Out of stock

the direct link to poor sales forecasting—and poorly coordinated pricing

Stockouts: A Symptom, Not an Isolated Incident

An empty shelf is immediately noticeable, whether in a store or online. What’s less obvious is what led up to it: an order placed based on a forecast that underestimated actual demand.

A shortage is almost never purely a logistical mishap—a late truck, a supplier that fails to deliver. More often than not, it’s the inevitable consequence of a figure that, just a few weeks earlier, suggested, “That’s all we’ll need.”

It is important to distinguish from the outset between stockouts and another problem that appears similar: markdowns or inventory clearance. A stockout occurs when there is insufficient inventory to meet demand.

A markdown is the opposite—too much unsold inventory, which eats into the margin. Both are symptoms of an inaccurate forecast, but in opposite ways, and they cannot be addressed using the same strategies—a point discussed in more detail below (§5).

A disruption can also be a pricing decision, not just a forecasting error

There is a third cause, one that is less often cited: stockouts triggered by pricing. An aggressive promotion or a price reduction launched without being based on reliable forecasts and available inventory can cause demand to surge beyond what inventory can handle—a stockout brought on by a pricing decision, not just by inadequate forecasting upstream.

The problem here is not that the forecast was wrong; it is that the pricing decision did not take it into account.

A previously published Booper article, “AI-Powered Sales Forecasting: Methodology and KPIs,” details the mechanics of calculating a forecast—and mentions data breaks as a technical factor that must be isolated before any calculations are made, so as not to skew the historical data used as input. This article delves deeper into this specific point: it explains why this technical precaution exists, the business cost of not taking it, and how an uncorrected data break becomes self-perpetuating.

The Cost of a Stockout, in Order of Magnitude

The retail industry talks a lot about “inventory distortion”—a term that encompasses two opposing problems: too little inventory (out-of-stock situations) and too much inventory (excess inventory, markdowns). Both are costly.

But one carries significantly more weight than the other, and it's not always the one you'd expect.

Two key figures to note in this study: the cost of stockouts ($1,200 billion) is more than double that of overstock ($572 billion). The retail intuition often leans toward overstock—an unsold product is visible; it sits on the shelf; it ends up on sale.

A missed sale, on the other hand, leaves almost no direct trace in the books: a sale that never took place generates neither an invoice nor a visible loss entry. Yet, on a global scale, it is the more costly of the two problems —precisely because it is the most difficult to measure without rigorous forecasting.

The vicious cycle: a breakup skews the sale it prevents

This is the core concept of this article, and it is rarely explained clearly: a data point outlier does more than just cost a sale in the moment. It permanently skews the data that will be used to calculate the next forecast.

Here's how bias takes hold, step by step:

• A product is out of stock for several days during a week of high demand—whether due to a seasonal peak, a sales promotion, or weather conditions.
• The system records this week's actual sales: they are low, for obvious reasons, since the product was simply no longer available for purchase.
• Nothing in this raw data distinguishes between “demand has fallen” and “demand existed but could not be realized.”
• A forecasting model—whether statistical or AI-based—that learns from this historical data without correcting it interprets this dip as a sign of weak demand and forecasts lower demand for the corresponding period in the future.
• Fewer forecasts mean fewer orders. Fewer orders mean another likely stockout of the same product during the same period.

The loop feeds back on itself. Each iteration amplifies the previous error.

Demand censoring. That is the name statisticians give to this bias: the observed data do not reflect true demand; they reflect demand filtered by available inventory.

Without explicit correction, a forecasting system—no matter how sophisticated it may be—optimizes based on a biased signal—and produces a consistent, reproducible bias that worsens over time when applied to the same benchmarks.

A side effect: cannibalization

The phenomenon becomes even more complicated due to cannibalization: when a product is out of stock, some of the demand for it shifts to a substitute product, whose sales rise artificially while those of the out-of-stock product plummet. Without proper adjustment, the forecast for the substitute product is also skewed—upward this time—for the following period.

The Cost to Customer Loyalty, Beyond the Lost Sale

The lost sale at the moment of stockout is the visible part of the cost. The less visible—but equally real—part is what goes through the customer’s mind when they find the shelf empty.

The same study estimates that the average share of revenue lost by a retailer due to out-of-stock situations is approximately 4 percent—a figure consistent with the proportion of global revenue lost to out-of-stock situations cited in §2. But that is not the most useful figure: it is 31 percent.

A missed sale isn't just the loss of a single line of business. In nearly one-third of cases, it means losing a customer who has just discovered—on the spot—that a competitor had something this store or website didn't.

