Demand Forecasting vs Sales Forecasting

You sold 700 units last month. But you stocked out on day 20, so how many would customers have bought if you hadn't? That missing number is the difference between the two forecasts most people treat as one, and buying off the wrong one quietly teaches your business to repeat its own shortages.

Demand forecasting and sales forecasting aren't the same: demand forecasting estimates unconstrained customer demand (what people would buy if stock never ran out), while sales forecasting estimates constrained sales (what you'll actually sell given limits). Buying off sales alone bakes in your past stockouts.

Key takeaways

• Demand is the want, sales is the outcome: the two numbers only match when nothing constrained the buying, and stockouts constrain it constantly.
• Sales history is censored evidence: a stockout week records low sales even when demand was at its peak, and models can't see the difference unaided.
• Buying off sales perpetuates shortages: each stockout teaches a sales-based forecast to order less of exactly what customers wanted most.
• The availability test tells you when it matters: near-perfect in-stock rates let sales stand in for demand; visible gaps mean you need demand, not sales.

Is demand forecasting the same as sales forecasting?

No. Demand forecasting estimates what customers want; sales forecasting estimates what you'll actually sell. They diverge whenever stock runs out, a size sells through, or a page goes down: any moment when a willing buyer met an unavailable product, demand kept happening while sales stopped being recorded.

Demand = unconstrained want; sales = constrained outcome

Think of sales as demand filtered through your own limitations. Finance is right to plan revenue off the constrained number, since that's the cash that will actually arrive. But inventory exists to remove the constraint, so planning stock off the constrained number is circular: you'd be sizing the shelf to the shortage the shelf created. (The broader discipline of estimating the want is covered in what is demand forecasting.)

Why does the difference matter for buying?

If you buy off sales history, every past stockout teaches the model to under-buy: you'd be planning to repeat the shortage. This is the practical heart of the distinction, and it compounds quietly over cycles.

Stockout-censored demand

Statisticians call it censored data: the true value existed but couldn't be observed past a cutoff. Here's the shape of it on one SKU:

• 1. In stock?: Yes, all month; Recorded sales: 240; Estimated true demand: ~240
• 2. In stock?: Yes; Recorded sales: 260; Estimated true demand: ~260
• 3. In stock?: Stocked out day 20; Recorded sales: 190; Estimated true demand: ~285
• 4. In stock?: Out until day 9; Recorded sales: 130; Estimated true demand: ~270

Read the sales column alone and demand looks like it collapsed in weeks 3 and 4. Read the story and demand was likely the strongest in exactly those weeks; the shelf just stopped recording it. A model fed the sales column sees a fading product and orders less.

How buying off sales perpetuates stockouts

The loop runs like this: a stockout suppresses recorded sales, the suppressed history lowers the forecast, the lower forecast shrinks the next buy, and the smaller buy stocks out sooner. Each turn looks locally sensible, and the SKU spirals from best-seller to "declining product" without customer demand ever actually falling. Best-sellers are the most exposed, because they stock out most often, which means your sales data is most wrong about precisely your most important products. (What each of those unserved weeks costs, in revenue and in customers, is the subject of the real cost of stockouts.)

When can you treat them as the same?

When you almost never stock out, sales closely track demand, and the distinction is academic. The moment availability gaps appear, the two split, and you need demand, not sales, driving the buy.

The availability test

Before trusting sales history as a demand signal, run three checks per SKU:

• In-stock rate over the history window: anything meaningfully below full availability means the history is censored somewhere.
• Stockout timing: gaps during peak weeks distort far more than gaps during quiet ones.
• Variant availability: a product "in stock" with its top size missing is half-censored, and blended numbers hide it.

Where the test fails, correct before forecasting: estimate what stockout periods would have sold (from run-rates before the gap, or comparable weeks) and forecast from that reconstructed demand. It doesn't need to be perfect; even a rough uncensoring beats teaching next season's buy to repeat last season's shortage.

Frequently asked questions

What is censored demand in forecasting?

Censored demand is demand that happened but couldn't be recorded because a constraint cut off the observation, almost always a stockout. Sales data from those periods understates true demand: customers wanted the product and met an empty shelf. Forecasting from censored history without correcting it systematically under-estimates exactly the products that sell hardest.

How do you recover lost sales from stockouts in a forecast?

Reconstruct the gap: take the SKU's run-rate just before the stockout (or matched comparable weeks) and extend it across the out-of-stock days, then use that estimate in place of the recorded low. Signals like page traffic, back-in-stock signups, and search volume during the gap sharpen the estimate. Approximate honestly; don't leave the zero.

Should I plan inventory off demand or sales?

Inventory off demand, revenue off sales. Stock decisions exist to serve what customers want, so they should chase the unconstrained number. Revenue and cash plans should stay on the constrained number, since that's what will actually transact. Using one number for both jobs quietly bakes yesterday's availability problems into tomorrow's plan.

Does sales forecasting underestimate demand?

Whenever availability was imperfect, yes, and always in the same direction: downward, and most severely on best-sellers, which stock out most. That's what makes the error dangerous; it's not random noise but a systematic bias that compounds through reorder cycles if nobody corrects the history.

How do you estimate demand during a stockout?

Start with the SKU's velocity immediately before the gap and carry it through, adjusted for anything unusual (a promo, a season turning). Cross-check with demand signals that kept recording while sales couldn't: product-page sessions, waitlist and back-in-stock signups, marketplace search interest. The blend gives a defensible weekly estimate to patch the history with.

Which forecast do finance teams usually use?

Finance runs on the sales (constrained) forecast, correctly: revenue projections should reflect what will actually transact given real availability. The friction appears when that same constrained number drifts into inventory planning. Mature teams keep both: demand for deciding stock, sales for planning cash, reconciled so each side knows which number it's holding.