Picture the quarterly business review at most large retailers. The planning team presents forecast accuracy by category, the trend has improved since last year, and the room moves on. Finance then walks through the quarter: a promotion discounted deeper than demand required, a run of stockouts in the stores where a new line sold fastest, and inventory sitting in locations where it never moved. Both presentations are accurate, and neither one explains the other, which is why so many retail leaders who have invested heavily in AI forecasting still find it hard to say what that investment has returned.
The Scorecard Everyone Reviews and Few Can Act On
Forecasting is usually graded with statistical error measures such as mean absolute percentage error (MAPE) and weighted absolute percentage error (WAPE). They do one job well, telling a planning team how far a forecast landed from actual sales. What they cannot tell anyone is why the miss happened, how to correct it, or what it cost the business.
That gap has a predictable effect inside a planning organization. Planners who cannot see why a number moved begin to distrust it and move their adjustments back into spreadsheets. Over time, the AI investment meant to change how the business plans ends up running quietly in the background.
Ownership makes the problem harder to fix. Data science owns the forecast range, planners own the quantity ordered, and finance absorbs the markdown when the call goes wrong. Each team touches the outcome, and the question of whether the forecast actually improved the decision belongs to none of them.
When a forecast scores well and the quarter still disappoints, who in your organization is accountable for finding out why?
A Forecast Miss Rarely Costs the Same Thing Twice
The cost of a forecast miss depends entirely on the decision it touched. A shortfall in one store and a surplus in another both register as error, yet one cost the sale while the other tied up capital and headed toward clearance. A promotion forecast built on blurred history gives away margin on units customers would have bought anyway. An assortment plan that drifts from demand shows up in missed targets and in planning teams reworking the same numbers several times a season.
Three costs this different cannot sit behind a single accuracy score, and every quarter they are measured that way, the business loses sight of where its margin went.
Start with the Decision, Then Judge the Forecast
The white paper sets out a different starting point. Impact Analytics built decision accuracy around the question executives ask of every forecasting investment, which is what it changed in the business. Each forecast is judged by the allocation, pricing, or assortment decision it informs, and each of those decisions is scored on its own business outcomes, from stock kept on the shelf to margin protected at markdown.
Two features make the approach practical for leadership. Every result is benchmarked against the retailer’s own documented starting point, so improvement can be traced back to where it began. Progress is reviewed in quarterly business reviews, where data science, planning, and finance read from the same scorecard. The paper walks through each scorecard in detail, including the specific measures tracked for allocation and replenishment, pricing and promotion, and assortment and planning.
The accuracy question most teams get wrong
What counts as a good forecast depends on where it sits. The paper explains why a store, subclass, and week forecast near 40% accuracy can be a strong result, while a department and month forecast needs to clear a far higher bar, and what that means for how each decision should be judged.
Three Problems Hiding Inside Most Forecasts
A measurement framework is only as strong as the forecasts underneath it. The paper describes the Impact Analytics forecasting engine behind DemandSmart and the retail problems it was built to handle, three of which shape almost every planning calendar.
- Stockouts that disguise demand: When a shelf runs empty, recorded sales fall to zero, and a model trained on that history learns to expect less from an item customers were actively trying to buy. Every later forecast inherits the understatement unless the lost demand is restored first.
- Promotions that blur the baseline: A promoted item sells at one rate in an ordinary week and another during an event. Deep holiday discounts, demand pulled forward from the weeks after a promotion, and mid-event stockouts all distort the signal a model learns from.
- Launches without any history: New products arrive without the sales record most forecasting methods depend on, at the very moment the buy is committed and the stakes are highest.
The paper explains how the engine handles each of these, and how a model tournament selects the best-performing approach for every product, store, and season instead of forcing one method across the whole assortment.
The Cost of Another Season on Error Rates
When forecasting is judged only on error rates, the allocation, pricing, and assortment decisions made each season go unmeasured. The markdowns still get taken and the stockouts still happen, while the reasons behind them stay spread across three teams and several spreadsheets. When the next budget conversation arrives, leadership is asked to fund forecasting again with little evidence of what the last round delivered.
The retailers moving ahead are treating this as a measurement decision as much as a technology decision. They are setting a baseline, defining what each decision should be judged on, and holding forecasting to the same standard as any other investment on the P&L.
If your board asked this quarter what AI forecasting has returned to the business, how confident would the answer be?
Your Takeaways from the Paper
- The scorecard for each planning decision, and the measures leadership should track for allocation, pricing, and assortment
- The phased path from a documented baseline to a planning team managing by exception, and what the first phase requires from your data
- The results retailers see in margin, availability, and planning time when forecasting is measured this way
See What Your Forecasts Are Worth
Every buy, allocation, and markdown your teams make this season will start from a forecast. The Forecast Dividend shows how to measure what each of those forecasts returns, and how Impact Analytics helps retail leaders turn that measurement into margin they can report with confidence.
