- Imperia
- Blog
- Demand Planning
- Forecast Value Added (FVA): how to measure what adds value to your demand forecast
Forecast Value Added (FVA): how to measure what adds value to your demand forecast
- Updated
- November 19, 2026
- Reading time
- 10 min read

Table of contents
- What Forecast Value Added means and why it is useful
- The data you need to analyze FVA
- How Forecast Value Added is calculated
- An FVA example using a statistical forecast and Sales adjustments
- Using FVA to improve demand planning
- Common mistakes when interpreting FVA
- Turn FVA insights into a better-informed forecasting process
Forecast Value Added (FVA) lets you check whether each step in creating a forecast improves or worsens its results relative to a benchmark. A forecast may go through a statistical model, a planner’s review, Sales adjustments, and a consensus meeting. Approving the final version, however, does not make it more accurate than the first.
Many companies track the approved forecast’s error but do not keep enough information to understand what happened during the reviews. When a forecast is wrong, it is hard to tell whether the issue came from the model, a Sales adjustment, or a decision made during consensus.
Analyzing FVA answers a practical question: which steps actually improve the demand forecast, and which should you revisit?
What Forecast Value Added means and why it is useful
Forecast Value Added measures the change in a forecasting performance metric when a process step is compared with a defined benchmark. It helps determine whether work on the forecast produces an observable improvement. The approach covers both models and later interventions, as described in SAS’s Forecast Value Added analysis.
What FVA measures across the forecasting process
FVA measures the performance difference between two comparable versions. You can use it to check whether a statistical model beats a simple forecast, whether the planner’s review improves the model, or whether Sales input lowers the previous version’s error.
Your benchmark determines what you are evaluating. Comparing the consensus forecast with the statistical forecast captures the combined impact of later reviews. Comparing it only with the latest Sales forecast lets you examine the change made during consensus.
Saying a forecast has positive FVA means little without specifying the comparison version, the metric, and the dataset used in the calculation.
The differences between FVA, forecast accuracy, and BIAS
Forecast accuracy describes how accurate a forecast is under the selected metric. Forecast BIAS reveals a systematic tendency to forecast higher or lower than observed demand. FVA tracks how that performance changes from one version to another.
A forecast may have a fairly low error yet perform worse than a simpler benchmark. It may also improve significantly on the starting point and still fall short of what purchasing or production decisions require.
FVA depends on performance metrics rather than replacing them. Evaluate both the size and direction of error: lower bias does not necessarily mean lower absolute error.

The data you need to analyze FVA
FVA analysis requires saved forecast versions, their creation dates, and the observed outcomes used to assess them. The records must let you reconstruct the comparisons without bringing in information that was not available when the decisions were made.
Save the benchmark forecast and later versions
Your analysis may follow this sequence:
- Naïve forecast: a simple benchmark, such as the last known value or the value for the same period in the previous year.
- Statistical forecast: the forecast produced by the model before human adjustments.
- Planner-reviewed forecast: the version containing changes based on the planner’s analysis.
- Sales forecast: the forecast updated with information about customers, promotions, or other events.
- Consensus forecast: the approved version used in planning.
Choose a naïve method that fits the series. Simply repeating last month may provide a weak benchmark for highly seasonal demand.
Record the date, owner, process step, and reason for each change, along with the forecast value. These details connect the outcome to the type of intervention. Arkieva’s forecast performance management documentation describes this comparison of the initial forecast and subsequent modifications.
Keep the horizon, SKUs, and level of detail consistent
Evaluate each version using the same products, locations, periods, and units of measure. A monthly product-family forecast and a weekly SKU forecast differ in ways that prevent a direct FVA comparison.
You also need to control when forecasts are generated. A review completed shortly before month-end may include orders that were unavailable when the original forecast was calculated. Better judgment is not automatically the reason it performs better.
When new information enters between steps, document it explicitly. Where possible, generate an updated benchmark to evaluate the adjustment. Consider the information available to each version when interpreting the improvement.
