Demand Forecasting: How to Avoid Both Stockouts and Dead Stock
Every inventory decision is a bet on the future. Order too little and you stock out, losing sales you'll never recover and customers you may never win back. Order too much and you're sitting on dead stock, cash frozen on a shelf, quietly bleeding storage fees and heading for markdown.
Demand forecasting is how you stop guessing and start planning. Done well, it's the single highest-leverage discipline in inventory management, because it prevents the two most expensive mistakes a brand can make, at the same time.
This guide covers what forecasting actually is, why brands get it wrong, the difference between accuracy and bias, and how to build a planning process that keeps you in stock without drowning in inventory.
The key shift: forecasting isn't about predicting the future perfectly. It's about being wrong by less, and in ways you plan for. The goal isn't a perfect number, it's a number you can build a resilient inventory plan around.
Jump to what matters most
The two failures forecasting has to prevent
Inventory planning is a balancing act between two expensive failures, and the scale of getting it wrong is staggering. Inventory distortion, the combined cost of stockouts and overstocks, cost global retail $1.77 trillion in 2025: $1.2 trillion from stockouts and $572 billion from overstocks. Even after decades of technology investment, retailers still lose about 6.5% of sales to these two problems.
Failure 1: The stockout
- Lost sales you can't recover
- Customers who buy from a competitor instead
- On marketplaces, suppressed listings and lost ranking
- Wasted ad spend driving traffic to an out-of-stock item
Failure 2: The overstock / dead stock
- Cash frozen in inventory you can't sell
- Storage fees piling up month after month
- Eventual markdowns that erase the margin
- Write-offs on product that never moves
These two failures are two sides of one coin: demand you didn't predict correctly. Forecasting is the discipline that keeps you off both rocks, and it starts from accurate inventory data, the foundation covered in the fulfillment KPIs that predict problems.
Why forecasting is harder than it looks
"Just look at last year" feels like a forecast, but it isn't one, and relying on it is why so many brands stay stuck at roughly 83% inventory accuracy when world-class operations hit 95%. Real demand is pushed around by forces a naive average ignores.
- Seasonality and holiday peaks that shift year to year
- Promotions and price changes, yours and competitors'
- Marketing spend and viral moments that spike demand
- New channels, new markets, and new SKUs with no history
- Supply-side lead times that force you to commit far ahead of demand
- Trend shifts that make last year's winner this year's dead stock
Forecasting doesn't have to be perfect to pay off, modern approaches that blend history with real signals commonly reach 75–90% accuracy, and every point of accuracy pulls you off both the stockout and the overstock rocks.
Accuracy vs. bias: the distinction that matters most
Here's the nuance that separates real planners from spreadsheet jockeys: a forecast can be reasonably accurate and still quietly destroy your inventory, if it's biased. Accuracy and bias are two different things, and you have to watch both.
| Accuracy | Bias | |
|---|---|---|
| What it measures | How close the forecast is, in either direction | Whether you're consistently over or under |
| Common metric | MAPE (mean absolute % error) | Forecast bias %, aim within ±5% |
| What it misses | Can look fine while being persistently skewed | Ignores the size of individual misses |
| What it causes if bad | Volatile, unpredictable inventory | Steady drift into overstock OR stockouts |
What a good forecast actually uses
A useful forecast blends several signals rather than leaning on any single one. The more of these you feed it, and keep current, the better it holds up.
The core inputs
- Sales history — by SKU, channel, and season
- Trend & seasonality — direction and timing, not just averages
- Promo & marketing calendar — planned demand spikes
- Lead times — how far ahead you must commit
- Channel mix — DTC vs marketplace vs retail velocity
The judgment layer
- New-product launches with no history
- Known competitor moves or supply disruptions
- Business goals (a push into a category or market)
- Sanity checks against capacity and cash
- Do we forecast at the SKU-and-channel level, or just in aggregate?
- How often do we revisit and adjust the forecast?
- Are we measuring both accuracy and bias, by SKU?
- Does the forecast actually drive purchasing, or sit in a file?
The forecast is only as good as the inventory data underneath it. If your counts are wrong, your forecast is built on sand, which is why single-source-of-truth inventory across every channel you sell on is a forecasting prerequisite, not a separate project.
Safety stock: planning for being wrong
No forecast is perfect, so the smartest planners build in a buffer, safety stock, sized to the risk of running out. Safety stock is how you absorb forecast error without stocking out, and doing it deliberately is far better than the two common defaults: no buffer (frequent stockouts) or a blanket buffer on everything (expensive dead stock).
The goal is to hold buffer where it prevents lost sales and cut it where it just ties up cash, which is the same per-SKU thinking behind calculating true landed cost per unit. Not every SKU deserves the same treatment.
Stockout vs. overstock cost calculator
Both errors cost money, but rarely the same amount. Adjust the inputs to see which failure is more expensive for your product, and therefore which way to lean.
If you under-order and stock out: $9,000 in lost margin
If you over-order and hold dead stock: $3,500 in carrying + markdown cost
For this SKU, stocking out costs more, so lean toward a bigger safety buffer here. The right buffer isn't the same for every product.
Forecasting readiness scorecard
Check the statements that are true for your operation today. This is a fast read on whether your planning is preventing the stockout/overstock whipsaw, or feeding it.
Tip: start checking boxes to see guidance.
Final insight: you'll never forecast perfectly, and you don't need to. You need to be wrong by less, know which direction you're biased, and buffer deliberately where it matters. That's the difference between the brands that whipsaw between stockouts and dead stock and the ones that just stay in stock.
Nautical gives planning the foundation it needs, accurate real-time inventory, clean sales data across channels, and the fulfillment flexibility to act on the forecast, so your demand plan becomes decisions, not just a file nobody trusts.
FAQ: demand forecasting & inventory planning
What is demand forecasting?
Demand forecasting is the process of predicting how much of each product you'll sell over a future period, so you can buy and stock the right amount. Done well, it prevents both stockouts (lost sales) and overstocks (dead stock), the two failures that cost global retail $1.77 trillion in 2025.
Why isn't "last year's sales" a good enough forecast?
Because real demand is moved by seasonality shifts, promotions, marketing, new channels and SKUs, competitor moves, and trends, none of which a simple historical average captures. Relying on last year is a big reason many brands stay near 83% inventory accuracy when world-class operations reach 95%.
What's the difference between forecast accuracy and bias?
Accuracy is how close your forecast is in either direction (often measured with MAPE). Bias is whether you're consistently over or under. Bias is more dangerous because it's invisible: a small error that's always in the same direction compounds cycle after cycle into overstock or stockouts. A common target is keeping bias within ±5%.
What is safety stock and how much should I hold?
Safety stock is a buffer that absorbs forecast error so you don't stock out. It should be sized per SKU based on how variable that item's demand is and how long its lead time is, not applied as a flat "two weeks of everything" rule, which over-buffers safe items and under-buffers risky ones.
How accurate can demand forecasting realistically get?
Modern approaches that blend sales history with real demand signals commonly reach 75–90% accuracy. Perfection isn't the goal, being wrong by less, knowing your bias, and buffering deliberately is what keeps you off both the stockout and overstock rocks.