The real cost isn't today's sale. It's the customer who, once they've gone to a competitor, doesn't automatically come back—even when the shelves are full the following week.

When this experience is repeated with the same products and the same customers, it cannot be remedied by a promotion the following week: it builds up over time and erodes the likelihood that a customer will return to the same store to buy that product the next time.

That is why stockouts are not just a supply chain issue. They are a forecasting issue—because it is forecasting that determines, in advance, whether inventory will be sufficient—and a customer retention issue, because it is customer loyalty, not just immediate revenue, that is at stake every time a shelf is empty.

Ultimately, it comes down to pricing. The speed at which a price readjusts once inventory is restocked—neither set too high due to a scarcity bias nor sold off at a fire-sale price out of panic—directly affects the likelihood of winning back a customer who has already gone elsewhere.

Stockouts and markdowns: two opposing symptoms of the same poor forecasting

Stockouts and markdowns share the same root cause—a forecast that didn't match actual demand—but they manifest in opposite ways, and confusing the two leads to applying the wrong solution to the wrong problem.

For the opposite problem of the one discussed here—excess inventory that is weighing on margins—Booper offers specialized support: markdowns and inventory clearance. But that’s not the subject of this article: a stockout is never resolved by selling a product at a discount that, by definition, is no longer in stock.

Both warrant a separate diagnosis, even though the root cause—a poorly calibrated forecast—is often the same.

The Wrong Reflex: A Breakaway That Escalates Into a Poorly Controlled Dribble

The confusion between the two sometimes takes a more insidious form: a poor transition from one to the other. A stockout creates pressure to react quickly as soon as the restock arrives—placing an urgent order to make up for the shortfall, then selling it at a deep discount once demand has dropped, or, conversely, maintaining a high price out of a scarcity bias.

In both cases, this isn't a new mistake: it's the same underlying governance issue—a pricing process that makes decisions without consulting the forecast—replaying itself in a different form.

Breaking the Vicious Cycle: Four Steps

The vicious cycle described in §3 is not a statistical inevitability. It can be broken through a methodical analysis of historical data and by detecting the break itself more quickly.

What Changes When a Forecast Explains Its Deviations

Correcting historical data after the fact is necessary, but insufficient if the next forecast remains a black box. What really changes the equation is a forecast capable of explaining why it predicts a particular figure—and monitoring capable of flagging a disruption as it’s forming, not after it’s been going on for three weeks.

Price as a Tool for Managing Demand

This isn’t the only option. For SKUs with tight inventory or low forecast confidence, price is also a tool for managing demand: adjusting the price or scaling back a promotion helps keep demand within the limits of available inventory—a form of demand shaping through price that complements pure supply-chain arbitrage.

A pricing team that coordinates with the forecasting team can defuse the risk of a stockout before it occurs.

The break is reversible

The vicious cycle described in this article is not inevitable, provided we treat it for what it is: a forecasting problem, not just a logistics or restocking problem.

This potential does not materialize on its own. It requires an organization that actively corrects the bias introduced by its past disruptions, rather than allowing its model to learn indefinitely from historical data that it already knows to be skewed.

This is as much a matter of governance as it is of statistical models— the reference guide for this topic explains what turns a forecast into an actionable decision, rather than a number that’s simply observed without being acted upon. And for the complete calculation process—methodology, KPIs, error formulas—the Booper article “Sales Forecasting with AI: Methodology and KPIs” remains the go-to technical resource.

Want to know where your data breaks are skewing your forecast?

Spend 30 minutes with our team to identify products that are chronically out of stock and determine how much they’re actually costing you in terms of both your sales forecasts and customer retention.

Let's plan an exchange →

Frequently Asked Questions

Also in this series

• Sales Forecasting: Methods, AI, and Best Practices (Reference Guide)
• Who Should Lead Sales Forecasting? Governance Among Sales, Purchasing, and Supply

Sources

• IHL Group, 2026 Inventory Distortion Study — ihlservices.com
• Gruen, T. W., Corsten, D., and Bharadwaj, S., “Retail Out-of-Stocks: A Worldwide Examination of Causes, Rates, and Consumer Responses,” Grocery Manufacturers of America / ECR, 2002 — featured in the Harvard Business Review, “Stock-Outs Cause Walkouts,” May 2004 — hbr.org
• McKinsey & Company, “AI-Driven Operations Forecasting in Data-Light Environments,” February 2022 — mckinsey.com
• McKinsey & Company, Dynamic Pricing in E-Commerce — mckinsey.com
• Booper, internal product data (`context/socle_booper.md` §3) — GENIUS Predict, GENIUS Monitoring, GENIUS Price.