Apply a consistent definition of actual demand
Before measuring error, define observed demand: incoming orders, sales, units shipped, or another variable that matches the purpose of your forecast.
These measures are not interchangeable. During a stockout, sales can fall below unmet demand. Assessing forecasts against those sales without accounting for the constraint may favor forecasts that are too low.
Returns, cancellations, and historical data corrections need consistent rules as well. If you estimate lost demand, label it as an estimate and apply it consistently across all versions being compared.
How Forecast Value Added is calculated
For an error metric where lower is better, subtract the evaluated version’s error from the benchmark error to calculate FVA. Both measurements must use the same metric and observations for the result to be meaningful.
Select a metric and define what the sign means
Use this basic formula:
FVA = benchmark error − evaluated version error
With this convention:
- Positive FVA: the evaluated version lowers error.
- Negative FVA: the evaluated version raises error.
- Zero FVA: performance stays the same under that metric.
If you use MAE, the result keeps the units of the forecast variable. Differences between percentage errors are expressed in percentage points. Going from 20% to 15% improves the result by 5 percentage points; it is not a 5% relative reduction.
Match the metric to the data. MAPE has problems when actual demand is zero and may not suit series with many periods of zero consumption. Use the same metric and handling rules for these cases throughout the analysis.
Measure FVA at each step and across the full process
Distinguish between two comparisons:
Incremental FVA compares a step with the one directly before it, showing where the forecast improves or gets worse.
Cumulative FVA compares every version with a shared benchmark, such as the naïve forecast. It indicates how much improvement over the starting point the process retains.
A step may have negative incremental FVA while cumulative FVA remains positive. This occurs when it worsens a previous forecast that already outperformed the original benchmark by a wide margin.

An FVA example using a statistical forecast and Sales adjustments
This example follows the error for one SKU as it progresses through the process. All data is hypothetical and covers the same product, location, and month, with observed demand of 1,000 units.
Compare the benchmark, model, and adjusted forecast
Assume every version is recorded before the forecast period, and no new orders are added between the post-model adjustment steps. Absolute error is the distance between each forecast and the observed demand of 1,000 units.
| Stage | Forecast | Absolute error | Absolute percentage error | FVA vs previous stage | FVA vs naïve forecast |
| Naïve benchmark | 800 | 200 | 20% | — | 0 points |
| Statistical forecast | 950 | 50 | 5% | +15 points | +15 points |
| Planner review | 980 | 20 | 2% | +3 points | +18 points |
| Sales adjustment | 1,150 | 150 | 15% | −13 points | +5 points |
| Consensus forecast | 1,100 | 100 | 10% | +5 points | +10 points |
Forecasts and absolute errors are shown in units. FVA is shown in percentage points. Because there is only one observation, the table reports absolute percentage error, not an average of errors over multiple periods.
Calculate the Sales adjustment as follows:
Incremental FVA = 2% − 15% = −13 percentage points
For the full process relative to the naïve benchmark:
Final cumulative FVA = 20% − 10% = +10 percentage points
Identify which intervention helps or hurts performance
The statistical model beats the naïve benchmark, and the planner brings error down further. The Sales adjustment makes the result worse, while consensus recovers some of the loss.
The final forecast is better than the naïve version but less accurate than both the statistical forecast and the planner-reviewed version. Focusing only on the final result would hide that difference.
Before changing the process, find out why the forecast increased to 1,150 units. Was a promotion expected? Was an unconfirmed sales opportunity involved? Had a sales target been built into the forecast?
One poor outcome does not establish that all Sales adjustments are harmful. A single successful outcome does not prove the assumption behind it was right, either. This example points to an intervention worth investigating, not a department that should stop participating.
Using FVA to improve demand planning
FVA is useful when the findings shape review rules: what needs attention, what information supports a change, and which interventions deserve further investigation. Applying it should help you allocate demand forecasting effort more effectively.
