The Stockout Bias Problem

There is a quiet bug in how most retailers plan their inventory. It does not show up as an error in any report. It hides in plain sight, compounding season after season, silently eroding margin. It is called stockout bias, and if you run a multi-branch modern retail operation, it is almost certainly affecting your business right now.

What stockout bias is

Stockout bias is a feedback loop. It works like this:

1. Size 7 sells out at Store A in week three of the season.
2. The POS records zero sales for size 7 at Store A from week three onward. Not because demand stopped. Because there was nothing to sell.
3. The planning system reads historical sales at the end of the season. It sees that Store A sold fewer size 7s than the national average.
4. Next season, the planner allocates fewer size 7s to Store A, based on the "low demand" signal.
5. Size 7 sells out even earlier next season because it started with fewer units.
6. The cycle deepens. Each season, the system becomes more confident that Store A does not need size 7. Each season, it becomes more wrong.

This is not a theoretical risk. It is the default behaviour of any allocation system that uses unadjusted historical sales as its demand signal. And that includes most allocation processes in South African mid-market retail.

Why it is so dangerous

Stockout bias is self-reinforcing. Unlike most data quality issues that produce random noise, stockout bias produces systematic error in one direction: under-allocation of popular items. The items that sell best are the ones most likely to stock out, which means the items with the most demand are the ones most likely to be under-allocated next season.

The damage compounds:

Revenue loss

Every stockout is a lost sale. In fashion retail, where most purchases are impulse-driven and substitution rates are low (a customer who wants size 7 will not buy size 9), stockouts translate almost directly to lost revenue. The customer walks to a competitor, buys online, or simply goes home empty-handed.

Margin erosion through markdowns

The sizes that were over-allocated (because the system misread the demand signal) end up being marked down. You take a hit on sizes that nobody at that store wanted, while the sizes everyone wanted were out of stock. It is the worst possible combination: lost full-price sales on your best sellers and forced markdowns on your worst.

Distorted buying decisions

Stockout bias does not just affect allocation. It feeds upstream into buying decisions. If the data says size 7 demand across the network is declining (because stockouts suppressed recorded sales), buyers may reduce size 7 in the buy plan for next season. Now the problem is no longer confined to store-level allocation. It is baked into the entire range.

Invisible in standard reporting

Most retail reporting measures sell-through rates, GMROI, and markdown percentages. None of these metrics directly reveal stockout bias. A store that stocks out of size 7 in week three and marks down size 10 in week ten will show up as having a decent sell-through rate (size 7 sold fast) and an acceptable markdown rate (size 10 was managed). The fact that the store could have sold twice as many size 7s and needed zero size 10s is invisible.

How it specifically affects modern retail in South Africa

Stockout bias is a universal problem, but it has specific characteristics in the South African market that make it particularly acute.

Size-driven categories

Footwear and apparel are size-driven categories. A customer needs a specific size; there is no substitution. This makes stockout bias far more damaging than in categories where customers can switch to an alternative product. In homewares, a customer might accept a different colour. In footwear, size 7 means size 7.

Long import lead times

Most mid-market South African modern retailers import a significant portion of their range from Asia, with lead times of 60-120 days. This means you cannot respond to stockout signals by reordering quickly. By the time you realise size 7 is understocked, it is too late to get more. The next opportunity to correct is next season -- and by then, the stockout bias has already contaminated your demand signal.

Limited domestic manufacturing

Unlike markets with strong domestic manufacturing (where you can do quick-response replenishment), South Africa's modern retail supply chain is import-heavy. This extends the feedback loop and makes each iteration of the bias harder to correct.

Store network diversity

South African retail networks span dramatically different demographics. A store in Sandton City serves a very different customer profile than a store in Mamelodi. National-average size curves are practically meaningless when your stores span this range of demographics. Yet many retailers allocate using national or regional averages, embedding stockout bias from day one.

How to detect stockout bias

Detecting stockout bias requires looking at data differently than most retailers are accustomed to.

Lost sales estimation

The foundational metric is lost sales: the sales you would have made if stock had been available. This requires estimating what demand would have been during stockout periods, based on:

• Pre-stockout sales velocity at the same store
• Sales velocity at other stores that had stock during the same period
• Seasonal demand curve position
• Size curve intelligence for that specific store

Demand signal correction

Once you can estimate lost sales, you can correct the demand signal before feeding it into allocation. This means the planning system sees "Store A has high demand for size 7" rather than "Store A has moderate demand for size 7 (because it was out of stock for 60% of the season)."

Cross-store comparison

If Store A sold out of size 7 in week three while Store B still has size 7 stock in week ten, the raw sales data will show Store B selling more size 7s. But Store A might actually have stronger demand -- it just proved it by selling through faster. Comparing rate of sale (units per day of availability) rather than total units sold reveals the true demand picture.

How Replenify breaks the cycle

Replenify's approach to stockout bias operates at three levels.

In-season: Rebalancing

The rebalancing engine catches and corrects stockout situations as they develop. When size 7 is selling fast at Store A and slow at Store B, the engine recommends a transfer before the stockout occurs. This prevents the stockout that creates the bias in the first place.

The engine's 3-pass architecture is described in detail in How 3-Pass Rebalancing Works. The key point for stockout bias: by keeping stock available where demand exists, you generate clean demand signals for future planning.

Pre-season: Allocation intelligence

The allocation module builds stockout-corrected demand signals into every allocation decision. Size curve intelligence is calculated per store, using rate-of-sale data adjusted for availability rather than raw sales totals. This means a store that consistently sells out of size 7 gets more size 7s next season, not fewer.

The allocation module also addresses a related problem: style-store affinity. Some styles sell better in some stores. If a style sold poorly at Store C but was also allocated the wrong sizes, was it a store-style mismatch or a size allocation error? Disentangling these effects requires the kind of granular analysis that spreadsheet planning cannot deliver.

Strategic: Assortment planning

At the buying level, assortment planning uses corrected demand data to inform the size/colour/style matrix for next season's buy. This prevents stockout bias from contaminating the buy plan and perpetuating the cycle at the most strategic level.

The broader context

Stockout bias is one aspect of a larger challenge in merchandise planning: how do you make decisions based on data when the data itself is shaped by your previous decisions? This circularity is inherent in retail planning, and it requires analytical methods that account for it explicitly rather than treating historical sales as ground truth.

The cost of ignoring stockout bias is not dramatic. It does not cause a single catastrophic event. It slowly, season after season, erodes your margin by a few percentage points. For a retailer doing R300 million in revenue, a few percentage points is R6-9 million in annual profit. It is worth fixing.

Where to start

If you suspect stockout bias is affecting your business (and if you run a multi-branch modern retailer, it almost certainly is), start with measurement. Before you can fix the problem, you need to quantify it.

1. Pick your top 10 selling SKUs by rate of sale (not total units).
2. For each, identify which stores stocked out during the season and when.
3. Estimate the lost sales at each store using the rate of sale prior to stockout.
4. Compare those lost sales to the markdown cost of over-allocated sizes at the same store.

That comparison will tell you the cost of stockout bias in your business. For most mid-market modern retailers, the number is surprising.

For a comprehensive view of how inventory optimisation addresses problems like stockout bias, see our guide to inventory optimization for modern retail.