reorder timing, safety stock, and the aged-inventory and storage-fee decisions that follow from getting it wrong — plus Orbit's BSR, buy box, price and fee tracking included at no extra cost, so the "bad week" input isn't a guess. Purchasing decisions always go to a human, whatever autonomy level you choose: full approval, supervised, or fully autonomous inside agreed guardrails. There's no published price; it's a demo and a call, with the first 30 days free. Whether or not that's the right fit for you, the formula above is the one to check any tool or spreadsheet against.