Forecast Value Added measures the value, positive or negative, that each step of the forecasting process adds relative to a naïve reference forecast. It is the only diagnostic that dares to answer an uncomfortable question: does this step, this meeting, this person, actually improve the forecast, or waste time?
Why it matters
Most indicators measure the quality of a forecast once it is produced. Forecast Value Added measures something else, far more uncomfortable: the real contribution of each person and each step to that quality. Does the statistical model improve on the naïve forecast? Does the planner's adjustment add value, or remove it? Was the monthly consensus meeting truly worth the time it cost all its participants?
The answer often surprises, and always unsettles. Published studies on real data, aggregating several companies, converge: manual interventions improve the forecast only to a limited extent, and a significant share of them plainly degrade it. A statistical number that was already good gets retouched by hand and made worse. Forecast Value Added makes this invisible waste finally visible, quantified, and therefore fixable.
One counter-intuitive result comes up systematically in the research: downward adjustments are often more effective than upward ones. When a planner revises a forecast down, they usually correct commercial optimism; when they revise up, they often add their own optimism.
Another expert lesson: the size of the adjustment is no guarantee of value. Large adjustments are not the most useful. A small adjustment based on precise field information (a known promotion, an exceptional order) beats ten large mood-driven ones. FVA lets you tell the two apart, objectively.
The mechanism
The principle is clear: define a naïve reference forecast, capture the frozen forecast at each step of the process, measure each one's error against actuals, and attribute to each step the gain or loss it produces.
On the chart, each bar is the forecast error at one step of the process, lower being better. The naïve reference, on the left, starts at 20%. The statistical model brings it down to 15%: it adds five points of value, it earns its place. The planner's adjustment gains another two points, down to 13%.
Then comes the painful step, the one only FVA unmasks: the marketing adjustment pushes the error back up to 18%, in red. This step removes five points of value: it would objectively have been better left out. That is exactly what the stairstep report reveals, and what no other indicator shows. An uncomfortable intuition becomes a quantified fact, calmly debatable.
Figure 1. The stairstep report: forecast error at each step. The statistical and the planner add value; the marketing adjustment removes it. Illustrative values inspired by published cases.
The traps
Forecast Value Added is simple in principle, but poorly run it wrongly accuses or wrongly excuses.
A poorly chosen naïve reference makes the whole process look brilliant by contrast. The naïve must be honest and fit the demand: the last actual for stable demand, or a seasonal naïve if demand has a marked season. Too dumb a naïve flatters artificially.
A step can be right once by luck and wrong ten times. Forecast Value Added is validly read only over a series of periods, segmented by product and channel, never on an isolated shot that proves nothing.
Retouching a lot is not adding value, often the reverse. Research shows large adjustments are not the most useful, and downward corrections are generally more effective than upward ones.
Setting it up
Installing Forecast Value Added needs no specialized tool: forecast history at each step, and the discipline to compare honestly.
Choose a forecast with no human judgment: the last actual (random walk), or a seasonal naïve. It is the yardstick against which everything is measured: F = D of the previous period.
Record the forecast as it leaves each step, frozen before the next: raw statistical, planner adjustment, marketing input, S&OP consensus. Without this systematic capture, no analysis is possible.
Compute the MAPE or WMAPE of each version against actuals, over the same periods. The comparison to the reference gives the step's value added: FVA = error(naïve) − error(step).
A step that repeatedly adds value is reinforced and equipped; a step that systematically removes it is dropped or reframed without hesitation. Planners' scarce time refocuses where judgment truly pays.
Neighboring concepts
Forecast Value Added builds on error measures and sheds light on the value of consensus.
From knowledge to action
The question deserves a quantified answer, not a conviction. Our Planning, Forecasting & S&OP file installs Forecast Value Added and draws the decisions from it, sometimes radical ones.