Make the regional and product forecasts describe one total

Establish which quantities genuinely add together

Choose one target measure and define its unit, period, currency treatment and population. Map the lowest-level records to the views planners need. Distinguish a nested structure, such as stores within regions, from crossed attributes, such as product and region. Check that the underlying observations obey the proposed aggregation before asking forecasts to obey it.

Hyndman and Athanasopoulos distinguish hierarchical and grouped series and describe coherent forecasts as respecting the same aggregation relationships as the data. Apply that principle carefully: counts of unique customers across overlapping audiences do not necessarily add, and conversion rates cannot be summed like orders. Resolve missing categories, duplicate membership and changing boundaries with the data owner instead of hiding them in a reconciliation adjustment.

Separate base predictions from targets and adjustments

Retain the independently produced base forecast for each series with its issue date and horizon. Explain which inputs and information were available at that moment. Identify any manual change as a separate judgment rather than silently rewriting the model output. A commercial target is an aspiration or commitment process; it should not become an observed fact the reconciliation must pretend to predict.

Ask the analyst to explain the selected reconciliation approach in terms of what evidence it uses and which constraints it imposes. A bottom-up sum, a top-down allocation and a method combining forecasts from several levels are not interchangeable. Do not select one merely because its national total is most attractive. Keep the original forecasts available so the team can see where the coherent set moved and why.

Compare performance across the structure

Recreate prior planning dates and compare the candidate coherent forecasts with the retained base forecasts and a defensible simple alternative. Use the same information availability and decision horizon throughout. Inspect performance at the levels where action occurs, not only the national total. A method that improves the aggregate may still produce unusable estimates for a smaller operational unit.

Ask the analyst to explain how forecast-error relationships, sparse history and newly introduced series affect the method. Inspect unexpectedly large adjustments and impossible values instead of treating an algorithm name as sufficient assurance. If practical constraints such as nonnegative demand are needed, have them incorporated and evaluated explicitly. Ad hoc clipping or changing one regional number after reconciliation can break the accepted relationships again.

Release a connected forecast package

Check every required aggregation relationship on the final delivered package for each period and horizon. Confirm that dashboards and spreadsheets use the same revision rather than mixing an updated regional forecast with an older national view. Label base, reconciled and judgmentally adjusted values so recipients understand what they are reading.

Keep uncertainty treatment separate from a simple check that point estimates add. Do not invent an interval for the total by casually adding individual upper and lower bounds; ask the analyst for a method that reflects the relevant dependence. Record category changes and new products as triggers to revisit the structure. Preserve the issued package for later evaluation, and report coherence and predictive performance as different acceptance findings.

Make the reasoning concrete.

A fictional subscription retailer forecasts next-month orders by product and by region. The regional sum exceeds the national forecast, while a new product appears in two commercial categories. A planner proposes scaling every row until the presentation total matches the sales target.

The team fixes the duplicated category and asks the analyst to compare coherent forecasts. It keeps the commercial target separately labeled and records where accuracy improves or worsens across the structure.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Measure

   Which quantities and units can legitimately be added?

2. Map

   Are the business views nested, crossed or overlapping?

3. Baseline

   Where are the issued base forecasts and commercial targets separated?

4. Evaluation

   At which levels and horizons must the coherent method work?

5. Release

   Do all delivered views use the same accepted revision?

Watch for the wrong conclusion.

• Summing overlapping unique-audience counts as though they were disjoint orders.
• Forcing forecasts to equal an aspirational target.
• Changing one component after reconciliation without checking the rest.

Sources & scope

Workflows, worksheets and fictional examples are original editorial analysis. The forecasting text supports the specifically attributed statistical distinctions. No observed result, universal accuracy guarantee or named human reviewer is claimed.

• Hyndman and Athanasopoulos: Forecasting hierarchical and grouped time series →Authoritative forecasting textbook · Supports: Hierarchical versus grouped aggregation, coherent forecasts and alternative reconciliation approaches. The acceptance contract and marketing example are original editorial analysis.