A grocery leader should not approve a value claim because a forecast score improves. The decision is to require evidence that the revised forecast was available before replenishment was set, reached the ordering workflow and altered the order path. Its business effect must then be judged across every store-day eligible to use it, not just the cases where a quantity moved. A forecast measures expected demand. It is not an operating outcome.
The order is the transmission gate. A revised signal may arrive after the cut-off, sit outside the planner’s workflow, or be absorbed by case packs, minimum quantities, safety stock, shelf limits and supplier constraints. A recommendation may change while the submitted order stays the same because of system rounding or a planner override. The record must distinguish the forecast available at the decision, the recommended quantity, the submitted quantity and any override.
Even a submitted order is not sellable stock. Supplier fill, delivery timing, inventory records and shelf execution determine what becomes available. More stock may protect availability and raise spoilage or markdown exposure. Less may reduce waste and create an out-of-stock period. Promotions, seasonality, price changes and customer substitution can mimic or mask the result. Forecast accuracy cannot resolve these grocery trade-offs.
Forecast accuracy is upstream of the grocery outcome
The decision to make now
Before scale-up, write the evidence bar into the pilot charter. Define eligible decisions using facts known before outcomes are visible. Include every eligible treatment and comparison store-day in the primary business readout, whether or not the recommendation or submitted order changed. Changed orders show whether the forecast was transmitted through replenishment; they are mechanism evidence, not a selection filter. The eligible-population comparison answers the business question: did the chosen outcome improve relative to a credible comparable group?
Choose one category-appropriate primary outcome and retain the necessary service and economic guardrails. Availability is a service result, not recognised revenue. An available unit may be bought, substituted for another product, purchased elsewhere or later, or not sold. Revenue requires an observed, recognised sale and support for the treatment’s incremental contribution. Cash requires attributable net money paid or received. Shrink and markdown may affect margin, waste or inventory without automatically becoming cash.
Keep each value register separate. Cash records only observed, attributable net money movement. Capacity records measurable planner or store headroom, not assumed savings. Structural value records durable improvements to ordering discipline, data lineage and override governance. Modeled upside holds estimated lost sales, rollout scenarios and future margin effects until they are observed and attributed. These registers answer different questions on different evidence, so they should never be summed.
What would count as proof?
Proof is a linked record from decision to outcome. It starts with the forecast version and timestamp available before the order cut-off and a documented baseline ordering rule. It preserves the recommendation, submitted order, override and reason, with eligibility and hard constraints classified before results are read. A comparable group should be formed using information available before the decision, such as the promotion plan, season, prior sales and prior availability. Supplier fill, delivery, shelf availability, shrink and markdown stay visible as later events in the chain.
The chosen outcome should be followed over the product-appropriate order cycle and reported across all eligible store-days. Leaders should see the result against the comparable group, its uncertainty, how stable it is under reasonable alternative assumptions and the known attribution boundary. An accountable owner should record the resulting decision to scale, adjust, extend or stop. Order changes then help explain the result without determining which store-days count.
What remains unclaimed?
An accuracy gain alone does not establish adoption. An order change alone does not establish incremental margin, recognised revenue, realised cash or payback. Better availability remains an operating result until sales are observed and attributed, and lower shrink or markdown is not automatically cash. Released time remains capacity rather than savings. Evidence from one category, store group, order cycle or season does not establish an estate-wide result; expansion remains modeled upside until observed. A changed-order subgroup can illuminate the mechanism, but it cannot independently establish the business effect.
The claim is narrower than a forecast headline, and more useful: the signal reached a controllable decision, and the full eligible comparison shows whether grocery economics moved with it.