Did The Planner Improve The Forecast? How To Measure Forecast Value Added

On Thursday a retailer learns that a partner has moved an email campaign forward by two weeks, so the demand planner increases next week’s forecast for the featured jacket. On Friday the commercial director nudges the number up again because the quarter is running behind. Both changes survive into the purchase plan. Six weeks later the company has a final forecast, an actual sales figure and a roomful of people with quite different recollections of who knew what.
A better question for that room than whether the final number was accurate is how much error each intervention removed or introduced. Forecast value added, usually shortened to FVA, measures the change in a chosen forecast performance metric from one step to the next. Done well, it helps a retailer decide which decisions deserve a planner’s time. That matters when the same person has half a morning to review six hundred lines and happens to know something genuinely important about three of them.
Keep The Number That Would Otherwise Disappear
Start by saving the forecast exactly as it stood before anyone touched it. A reproducible baseline needs a forecast date, a target period, a product and location or channel, a unit of measure and a lead time, and each material revision should carry the same identifiers. A number generated on the Monday of launch week has an information advantage over one generated at purchase-order cutoff eight weeks earlier. Comparing the two can tell you what the later information was worth. It cannot tell you how much better the planner was at the earlier buying decision.
Record each intervention while its rationale is still available. “Campaign email moved from week 24 to week 22, confirmed by channel owner” gives a reviewer something to investigate. “Commercial input” could describe anything from a signed order to optimism about the weather. A short reason and the time it became known are enough. Demanding an essay for every adjustment makes it harder to learn which ones pay for the trouble.
Keep a simple reference forecast alongside the operational one. That might be recent sales velocity for a continuing line or the comparable week of a prior season, depending on the decision. A weak statistical baseline can make an ordinary intervention look exceptional, while a strong seasonal reference can reveal that expensive steps in the process are merely restoring a calendar pattern that was available all along. Choose a reference someone could actually have used when the decision was due, with the same treatment of trading weeks and availability.
Three Adjustments Take Ten Units Out of Forecast Error
Imagine three fully available products in one selling week. At the same cutoff, the system predicts 90, 60 and 50 units. A planner revises those figures to 110, 75 and 45. Customers ultimately buy 120, 60 and 40. Use absolute error in units, so that an underestimate on one product cannot cancel an overestimate on another. The original errors are 30, zero and 10 units, totaling 40. The revised errors are 10, 15 and five, totaling 30.

The planner removed 10 units of absolute error across the three products. The first adjustment alone removed 20. The second introduced 15 where the original forecast was exactly right, and the third removed five. If the team reports mean absolute error, the improvement is 10 divided by three, or about 3.3 units per product. It is tempting to report a single 25 percent improvement and move on. The operational question is why the strongest revision was worth making, and what prompted the costly one.
Suppose the first revision reflected that confirmed campaign move, the second a desire to make the month’s sales target, and the third a correction for the loss of a display. Those details suggest an experiment for the next review cycle: preserve a route for confirmed events and changes in distribution, and track target-driven adjustments separately. Three observations cannot establish a durable rule, and a well-informed call can lose in a particular week. They can show the team which reasons deserve a larger sample.
The Number on The Till Can Give The Wrong Verdict
Consider a boot with a forecast of 80 pairs for the week. The warehouse has 45 pairs, and the last one ships on Wednesday. Comparing the forecast with 45 completed sales would favor a lower estimate, although the brand has no direct observation of demand on the other four days. Mark that line as availability-constrained. If the business estimates unmet demand, keep the method and its uncertainty visible, and see whether the conclusion survives under plausible alternatives. A category dominated by lines like this may not yet support a confident judgment about its overrides.
Selection sets another trap. Planners routinely touch the products with unreliable supply, short seasons or an unusual campaign, while untouched lines can be boringly predictable. Comparing revised products with untouched ones hands the harder assignment to the people whose work is being scored. Instead, compare each intervention with its own frozen baseline on the same items and weeks, then show how much of the whole book the interventions cover. A spectacular gain on four lines is still a four-line decision about staffing or software.
Spend The Next Hour Differently
An FVA review becomes useful when somebody changes the agenda. For a few buying cycles, keep the usual process but preserve the forecasts it would previously have overwritten. Group the resulting comparisons by reason, horizon and merchandise type, and look for gains that recur across decisions, along with the labor consumed to achieve them. A category buyer who spends Thursday afternoon shaving one unit off the expected error on low-margin accessories might make a larger contribution by calling a supplier about a delayed delivery of a bestseller.
Treat money as a separate calculation. Ten fewer units of absolute error do not put ten units of margin in the bank. An improved forecast can arrive after the last order cutoff, or the supplier’s case pack can leave the purchase unchanged. Conversely, revising a single high-value size before a cancellable order closes might avoid an expensive mistake. Trace a few interventions all the way from the original forecast, through the order decision, to the eventual outcome. That is where a team can judge whether an accuracy gain changed the business.
Some overrides deserve protection even though they never improve the demand estimate. A retailer may knowingly buy above expected demand to maintain display standards in a flagship store. Record that as a buying decision with an owner and a cost, and keep the best estimate of customer demand intact. Separating the two records makes retail demand planning more candid: the merchant can defend a choice about presentation without having to defend a forecast nobody believed.
By the next campaign move, the Thursday conversation can be wonderfully short. The planner has evidence that a specific event has changed, a deadline by which the adjustment could affect the buy, and a way to see afterward whether it helped. The commercial director still has a sales target. Each can put a number in the right place, and neither has to depend on the room’s memory six weeks later.