Why Over-Guessing One Week and Under-Guessing the Next Still Hurts 100%
If your 4-week total forecast is 400 and actual sales are 400, your average bias looks like 0—but guessing wrong in every single week creates a 100% absolute timing error (WAPE).
Interactive calculator loads automatically in modern browsers. All plain-English explanations, units, and tables are shown directly on this page.
How This Rule Works in Simple Steps
What We Want to Achieve: separate accuracy from bias — Positive and negative errors cancel in the total.
- WAPE = 100 * sum absolute errors / sum actual demand.
- Forecast-minus-actual is the declared signed-error convention.
- If total actual is zero, WAPE is undefined and must be labeled.
- Keep pure calculation code separate from storage and UI.
Three Simple Steps We Compare on This Page
- 1. Normal Day (Baseline Plan): Apply the baseline policy to the frozen input state.
- 2. Problem Day (When Something Goes Wrong): Evaluate the alternate forecast against the same actual demand.
- 3. Fixed Plan (How We Solve It): Evaluate a proposed revision; do not treat it as a future guarantee.
Important Things to Remember Before Using in Real Life
- Four toy periods; no claim of causal forecasting improvement.
- Passing this model does not validate a complete real supply chain.
- A case must disclose every assumption used in the calculation.
7 Worked Examples of "Why Over-Guessing One Week and Under-Guessing the Next Still Hurts 100%"
Click any of the 7 worked examples below to open its own separate page with step-by-step numbers and a visual bar chart:
- SCI-106: Zero bias large error — Positive and negative period errors cancel. (A correct total does not prove accurate timing.)
- SCI-107: Zero demand period — One period has zero observed demand. (The example reports WAPE with a positive horizon denominator.)
- SCI-108: Forecast shifted one period — The forecast pattern occurs in the wrong periods. (Timing error remains visible even if the horizon totals match.)
- SCI-109: Revision evaluated fairly — A revised forecast uses the same evaluation window. (A revision must not choose only favorable periods.)
- SCI-110: Outlier dominates aggregate — Large periods dominate a volume-weighted score. (No single metric fully describes every business objective.)
- SCI-111: Sign convention disagreement — Teams use opposite definitions for bias signs. (An unexplained sign convention can reverse the verbal interpretation.)
- SCI-112: Accuracy is not a service plan — Forecast scoring is separate from inventory decisions. (The case teaches measurement rather than an automatic stock policy.)