ORIGINAL COMPANY-CONTEXT PRACTICE
Reconcile demand forecasts across products and regions
Original practice scenario: A hypothetical household-products planner forecasts individual SKUs, brand totals and regional totals independently. The forecasts disagree. How would you construct a coherent evaluation?
Written for independent practice using company context; this is not a verified question from a company interview.
A REASONED APPROACH
Model answer.
Define the hierarchy and decision horizon, then assess both accuracy and consistency. Independently good forecasts can still imply impossible totals when their aggregation relationships are ignored.
Work through it
- Create an effective-dated product-region hierarchy, accounting for discontinued products and new pack sizes before building comparable historical series.
- Backtest bottom-up and other justified reconciliation approaches using the same rolling cutoffs, evaluating error at the levels where planning decisions are made.
- Preserve nonnegativity or other required constraints and inspect intermittent-demand products separately, comparing coherence and error with unreconciled baselines.
Tradeoffs to discuss
- Bottom-up forecasts preserve detailed relationships but can be noisy for sparse SKUs; aggregate information may stabilize estimates while reducing local flexibility.
Common pitfalls
- Do not reconcile against hierarchy memberships that were unknown at the historical forecast date.
- A coherent set of totals is not necessarily an accurate forecast, so both properties need separate checks.