What FVA and the 50/50 Rule are
The concept of Forecast Value Added and the 50/50 Rule: when human intervention adds or destroys value versus the Baseline.
This page explains the purpose of FVA and the logic of the 50/50 Rule. Calculation variants, formulas, and numerical examples live in Reference → Forecast Value Added.
The question FVA answers
A typical S&OP process passes the AI forecast through several review stages: Sales adjusts it, Marketing modifies it, Consensus ratifies or changes it again. Each stage has costs: meeting time, analysis, negotiation. The natural question is: do those interventions actually improve the number?
Forecast Value Added answers that question quantitatively. It measures whether each area's intervention (Sales, Marketing, Consensus) moves the final forecast closer to or further from actual demand, always using the Baseline — the pure AI-generated forecast with no human adjustments — as the reference point.
- Positive FVA: the stage adds value; its intervention improved the forecast relative to the Baseline.
- Negative FVA: the stage destroys value; the forecast would have been more accurate without that intervention.
- FVA ≈ 0: the stage is neutral; it neither improves nor meaningfully worsens the forecast.
The Baseline is not a rival; it is the neutral comparison point that makes it possible to distinguish where human judgment contributes and where it introduces noise.
The 50/50 Rule
The 50/50 Rule comes from empirical evidence on manual adjustments in S&OP processes: statistically, of every 100 manual adjustments a team makes, approximately 50 improve the forecast and 50 worsen it. The net result is close to zero, but the cost of making those adjustments is real.
This does not mean manual adjustments are useless. It means that an adjustment without solid justification has roughly a coin-toss probability of improving the outcome. Market knowledge, when well-grounded, can tilt that distribution favorably; acting on noise leaves it the same or makes it worse.
How SofIA uses the 50/50 Rule
AInventory incorporates the 50/50 Rule directly into the workflow. Before recommending or allowing a manual adjustment, SofIA requires qualitative justification through five questions:
- Have new markets or customers been gained that the model does not yet know about?
- Have markets or customers been lost?
- Is a new product being launched (phase-in) with no comparable history?
- Is a product being discontinued (phase-out)?
- Is there an active transition from one product to another that the model has not yet captured?
If the answer to all five is "no," the manual adjustment has no justification based on new information; it simply introduces the team's opinion about randomness, which tends to worsen the forecast on average.
If the FVA of a stage (for example, Marketing) is negative consistently over several periods, SofIA triggers an alert and recommends reducing manual intervention at that stage. Additionally, the justification standard for new adjustments is raised: more explicit evidence is required before proceeding.
Why FVA is read alongside accuracy metrics
FVA compares stages against each other, but always on the basis of some error metric. The most common variants are:
- BIAS FVA: did the intervention correct or introduce directional bias?
- MAE FVA: did the intervention reduce or amplify the magnitude of the error?
Reading both together enables finer diagnoses: a stage may eliminate bias (positive BIAS FVA) but increase variability (negative MAE FVA), indicating the adjustment pointed in the right direction but was too aggressive in magnitude.
Relationship with the glossary
The terms Forecast Value Added, Baseline, phase-in, phase-out, and 50/50 Rule have entries in the bilingual glossary.