Recognize when a forecast adjustment is worth making
An adjustment should use relevant information missing from the model, such as a confirmed promotion, a lost customer, or an assortment change. Check whether that effect is already accounted for so you do not count it twice.
To assess the adjustment later, document its triggering event, the size of the change, and the period when the impact was expected. This follows the advice in Forecasting: Principles and Practice to document and systematically evaluate expert judgment.
FVA can highlight patterns, but no universal rule makes every large adjustment worthwhile or every small adjustment unnecessary. Fildes, Goodwin, and De Baets’ research on FVA in demand planning shows that interventions have different effects across datasets and adjustment types.
Prioritize reviews by product, horizon, and event type
An overall average can mask an intervention that improves promotional forecasts but worsens those for stable products. Separate the results into segments with enough observations for analysis.
Start by distinguishing stable, seasonal, and promotional products from new launches. You can then examine horizons, channels, or reasons for adjustment.
You might keep Sales reviews for confirmed promotions while restricting repeated corrections to stable SKUs when no new information is available. Collaborative forecasting adds more value when contributions have a clear purpose and teams review their results.
Balance improvement against review time and cost
Positive FVA does not automatically translate into profitability. A slight improvement may consume many hours without materially changing downstream decisions. Another may be important because it affects a critical material or a service commitment.
Consider the time invested, frequency of intervention, and operational effects. Does the improvement avoid purchasing changes? Does it reduce urgent action? Does it affect SKUs with a meaningful financial impact?
FVA provides evidence about predictive performance. To establish financial value, you also need to examine how the forecast is used.

Common mistakes when interpreting FVA
Interpretation goes wrong when versions differ, results that need separate analysis are aggregated, or a small sample is given too much weight. Decisions need consistent criteria and enough context.
Relying on a single cycle or an aggregate average
An unusual month may favor one version for reasons that will not happen again. Review multiple cycles and check whether improvement holds before changing a working rule.
Keep enough detail to spot opposite outcomes across segments. Overall positive FVA may sit alongside ongoing deterioration in a critical product family.
No fixed number of periods works for every situation. Demand frequency, seasonality, and the number of available adjustments determine how much evidence you need.
Combining adjusted and unadjusted SKUs without separating them
When only a small share of the product range is adjusted, measuring FVA across all SKUs can dilute the effect of the interventions.
Suppose you review 50 of 1,000 SKUs. To analyze the adjustments, compare the before-and-after results for those same 50 SKUs. The result across all 1,000 still helps evaluate the entire process, but it answers another question.
Adjusted SKUs may also be the hardest to forecast. Comparing their error directly with unadjusted SKUs would penalize the intervention without measuring what would have happened if the original forecast had stayed in place.
Using FVA to rank individuals
Two planners may handle portfolios with very different forecasting challenges. Comparing them only on final error or average FVA can hide differences in horizon, volatility, available information, and responsibilities.
The goal is to understand comparable decisions: what changed, the evidence behind the change, and the outcome. Negative FVA may uncover an information issue, an incorrect assumption, or an unclear approval rule.
Ranking individuals with this metric can discourage necessary adjustments and weaken the information shared across teams. FVA is valuable for improving the process and learning from decisions.
Turn FVA insights into a better-informed forecasting process
Forecast Value Added helps you assess whether the work spent reviewing a forecast improves the outcome. You need saved versions, comparable data, and explanations for the changes. These foundations help a company identify useful interventions, revisit habits that do not improve results, and focus attention where relevant information is available.
As the product range expands and more teams get involved, this discipline depends on shared data and a well-organized planning process. Demand planning software can make it easier to combine models, business knowledge, and forecast monitoring.
SCP Studio builds forecasts by combining historical data, advanced models, and business input through our demand planning software. To explore how to organize this process in your company, request a demo with our experts. Start with a specific situation: which forecasts you use, who adjusts them, and how you evaluate the results today.
Related articles
Subscribe to our newsletter and transform your management!
Receive updates and valuable resources that will help you optimize your purchasing and procurement process.





