Peak Season Inventory Forecasting: How to Plan Before Demand Spikes

Peak season compresses a lot of demand into a very short window. It also compresses the time available to fix mistakes.

By the time a retailer knows which promotion worked, which color took off or which stores are running through inventory faster than expected, a large part of the inventory commitment has already been made. Purchase orders are placed. Receipts are scheduled. In some categories, the realistic opportunity to chase more stock has passed.

That is why peak forecasting is as much an inventory risk problem as a demand forecasting problem.

Using last year's sales and adding a seasonal uplift is not enough. The peak calendar itself is shifting. NRF says roughly two in five holiday shoppers have been browsing and buying before November in recent years, with retailers responding by stocking earlier and running sales events as early as October.

So the useful question isn't only, "How much will we sell?"

It is also: "What happens to the business if we're 15% or 20% wrong?"

That answer should change how much inventory you buy, where you put it, how much safety stock you carry and how much flexibility you preserve for later.

Start With a Demand Baseline You Can Actually Trust

Last year's sales are an input. They are not a forecast.

Historical POS data contains all sorts of decisions and constraints that can masquerade as demand. Peak season makes this particularly dangerous because you're taking those historical signals and using them to justify much larger inventory positions.

Suppose a jacket sold 2,000 units during last year's Black Friday period. On paper, 2,000 looks like the starting point for this year's plan.

But what if the black colorway stocked out halfway through the event? What if medium and large disappeared even earlier? The business didn't necessarily have demand for 2,000 jackets. It had recorded sales of 2,000 jackets before availability started constraining what customers could buy.

The opposite problem happens too. A SKU can look like a peak-season winner because it spent most of the event at 40% off. If this year's merchandise plan assumes full-price selling for longer, blindly carrying that volume forward creates an obvious overstock risk.

Before building a peak forecast, clean the history for the things that distorted it: stockouts, unusual promotions, price changes, assortment changes, channel shifts, store openings and closures, and one-time events.

Separate sales history from unconstrained demand

This matters even more at size level.

A fashion style can technically remain "in stock" while the size break is destroyed. If XS and XXL are still sitting on the shelf but M and L sold out three days ago, observed sales will start slowing. That slowdown doesn't necessarily mean customers stopped wanting the style.

It means you stopped having the units they wanted.

The baseline also needs to reflect where the business is going. If ecommerce is growing while store traffic is declining, or if you've reduced the door count in a category, last year's demand curve needs context.

Start at a planning level where the historical signal is credible. Get to a sensible business-as-usual expectation first. Then add the peak effects.

The objective isn't a mathematically perfect baseline. It is one you trust enough to distinguish underlying demand from promotion, seasonality and event-driven demand.

Build Peak Demand From Drivers, Not One Seasonal Multiplier

"Normal demand × 3" is convenient. It is also a good way to hide bad assumptions.

A stronger peak forecast makes the components visible:

Baseline demand + seasonal uplift + promotional lift + calendar effects + known business changes.

That structure matters because planners can challenge it.

If a forecast jumps 70% over baseline, you should be able to see why. Is the increase coming from holiday seasonality? A planned 30% promotion? More marketing spend? Additional doors? A product launch? An earlier event?

Promotions deserve particular scrutiny.

Last year's promotion may have generated a large sales spike, but that doesn't mean every incremental unit represented incremental demand for the business. Customers may have pulled purchases forward. One promoted SKU may have cannibalized another. A traffic-driving offer may have created halo sales elsewhere.

McKinsey notes that promotion analysis needs to account for stock-up, cannibalization and halo effects. Its work with leading grocers found that when those effects are considered, 10% to 15% of promotions can actually dilute sales and margins.

So if last year's hero fleece sold exceptionally well at 30% off, don't automatically use that event as evidence for a substantially larger full-price buy this year. Ask what created the volume.

Account for a peak calendar that keeps moving

Retailers also need to stop treating Black Friday or another traditional event date as the beginning of peak.

NRF reports that approximately two in five holiday shoppers have started before November in recent years. Retailers have responded by putting inventory in place earlier and launching October events.

This creates a forecasting problem that simple seasonal curves handle badly: demand can move rather than grow.

A strong October event might pull purchases forward from November. If the forecast treats October's uplift as purely incremental and still expects the old November curve, the season gets overplanned.

Map the actual commercial calendar. Promotions. Holidays. Paydays where relevant. Email and paid-media pushes. Product drops. Marketplace events. Store events. Channel-specific campaigns.

Then ask what each one should do to demand and when.

The forecast becomes much easier to operate when the demand spike has an explanation rather than disappearing inside an uplift percentage.

Turn the Forecast Into Inventory Commitments Before the Window Closes

A forecast that never changes an inventory decision is mostly an analytical exercise.

Once expected demand is established, the planner has to translate it into initial buys, safety stock, receipt timing and replenishment capacity.

This is where forecast risk becomes asymmetric.

