Time Bucket Optimization

Written By Ashutosh Bansal

Traditional Forecast engines forecast at the time bucket level defined by the Demand Planner, then disaggregate and store the resulting forecasts at the base level (e.g., Week | Product ID | Customer ID).

Temporal Aggregation enables vyan.ai to evaluate Forecastability in all higher time buckets available in the Time Dimension.

For example, Time Dimension may have the following attributes within it:

  • Day

  • System Week

  • (Calendar) Week

  • Month

  • Quarter

  • Year

In this case, vyan.ai will automatically generate Forecasts in all aggregated time buckets per list below:

  • Month | Product ID | Customer ID

  • Quarter | Product ID | Customer ID

  • Year | Product ID | Customer ID

vyan.ai generates disaggregation factors to split such aggregated forecasts down to the Base Level: Week | Product ID | Customer ID. This is needed to ensure apples-to-apples comparison of all forecasts in terms of forecast error.

This enables vyan.ai to understand which levels of aggregation enable better forecastability.

vyan.ai Optimal Blending process then allocated optimal weights to all such competing forecasts from all time buckets (across all forecast models and all hierarchy levels).

Ashutosh Bansal
Previous

Forecast Level Optimization