Demand Planning & Forecasting Analytics

WMAPE, forecast bias, and tracking signals.
The SQL demand planners pull before every S&OP cycle.

WMAPE calculation, forecast bias detection, tracking signal computation, consensus vs statistical forecast value-add, promo lift attribution, and ABC×XYZ segmentation — the full demand planning analyst SQL curriculum across 30 missions.

See All Missions

30 missions · 13 free + 17 Pro · Starter → Master

How It Works

1

Read the briefing

A Slack message from your manager

2

Explore the schema

5 tables in a star schema

3

Write your query

Full SQL editor with autocomplete

4

Get expert feedback

Graduated hints, not just pass/fail

Why This Path

WMAPE vs MAPE

Learn why WMAPE (Σ|A−F|/ΣA) is the industry default — it weights by volume and avoids divide-by-zero on intermittent SKUs.

Forecast Bias Detection

Spot the systematic over-forecast on Category C that hides inside an acceptable WMAPE — the pattern demand planners escalate to leadership.

Tracking Signal

Compute rolling tracking signals and flag SKUs where |TS| > 4 — the statistical trigger for model recalibration.

Consensus Value-Add

Measure whether the human consensus override actually beats the statistical baseline — it only does 60% of the time in this dataset.

The Missions

Each mission is a real request from someone at the company. Difficulty increases as you go.

The Database

A demand planning star schema: 30 SKUs (4 intermittent, 2 NPI with no history), 5 locations, 4 forecast versions (STATISTICAL/CONSENSUS/OVERRIDE/NAÏVE), 480 forecast rows, 120 actuals. Planted data-quality issues: STATISTICAL version over-forecasts Category C by ~18% (bias pattern); CONSENSUS beats STATISTICAL on only 60% of periods; 2 promos with negative lift (pull-forward effect); 4 SKUs with zero-actual periods that destroy MAPE.

dim_skus (30)dim_locations (5)dim_forecast_versions (4)fact_forecasts (480)fact_actuals (120)fact_promotions (18)

dimension tables   fact tables

Ship the SQL every S&OP review demands

WMAPE, bias, tracking signals, consensus value-add — the queries that determine which models to trust.

Looking for something different?

|