Thesis: Observed sales are constrained by availability. A forecasting workflow that treats low sales during a stockout as low demand can recommend buying less of the product customers could not purchase. Predictive commerce needs to distinguish a weak product from a blocked buying opportunity before turning sales history into replenishment decisions.
This is especially important at variant level. A product can appear available while its most requested size, color, or pack is missing. The product-level sales total then hides the choice customers actually faced.
Reconstruct what was purchasable
Connect each sales period to the sellable availability of the relevant variant and channel. Warehouse quantity is not enough if units are reserved, awaiting inspection, or unavailable to the customer's delivery location. Record when the storefront could accept an order, with delivery promises and purchase restrictions where relevant.
A commerce ontology provides the relationships between product, variant, location, reservation, channel, and order. Without those relationships, a forecast may treat stock in one warehouse as an available choice for every shopper.
Historical snapshots matter. Today's stock level cannot reconstruct whether a variant was available during last week's campaign. Start collecting availability intervals before promising a precise account of earlier lost demand.
A replenishment error hidden in a weekly total
Illustrative example: A medium-size shirt sells 20 units in a week but is purchasable for only two days. Another variant sells 30 units while available all seven days. Ranking them by weekly units labels the medium-size shirt as the weaker seller.
Dividing 20 by two produces an observed rate of 10 units per available day. Multiplying that rate by seven would suggest 70 units, but that is only a crude scenario, not a validated demand estimate. The available days may have included a campaign burst, payday, or unusually high traffic. Demand after the stockout may also have shifted to substitutes or disappeared.
The useful first conclusion is narrower: the 20-unit total is not a fair measure of unrestricted weekly demand. The buying decision requires more evidence.
Triangulate unmet interest
- Compare availability with product traffic and attempted variant selections.
- Review back-in-stock requests, treating them as expressions of interest rather than guaranteed orders.
- Check whether substitute variants gained sales while the preferred option was unavailable.
- Compare similar in-stock periods while accounting for promotions, price, channel, and seasonality.
- Record whether a product was hidden from navigation or advertising after selling out; reduced visibility changes the observed opportunity.
These signals support a range of plausible demand, not a perfect count of lost orders. Do not add every unavailable selection, notification request, and substitute purchase together: the same shopper may appear in all three.
Account for substitution and return risk
A substitute purchase can preserve a sale, but it may transfer demand rather than create new demand. Forecasting the missing variant's full opportunity and retaining all substitute sales as independent demand can inflate the total buying plan.
Link substitution to post-purchase outcomes where the evidence allows. A shopper choosing a different size because the preferred size is unavailable may later return it, although that sequence alone does not establish the cause. Return intelligence helps teams investigate whether substitution is producing acceptable purchases or avoidable recovery work.
Translate uncertainty into a bounded buying decision
Keep observed sales, estimated unmet demand, and planned replenishment as separate quantities. Show the estimate's assumptions and range. Consider supplier lead time, minimum order quantity, season end, and the margin consequences of excess inventory before acting.
An effective predictive commerce workflow can recommend a limited replenishment test, a notification campaign after stock arrives, or a review when evidence is too weak. Automatically replacing missing sales with an optimistic estimate simply moves the error from underbuying to overbuying.
Check the estimate when availability returns
After replenishment, compare demand with the recorded estimate while accounting for changed traffic and promotions. Check how quickly stock sells, whether substitutes fall back, and whether returns increase. Avoid treating demand immediately after a restock as the permanent rate: a waiting list may concentrate purchases into the first few days.
Start with one category and its most consequential variant gaps. Within a Commerce Intelligence OS, connect availability evidence, demand estimates, buying decisions, and subsequent outcomes. The objective is a replenishment decision that can explain which sales were observed, which opportunities were inferred, and what the team learned afterward.
Book a demo to explore how iKawn can map availability and demand signals into a clearer replenishment workflow.