Definition
Stockout-censored demand estimation accounts for the fact that sales stop when sellable inventory runs out, even if customers still want the item. A zero-sales interval with no availability is not equivalent to a zero-sales interval with stock. The aim is to estimate missing demand while keeping observed orders and modeled quantities distinguishable.
Why It Matters
- Training a forecast directly on constrained sales can make a repeatedly unavailable product look unpopular. This can reinforce under-ordering.
- Searches, product views, and waitlists provide context but are not confirmed lost orders. Some shoppers substitute, return later, or would never have purchased.
- A Commerce Intelligence OS needs the availability conditions behind a demand signal before recommending replenishment or campaign changes.
How It Works
- Join SKU and location sales intervals to sellable availability, stockout start and end times, price, promotions, and traffic context. Distinguish unavailable from merely low stock.
- Identify comparable in-stock intervals and assess whether day, season, price, and exposure are sufficiently similar to support estimation.
- Estimate missing demand with documented assumptions and uncertainty. Keep substitutions and later purchases visible to avoid counting the same shopping mission twice.
- Evaluate estimates on held-out in-stock periods with artificially masked sales, then monitor forecast performance after availability recovers. A masked-data test does not prove every real stockout behaves the same way.
Ecommerce Example
Context: Illustrative example: a product sells 12 units before running out at noon. Comparable fully stocked days sell around 20 units.
Recommended move: Treat eight additional units as a provisional scenario, not eight verified lost sales. Check afternoon traffic, promotion changes, and substitute purchases before using it.
Why it matters: Planning receives observed sales plus an explicitly uncertain demand estimate. These hypothetical figures are not an iKawn customer result.
iKawn Framework
Observe
Within the iKawn framework, connect demand events with inventory availability in the commerce ontology.
Estimate
Model unavailable periods without relabeling estimates as actual transactions.
Decide
Use plausible demand ranges when considering inventory and campaign actions.
Validate
Compare later outcomes and revise assumptions when stock or shopper behavior changes.
Concise Summary
Availability constrains sales observations. Estimating censored demand can improve planning inputs, provided uncertainty, substitutions, and the distinction between actual and modeled demand remain explicit.