Let’s explore how hierarchy, level selection, aggregation, and disaggregation work and how manual adjustments and sales input combine with system-generated forecasts at different levels.
A forecasting hierarchy organizes data into different levels of detail. At the bottom, you have the most granular data, like SKUs or individual customers. As you move up, the data becomes more consolidated, grouping into categories like product families or regions. The level you choose to focus on depends on the types of decisions you’re trying to make.
Bottom level (SKU or customer level): This is where you’ll find the most detailed, and also the most noisy, data.
Middle level (product categories or regions): This level is useful for reporting and spotting trends, smoothing out individual variability.
Top level (corporate or executive summary): This is the aggregated, high-level view that helps guide big-picture decisions across the company.
Different stages of the demand planning process require different levels of detail:
Consensus and demand reviews usually rely on more aggregated data at higher levels, especially for long-term forecasts.
It’s about finding the right balance between detail and decision-making needs. The latest forecasting tools can work across multiple levels, aggregating or disaggregating as needed to find the level that works best for your situation.
Aggregation: This is where data from lower levels (like SKU-level forecasts) is rolled up into higher-level forecasts, such as for product categories or regions.
Disaggregation: High-level forecasts are broken down into more specific forecasts at lower levels, often using historical ratios (e.g. proportionally by customer or region according to the last 12 months) to guide the process.
Higher-level forecasts tend to be more accurate because they reduce the noise found in granular data. However, lower-level forecasts provide the specificity needed for execution. Aggregation and disaggregation help bridge the gap between accuracy and detail.
It’s common to measure forecast accuracy at different levels, with higher-level forecasts typically being more accurate. When using aggregation and disaggregation techniques, it’s important to define which level will be the key measure of success (as set out in your KPIs).
Manual adjustments can be made at any level, but making changes at the most granular level, like SKU or customer level, is generally preferred. It reduces the risk of making inaccurate assumptions when disaggregating data later. However, this is also the most time-consuming and complex level to adjust accurately.
Most companies use a mix of top-down and bottom-up forecasting: Top-down forecasting starts at the highest level and disaggregates proportionally to the lower levels. It’s easier and faster but can lose detail when breaking down the data. Bottom-up forecasting works from the most detailed level, aggregating forecasts upwards. This method captures more specificity but can get noisy and complex, especially if data is sparse.
How to define your forecasting hierarchy by Arkieva: https://blog.arkieva.com/how-to-define-your-forecasting-hierarchy/
How Does a Supply Chain Planning System Aggregate and Disaggregate Data? By Arkieva: https://blog.arkieva.com/supply-chain-data-aggregate-and-disaggregate/
Bottom-up, top-down and optimal combination methods for identifying forecasting hierarchy by Lancaster University: https://www.lancaster.ac.uk/~morganle/images/HierarchicalModels.pdf
The importance of hierarchy levels in demand planning by O9, both video & article:
Disaggregation in Anaplan simply explained with an example: https://community.anaplan.com/discussion/156263/can-anyone-describe-the-disaggregation-in-anaplan#:~:text=Disaggregation%20in%20demand%20planning%20is,time%20periods%2C%20or%20customer%20segments.
Disaggregation of the forecast by Valtitude: https://valuechainplanning.com/knowledge-base/SCM/Disaggregation
Arkieva on aggregation & disaggregation: https://blog.arkieva.com/supply-chain-data-aggregate-and-disaggregate/