Skip to main content
Demand Forecasting

Forecast Value Added (FVA): how to measure what improves your demand forecast

Updated
19 November 2026
Reading time
11 min read
What Forecast Value Added (FVA) is and how to calculate it.
Back to blog

Forecast Value Added (FVA) helps you assess whether each stage of producing a forecast improves or worsens its performance against a reference. A forecast may pass through the statistical model, the planner’s review, Sales adjustments and a consensus meeting. Yet approving a final version does not mean it is more accurate than the original.

Many businesses measure the error of the approved forecast without retaining enough information to understand what happened during those reviews. When the forecast misses the mark, it becomes difficult to establish whether the problem lay in the model, a Sales adjustment or a decision made during consensus.

FVA analysis helps answer a specific question: which steps genuinely improve the demand forecast, and which need reviewing?

What Forecast Value Added is and why it matters

Forecast Value Added is the change in a forecast performance metric when a process stage is compared with a defined reference. It helps assess whether the work carried out on the forecast delivers an observable improvement. The method examines both models and subsequent interventions, as outlined in SAS’s Forecast Value Added analysis.

What FVA measures in the forecasting process

FVA measures the difference in performance between two comparable versions. It can establish whether a statistical model outperforms a simple forecast, whether the planner’s review improves that model or whether Sales input reduces the error of the previous version.

The chosen reference determines what is being evaluated. Comparing the consensus forecast with the statistical forecast shows the combined effect of subsequent reviews. Comparing it only with the latest Sales version isolates the change introduced during consensus.

A positive FVA therefore tells you little unless you specify which version it is measured against, which metric is used and which dataset is being assessed.

How FVA differs from forecast accuracy and BIAS

Forecast accuracy expresses how accurate a forecast is according to the chosen metric. Forecast BIAS identifies a systematic tendency to forecast above or below observed demand. FVA examines how that performance changes between versions.

A forecast can have a relatively low error and still perform worse than a simpler reference. Equally, it can improve substantially on its starting point and remain inadequate for certain purchasing or production decisions.

FVA relies on performance metrics; it does not replace them. Both the magnitude and direction of error matter, because reducing bias does not necessarily reduce absolute error.

A forecast review meeting focused on FVA.

What data you need for FVA analysis

To analyse FVA, you need to retain forecast versions, know when they were produced and have the observed outcome against which they will be evaluated. Your records must allow you to reconstruct comparisons without using information that was unavailable when each decision was made.

Record the reference forecast and subsequent versions

An analysis sequence might include:

  • Naïve forecast: a simple reference, such as repeating the last known value or the value from the same period a year earlier.
  • Statistical forecast: the model’s forecast before any human adjustments.
  • Planner-reviewed forecast: the version incorporating changes arising from the planner’s analysis.
  • Sales forecast: the forecast enriched with information about customers, promotions or other events.
  • Consensus forecast: the approved version that feeds into planning.

The naïve method should be appropriate for the series. Repeating the previous month may offer a poor reference where seasonality is pronounced.

Alongside the forecast value, record the date, owner, stage and reason for each change. This makes it possible to relate the outcome to the type of intervention. Arkieva’s documentation on forecast performance management describes this comparison between the initial forecast and its subsequent adjustments.

Compare the same horizon, SKUs and level of detail

Versions must be evaluated across the same products, locations, periods and units of measurement. Comparing a monthly forecast by product family with a weekly forecast by SKU introduces differences that prevent direct interpretation of FVA.

The time of generation also needs controlling. A review just before month-end may include orders that did not exist when the initial forecast was calculated. Its advantage should not automatically be attributed to better judgement.

If new information enters the process between stages, make that explicit and, where possible, produce an updated reference against which to assess the adjustment. Interpret the observed improvement in light of what each version knew.

Use a consistent measure of actual demand

Before calculating error, define what counts as observed demand: orders received, sales, units dispatched or another measure appropriate to the purpose of the forecast.

