top of page

Forecast Accuracy for Free

Writer's picture: James TaylorJames Taylor

Updated: Oct 23, 2024


Fellow Gen Xer’s, do you remember blasting REM’s "What’s the Frequency, Kenneth?" on your cassette Walkman? Have you considered Michael Stipe’s cryptic lyrics when searching for forecast signals in the radio static of promotions, seasonality, and stockouts? If you want to filter out the distracting noise and get your forecast humming, there’s a way to do that. It’s called aggregation.


As aggregation increases, your plans benefit from statistical risk pooling—your forecasts become more accurate as the noise gets diluted. It's a quick win in supply chain forecasting, but often overlooked in favor of obsessing over algorithms. Setting the right forecast level can be more important than choosing the forecasting algorithm or software you use.

 

If you're focusing too granularly (think SKU level), you might be asking yourself, "What's the signal here?"—and missing it entirely.

 

While aggregation helps you with forecasting, you still need to disaggregate back down to a level that’s useful for actual decision-making. The reason we forecast is to support decisions about inventory, orders, or production. This balancing act—between forecasting and decision levels—is where the magic happens.

 

What’s the Right Level?

To figure out the sweet spot, planners need to understand two things:


1. The Decision Level: What level of detail is necessary to make decisions in the supply chain? Operations, purchasing, and sales teams will rely on this level of detail to act on the forecasts.


2. The Forecasting Level: What level of detail allows you to create an accurate and actionable forecast? Forecasting at too granular a level results in high variability and low accuracy, while too high a level may leave nuances left out.


Here’s the key: The decision level and the forecasting level can be different! When they are different, these two levels are connected by disaggregation logic—the methodology you use to break down the aggregated forecast back into actionable detail.

 

Questions to Ask Operations and Purchasing

To determine the correct decision level, ask your operations and purchasing teams the following questions:

  • What decisions are you using the forecast to make?

  • What level of granularity do you need to make those decisions? (e.g., SKU, category, or family)

  • How far in advance do you need to make those decisions?

  • Which locations or regions are critical for your decision-making?

  • Do you need forecasts broken down by specific channels?

  • What time buckets (weeks, months, or quarters) align with your planning cycle?

  • Are there particular products or regions where the lowest level of detail is essential?

 

Ways to Find the Right Forecast Level

Now that you know the questions to ask, let’s look at a few ways to find the appropriate level of forecasting aggregation:


1. Level Up to the Decision-Making Level: Ensure you’re forecasting at the level where users are actually making decisions. If your operations or procurement teams are making decisions based on product families rather than individual SKUs, then that’s where your forecast should focus.


2. Align Demand-Side and Supply-Side Groupings: In many businesses, there’s a many-to-one relationship between customer-facing products and the supply-side components used to make them. Small differences in demand-side products—like packaging or minor feature variations—often mask the fact that they share common supply-side materials or production processes. To improve forecast accuracy, aggregate demand-side products that have similar supply-side components, ensuring your forecasting aligns with how the supply chain operates.


3. Historical Data Analysis: Examine the volatility of your historical data at various levels of aggregation by looking at the Coefficient of Variation (CoV)—the ratio of the standard deviation to the mean. A high CoV indicates high variability, which can lead to poor forecast accuracy at detailed levels (e.g., SKU). If you notice that the CoV decreases significantly when you aggregate up (e.g., from SKU to product family), and your forecast accuracy improves, that’s a sign you’re forecasting at too granular a level. Aggregation smooths out the noise, making your forecasts more reliable.


4. Forecast Value Added (FVA) Analysis: FVA measures whether your current forecast model is actually improving accuracy over a simple naive forecast (like last year’s sales). If your detailed-level forecast isn’t beating the naive approach, it’s a sign you may need to move up to a higher aggregation level.

 

Setting the right forecast level can have more impact than choosing the forecasting algorithm or software you use.


5. Forecast Decomposition: Break your historical data down into its core components—level, trend, seasonality, and noise. At granular levels (like SKU or specific locations), the noise component can overwhelm the signal, making it difficult to detect meaningful patterns. As you aggregate to higher levels (e.g., product families or regional sales), the noise tends to smooth out, while the more predictable elements—like level, trend, and seasonality—become clearer. If your decomposition reveals that noise dominates at the detailed level but reduces significantly with aggregation, it’s a sign that you’re forecasting too granularly. Aggregation allows the signal (trend and seasonality) to shine through, improving forecast accuracy.


6. Pareto Analysis (80/20 Rule): While not strictly a way to identify forecasting level, an 80/20 analysis can help in two ways. First by identifying the most impactful product or product-location combinations to investigate with the other methods mentioned here. And secondly by identifying products or product locations that might be combined, rationalized, stocked together, or postponed; all strategies that could yield a more forecastable and profitable product portfolio.


7. Find the Right Number of Products to Forecast: It’s important to strike the right balance between too many and too few products in your forecast, especially when you are forecasting on a repeating weekly or monthly cycle. Aggregation can have the additional benefit of reducing planner workload. This number will vary depending on your industry, the number of demand planners you have, their other responsibilities, and the sophistication of your forecasting toolset. Forecasting too many products can overwhelm your team and dilute accuracy, while forecasting too few can lead to gaps in inventory planning.


8. Use More than One Level. The aggregation level for a 3-month S&OE horizon (weeks) may not be right when talking about plant capacity in S&OP year 3 (quarterly buckets). Use both.


Why This Matters

Getting the forecast level right means more than just improved accuracy—it means fewer stockouts, better inventory positioning, and reduced costs. If you’re looking for a quick win in your supply chain forecasting, stop overcomplicating things. Aggregation might get you some forecast accuracy for free.

 

Don’t forecast at a lower level than you need to. Always be talking to the users of your forecast about the decisions they are making with it.

 

At IBP2, our Demand Segmentation offering helps increase forecast accuracy by determining the right level of aggregation. We analyze your data across products, locations, channels, and time buckets to find the optimal balance between granularity and accuracy. Using advanced techniques like demand pattern classification and forecast decomposition, we help you categorize demand patterns, enabling you to forecast at the right level without sacrificing detail where it’s needed.


Would you like to explore these strategies in more depth? Reach out to IBP2 for customized solutions on optimizing your forecast levels and decision-making processes.

277 views0 comments

Comments


IBP2 Full 2024 Final Transparent.png

2 Mid America Plaza, Suite 902
Oakbrook Terrace, IL 60181

Thanks for submitting!

  • Linkedin
  • Facebook
  • X (formerly Twitter)

© 2025 by Integrated Business Planning Associates

bottom of page