Being 20% over forecast on an evergreen bestseller isn't necessarily disastrous. The inventory may carry into January and continue selling at full margin.

Being 20% over on a Christmas-specific fashion item is different. Once the selling window closes, those units become markdown exposure very quickly.

Likewise, underforecasting a long-lead-time hero product can be far more expensive than underforecasting a basic that can be replenished in two weeks.

So don't apply the same safety-stock logic across everything.

For long-lead-time, high-margin products with dependable demand and reasonable post-peak sellability, protecting the upside may make sense. For highly seasonal fashion with poor exit options, the better decision may be a tighter initial commitment even if that means accepting some stockout risk.

This is also where open-to-buy matters. Buying everything against the first forecast may improve theoretical availability, but it removes your ability to respond when the forecast inevitably gets something wrong.

Use downside, base and upside scenarios

A single forecast creates false confidence.

Build at least three views.

The base case drives the most likely initial inventory requirement. The upside case tells you where you need supplier capacity, replenishment options or extra raw material protected. The downside case shows where excess stock becomes uncomfortable.

Then overlay lead times.

There is a last meaningful replenishment date for every seasonal product. It isn't the date the supplier can technically ship another unit. It is the date another receipt can still arrive with enough selling time left to justify owning it.

Ordering more Christmas inventory on December 20 isn't much of a chase strategy if it lands after the customer has moved on.

BCG's holiday retail work similarly argues for planning early while preserving the ability to pivot as actual results emerge.

The forecast should ultimately answer three questions: How much should we own? When should we receive it? How much flexibility should we keep?

Forecast Where Demand Will Happen, Not Just How Much Demand Exists

You can have enough inventory across the chain and still lose sales every day.

A category forecast might correctly call for 50,000 units. That doesn't help much if the wrong colors are sitting in slow stores while ecommerce is stocked out.

Peak planning eventually has to move down the hierarchy:

Category → style → SKU → size/color → channel/location.

Not every decision needs to start at SKU-store-day level. In fact, forcing extreme granularity into every forecast often creates more noise than useful information.

But inventory eventually needs to be positioned where demand is expected to happen.

Fashion exposes this problem clearly. A style might show six WOS at total level and look perfectly healthy. Open the size curve and the picture changes. Core sizes could be approaching stockout while fringe sizes account for most of those six weeks.

The same thing happens geographically. Store A can be carrying eight WOS while Store B has one. The chain average looks fine. The customer experience does not.

McKinsey's retail analytics work specifically points to store-specific SKU selection as a way to move beyond broad assumptions about what every location should carry.

Initial allocation should reflect confidence. Proven high-volume doors can take more depth. Less predictable locations may need a shallower opening position, supported by DC inventory or replenishment.

And if the operating model allows it, transfers should be part of the plan rather than an emergency measure discovered halfway through peak.

A useful rule is simple: forecast at the level where a different forecast would cause you to make a different inventory decision.

Below that point, granularity can become workload rather than insight.

Reforecast During Peak and Manage Exceptions Before They Become Markdowns

The pre-season forecast doesn't become sacred because the inventory arrived.

Once peak starts, actual selling should begin replacing assumptions.

That doesn't mean planners need to stare at thousands of SKU rows every morning. The job is to find the exceptions that require a decision.

The operating view should be fairly tight: forecast versus actual sales, sell-through, WOS, stockout risk, core-size availability, location and channel imbalance, promotional performance, and projected end-of-season inventory.

From there, the actions are familiar:

Chase. Hold. Transfer. Cancel. Mark down.

The difficult part is identifying the need early enough that those actions are still available.

Consider a seasonal boot running materially below plan in the first week of peak. If the retailer spots the change immediately, there may still be time to cancel later receipts, shift inventory between channels, reduce replenishment or adjust promotional placement.

Find the same problem three weeks before the end of the season and most of those options have disappeared. Now the conversation is probably about markdown depth.

This is one reason spreadsheet-heavy peak planning becomes painful. By the time planners export data, reconcile different versions, update formulas and work through hundreds of SKU exceptions, another day of selling has passed. A forward-looking system such as Flagship can be useful here because the planner doesn't need another static report. They need early visibility into which inventory positions are moving away from plan and why.

BCG recommends a cross-functional holiday command-center model built around continuously assessing performance and making rapid pricing, promotional and operational decisions. In its footwear example, teams reviewed results across full-price, outlet and online channels and adjusted plans as the season developed.

That operating rhythm matters. Merchandising can't be chasing units while marketing is preparing to reduce support for the same product. Ecommerce can't be sitting short while stores hold excess inventory nobody has discussed transferring.

Peak season punishes slow information.

The best peak inventory plan isn't the one with the most accurate single forecast. It is the one that recognizes uncertainty before inventory is committed, protects upside where stockouts are expensive, limits exposure where excess is dangerous, and leaves the planning team enough room to react once customers show you what they actually want.

That is the point of peak forecasting.

Not to predict everything correctly.

To be wrong in places the business can afford.