These measures are not interchangeable. A stockout may leave sales below unmet demand. Evaluating forecasts against those sales without accounting for the constraint can favour forecasts that are too low.

Returns, cancellations and corrections to historical data also require stable rules. If lost demand is estimated, identify it as an estimate and apply it consistently to every version being compared.

Two colleagues reviewing the forecast.

How to calculate Forecast Value Added

When a lower error metric indicates better performance, calculate FVA by subtracting the evaluated version’s error from the reference error. Both measurements must use the same metric and observations for the calculation to be meaningful.

Choose the metric and define the sign of the result

The basic formula is:

FVA = reference error − evaluated version error

Under this convention:

  • Positive FVA: the evaluated version reduces error.
  • Negative FVA: the evaluated version increases error.
  • Zero FVA: performance is unchanged according to that metric.

With MAE, the result retains the units of the variable being forecast. When subtracting percentage errors, the difference is expressed in percentage points: moving from 20% to 15% is an improvement of 5 percentage points, not a relative reduction of 5%.

The metric must suit the data. MAPE, for instance, causes problems when actual demand is zero and may be unsuitable for series with many periods of no consumption. Keep both the choice of metric and the treatment of these cases consistent throughout the analysis.

Calculate FVA for each stage and for the whole process

It is useful to distinguish two comparisons:

Incremental FVA compares a stage with the one immediately before it. It pinpoints where the forecast improves or deteriorates.

Cumulative FVA compares each version with a common reference, such as the naïve forecast. It shows how much of the improvement achieved since the start remains in the process.

A stage can have negative incremental FVA and still show positive cumulative FVA. This happens when it worsens an earlier forecast that was already performing considerably better than the initial reference.

FVA example with a statistical forecast and Sales adjustments

The following example shows how error changes for a single SKU as it moves through the process. All figures are hypothetical and refer to the same product, location and month, with observed demand of 1,000 units.

Compare the reference, model and adjusted forecast

Assume that the versions are recorded before the forecast period and that no new orders are introduced between the adjustment stages after the model. Absolute error is the distance between each forecast and the 1,000 units observed.

StageForecastAbsolute errorAbsolute percentage errorFVA vs previous stageFVA vs naïve forecast
Naïve reference80020020%—0 points
Statistical forecast950505%+15 points+15 points
Planner’s review980202%+3 points+18 points
Sales adjustment1,15015015%−13 points+5 points
Consensus forecast1,10010010%+5 points+10 points

Forecasts and absolute errors are expressed in units. FVA is expressed in percentage points. As this is a single observation, the table shows absolute percentage error rather than an average of errors across several periods.

The Sales adjustment calculation is:

Incremental FVA = 2% − 15% = −13 percentage points

The calculation for the whole process against the naïve reference is:

Final cumulative FVA = 20% − 10% = +10 percentage points

Interpret which intervention improves or worsens performance

Here, the statistical model improves on the naïve reference and the planner reduces error further. The Sales adjustment worsens performance, while consensus recovers part of that loss.

The final version outperforms the naïve forecast but is less accurate than either the statistical forecast or the planner’s review. Looking only at the final result would have concealed that difference.

Before changing the process, investigate why the forecast was raised to 1,150 units. Was a promotion expected? Was there an unconfirmed sales opportunity? Had a sales target been transferred into the forecast?

One unfavourable result does not prove that every Sales adjustment is harmful. Nor does one success confirm that the underlying assumption was correct. The example identifies which intervention to investigate, not which department should stop contributing.

How to use FVA to improve demand planning

FVA becomes useful when its findings lead to review rules: what deserves attention, which information justifies a change and which interventions need closer examination. Applying it should help allocate forecasting effort more effectively.

Identify when adjusting the forecast is worthwhile

A change should reflect relevant information that the model has not incorporated: a confirmed promotion, the loss of a customer or an assortment change, for example. Check whether the effect is already included to avoid counting it twice.

