Definition
Temporal aggregation combines observations or forecasts across time, such as daily unit demand into a weekly total. The aggregation rule depends on the measure: unit flows can be summed, while a stock balance or conversion rate needs a different rule. A forecast for a period total must also represent uncertainty across its component periods.
Why It Matters
- A buyer ordering for a week needs evidence about that whole week, while warehouse staffing may need individual days.
- The Commerce Intelligence OS framework connects the forecast horizon to the commercial action it supports.
How It Works
- Define the decision window, calendar boundaries, units, and forecast origin. Keep incomplete periods and missing observations explicit.
- Compare forecasting the aggregate directly with aggregating finer-grained forecasts. Evaluate both on the same future windows using only information available at the origin.
- For expected unit totals, add the component means. For uncertainty, account for dependence across periods or aggregate jointly simulated paths; adding daily interval endpoints does not generally produce a valid weekly interval.
- Retain daily context when peaks matter to execution. Judge performance on both the purchasing window and any operational constraints hidden by the total.
Ecommerce Example
Context: Illustrative example: daily expected demand of 10, 12, 9, 11, 15, 18, and 15 units sums to 90 for the week.
Recommended move: Use a weekly uncertainty calculation that preserves dependence between days before setting a replenishment buffer.
Why it matters: The 90-unit total does not reveal whether a weekend staffing bottleneck exists. These hypothetical numbers are not an iKawn forecast result.
iKawn Framework
Align
The iKawn framework links planning windows to buying decisions.
Combine
Apply an aggregation rule appropriate to the commerce measure.
Quantify
Preserve dependence when calculating uncertainty for totals.
Review
Retain the finer detail needed for fulfillment execution.
Concise Summary
Aggregate forecasts to the decision window, preserve the correct measure semantics, and calculate uncertainty for the total rather than adding interval endpoints.