Harpy Glossary

Demand Forecasting

Amazon & D2C glossary · Harpy Media

Demand Forecasting is estimating how much of a product customers will buy in a future period — next month, next quarter, next season — and planning inventory to meet that number without drowning in it. It fuses sales history, seasonality, promotions, market trends, and lead times into a reorder plan.

What is Demand Forecasting?

It’s the hinge between two expensive failures: stock out and lose rank (the algorithm punishes empty shelves, and building back costs months), or over-order and pay storage, aged-inventory surcharges, and trapped capital. Every seller forecasts; the difference is whether it’s a spreadsheet guess or an instrumented plan with an error rate they actually track.

What a real forecast is built from

Layers, each measurable: baseline velocity (trailing sales per SKU, cleaned of promo distortions), seasonality (your own year-over-year curves plus category rhythms), promotion effects (event lift estimates with confidence ranges, not wishes), market signals (category trend, competitor stockouts, search volume shifts), and lead times (supplier production plus freight — the number that turns a forecast into an ORDER date). Amazon adds its own view: Seller Central’s inventory tools and vendor-side forecasts show the platform’s expectations — useful context, never gospel, since Amazon forecasts serve Amazon’s network first. The professional habit: forecast in ranges (base/bull/bear), track error monthly, and place orders against the plan, not against last month’s excitement.

The two failure modes, priced honestly

Stockouts cost more than the lost sales: rank decays, ad efficiency drops (you pay to send traffic to “currently unavailable”), and rebuilding takes repeated good weeks. Oversupply costs storage (with Q4 multipliers), aged-inventory surcharges at 180/270/365 days, markdown capital, and working-capital paralysis — the warehouse full of last season is why there’s no money for this season’s winner. The balance point isn’t perfect accuracy (nobody hits it); it’s asymmetric insurance: high-velocity hero SKUs carry safety stock; long-tail carries less; new launches get staged orders with contingencies. Forecast honestly, err toward availability on heroes, and never let yesterday’s optimism set today’s PO.

Forecast units = trailing baseline × seasonal index × promo multiplier · Order point = forecast lead-time demand + safety stockThen track forecast error (MAPE) monthly — a forecast you don’t grade never improves.

In practice

A brand builds a three-scenario forecast for Q4 and discovers its hero SKU’s lead time (62 days door-to-FC) means the Q4 order needed placing in July — two months before their usual September reorder. They place base + a bull tranche against a contingency supplier. Q4 arrives hot; hero stock holds through December; the SKU finishes the year with its rank intact and no storage hangover, while a competitor’s identical product goes unavailable on December 8 and loses its page-one position into January.

⚠️ Watch out. A seller reorders based on last month’s sales, ignoring the promotion distorting them. The deal units read as baseline, the PO doubles, and the “growth” returns as storage fees and a markdown. Forecast the underlying demand, not the anomaly you just paid for.

💡 Harpy tip. Write down your forecast before the month starts, then score it after it ends — both number and reasoning. Six months of scored forecasts beats any planning software you could buy.

How Harpy Media helps

Forecasting with tracked error rates and scenario planning is a standing deliverable on our accounts — inventory is where good demand guesses become real money.