To evaluate the adjustment later, record the event behind it, the size of the change and the period during which its effect was expected. This is consistent with the recommendation in Forecasting: Principles and Practice to document and systematically evaluate expert judgement.

FVA can reveal patterns, but there is no universal rule making all large adjustments beneficial or all small ones unnecessary. Research by Fildes, Goodwin and De Baets on FVA in demand planning shows that the effect of interventions varies between datasets and adjustment types.

Prioritise reviews by product, horizon and event type

An overall average may conceal an intervention that works for promotions but worsens forecasts for stable products. Break results down into segments with enough observations to support analysis.

An initial segmentation might separate stable, seasonal and promotional products from new launches. Horizons, channels or adjustment reasons can then be analysed.

One decision might be to retain Sales reviews for confirmed promotions while limiting recurring corrections to stable SKUs when there is no new information. Collaborative forecasting becomes more useful when contributions have a clear purpose and their results are reviewed.

Weigh the improvement against review time and cost

Positive FVA does not automatically mean profitability. A small improvement may take many hours and barely affect subsequent decisions. Another may matter considerably if it involves a critical material or service commitment.

Assess the time spent, intervention frequency and operational consequences. Does the improvement prevent changes to purchasing decisions? Does it reduce urgent action? Does it affect SKUs with a significant financial impact?

FVA provides evidence of predictive performance. Establishing financial benefit also requires checking how the forecast is used.

Executives discussing Forecast Value Added.

Mistakes that can lead to misinterpreting FVA

Interpretation errors arise when different versions are compared, results are aggregated when they should be separated or too much significance is attached to a small sample. Stable criteria and sufficient context are needed to guide decisions.

Drawing conclusions from one cycle or an aggregate average

An exceptional month may favour one version for reasons that will not recur. Observe several cycles and check whether improvement is consistent before changing a working rule.

Retain enough detail to identify opposing results across segments. Positive overall FVA can coexist with persistent deterioration in a critical product family.

There is no single number of periods suitable for every case. Demand frequency, seasonality and the number of available adjustments determine the evidence required.

Mixing adjusted and unadjusted SKUs without distinguishing them

If only a small part of the range is changed, calculating FVA across every SKU may dilute the effect of those interventions.

Imagine reviewing 50 out of 1,000 SKUs. Adjustment analysis must compare before and after for those same 50 SKUs. Results across all 1,000 remain useful for evaluating the whole process, but answer a different question.

The adjusted SKUs may also be the most difficult ones. Comparing their error directly with that of unadjusted items would penalise the intervention without measuring what would have happened had the original forecast been retained.

Turning FVA into a ranking of people

Two planners may manage portfolios with very different levels of difficulty. Comparing them solely by final error or average FVA can obscure differences in horizon, volatility, available information and responsibilities.

Focus on understanding comparable decisions: what changed, what evidence supported it and what happened afterwards. An intervention with negative FVA may expose an information problem, a mistaken assumption or an unclear approval rule.

Using the indicator to rank individuals can discourage necessary adjustments and reduce the quality of shared information. Its value lies in improving the process and learning from decisions.

From FVA analysis to a more considered forecasting process

Forecast Value Added helps assess whether the effort spent reviewing a forecast improves its performance. Making use of it requires retained versions, comparable data and explanations for changes. With that foundation, a business can identify useful interventions, question habits that do not improve results and focus attention where relevant information exists.

As the product range grows and more teams become involved, maintaining this discipline requires shared data and a well-organised planning process. Demand planning software can help bring models, business knowledge and forecast monitoring together.

SCP Studio combines historical data, advanced models and business input to build forecasts through our demand planning software. If you would like to review how to organise this process in your business, request a demo with our experts and start with a concrete situation: which forecasts you use, who changes them and how you currently evaluate the outcome.

Subscribe to our newsletter and transform your management!

Receive updates and valuable resources that will help you optimise your purchasing and procurement process.