Executive Guide to Forecasting in Manufacturing, Part 4: Understanding Your Data

Forecasting accuracy doesn’t begin with algorithms, it begins with understanding your data.

In manufacturing and supply chain management, leaders often search for more sophisticated models to improve forecast precision. Yet, as Stephan Kolassa, Bahman Rostami-Tabar and Enno Siemsen remind us in Demand Forecasting for Executives and Professionals, the foundation of every reliable forecast is not mathematical complexity, but data literacy — knowing what your numbers really represent.

Before we can predict the future, we need to understand the past. That means understanding time series data: what it is, what it’s telling us, and where it may be misleading.

Key Insights from the Book

1. Know your time series
A time series is simply data measured over time: monthly sales, daily output, or hourly machine utilization. But not all time series are equal. Leaders should start by asking:

• What exactly is being measured:  Shipments, sales or demand?
  
• Are there missing periods, stockouts, or anomalies?
  
• How far back does the history go, and how relevant is it today?
  

One of the book’s practical reminders is the difference between sales and demand. Sales reflect what was purchased; demand reflects what customers wanted to buy. When stockouts or constraints occur, sales underestimates true demand, leading forecasts astray and misaligning production plans.

2. More data is usually better… if it’s meaningful
Executives sometimes discard older data, assuming it’s outdated. Yet long time series capture valuable context (from market cycles to exceptional disruptions) that improves forecasting resilience. Models like Exponential Smoothing can adapt to recent changes without losing the benefit of long-term perspective. Old data doesn’t just describe the past, it anchors our understanding of variability. Unless the business has fundamentally transformed, retaining it helps identify recurring dynamics that might return.

3. Recognize the patterns: level, trend, and seasonality
Every time series contains three structural components:

• Level: the baseline around which demand fluctuates
  
• Trend: a long-term increase or decrease
  
• Seasonality: recurring cycles tied to time or events
  

Recognizing these patterns ensures that forecasts reflect reality. Executives don’t need to master the math, but they must ensure forecasting systems detect and account for these components. Ignoring them turns predictable cycles into preventable surprises.

4. Decomposition: seeing what’s really driving change
A useful management technique is time series decomposition: breaking a dataset into trend, seasonality, and remainder. This helps identify what part of demand variation is structural versus random, and communicate findings clearly across functions.
In manufacturing, decomposition can reveal whether fluctuations are market-driven or process-induced. For instance, separating genuine customer seasonality from internal scheduling effects. Once the structure is visible, teams can address volatility where it matters most.

5. Forecastability and scale
Not all demand is equally predictable. Low-volume or highly intermittent products — common in spare parts or custom manufacturing — are inherently noisy. Aggregated categories, however, tend to be more stable. This creates an economy of scale in forecasting: broader aggregation often means greater reliability. Supply chains can exploit this by forecasting at higher levels (families, regions, components) and postponing differentiation until later stages.
A classic example is postponement in product design: producing standardized subassemblies, then customizing closer to demand. This strategy shifts planning to levels where forecasts are more trustworthy, balancing accuracy with flexibility.

Quantics Perspective

From an enterprise manufacturing perspective, the principles outlined in the book hold true, but their application is often more nuanced in practice.

Define the Objective Before Choosing the Data

It starts with clarifying the information needs of the business. What exactly should the forecast support? For example, if the goal is to plan outbound logistics, shipment data may be the most relevant input. Ideally, however, companies would forecast true demand (what customers actually want) — but this is often not directly or reliably recorded, which is why sales data is frequently used as a proxy.

At the same time, it is important to recognize that both sales and shipments are not pure reflections of demand. They are significantly influenced by internal processes such as order acceptance rules, delivery date promising, and production or capacity constraints. As a result, the data used for forecasting already embeds operational decisions, which are then reflected in forecasts for the future. In other words, companies are not just forecasting demand from the market — they are forecasting the outcome of their own decisions.

When Demand Patterns Break the Rules

The implications of this become especially clear in B2B manufacturing environments, where demand patterns are inherently irregular. Customers tend to order in batches, leading to long periods of zero demand at a product level, followed by sudden spikes. The challenge is therefore twofold: not only estimating how much will be sold, but also when. At granular levels such as SKU or customer-product combinations, trends and signals are often weak or even absent, making traditional approaches less effective.

A few practical implications we consistently observe:

• Data selection is context-dependent. Demand, sales, and shipments each tell a different story — be aware of the story they tell.
• Internal processes shape the data. Order promising, interplant traffic, capacity constraints, and planning rules directly influence what is observed as “sales.”
• Intermittent demand is the norm, not the exception. Especially in B2B, forecasting requires handling both volume and timing uncertainty.
• Granular data is often very noisy. At SKU level, signals are weak, making it difficult to extract reliable patterns without additional structure.

Leveraging Structure: Forecasting Across Levels

To address this, Quantics applies hierarchical forecasting approaches, leveraging information across multiple aggregation levels, from product groups down to individual SKUs. By learning from more stable patterns at higher levels and reconciling them with granular data, accuracy can be improved across the entire hierarchy.

Rethinking Accuracy: From Targets to Value Added

Another important implication is that not all data is equally forecastable. In large manufacturing organizations, variability differs significantly across business units, product portfolios, and markets. Setting uniform accuracy targets across the board is therefore misleading. What is achievable for a high-volume, stable product line may be entirely unrealistic for low-volume or highly customized products. A more meaningful approach is to benchmark forecasts against simple baseline models and measure value added.

In practice, this means:

• Avoid uniform accuracy targets. Forecastability differs structurally across products and business units.
• Always benchmark against a baseline. Improvement matters more than absolute accuracy.
• Track value added consistently. This applies not only to statistical models, but also to overrides and consensus inputs.
• Prioritize robustness over complexity. Methods must handle noise, structural breaks, and process-driven distortions.

At Quantics, we bring together data scientists and supply chain experts to help manufacturing organizations navigate this complexity. Our focus is not just on automating processes, but on enabling better decisions — by helping teams understand their data, challenge assumptions, and quantify what truly adds value. In doing so, we turn forecasting from a technical exercise into a practical driver of better planning and more resilient supply chains.

Practical Takeaways

Executives can strengthen forecasting outcomes by:

• Start with the business question. Define whether you need to forecast demand, sales, or shipments — and choose the data accordingly.
• Separate demand from sales. Always understand whether your data reflects true customer demand or transactions constrained by stock and internal processes.
• Understand how processes shape data. Order acceptance, capacity constraints, and planning rules directly influence what you observe and forecast.
• Preserve and leverage historical data. Long time series improve robustness and capture rare but instructive events.
• Audit data quality routinely. Visualize and sanity-check key time series before relying on them for decisions.
• Recognize structure in the data. Identify level, trend, seasonality, and randomness to interpret forecasts correctly.
• Account for intermittency. Especially in B2B, expect irregular demand patterns with both timing and volume uncertainty.
• Forecast at the right level. Use aggregation, hierarchical forecasting, or postponement where granular data is too volatile.
• Measure value added, not just accuracy. Benchmark against baseline forecasts and avoid uniform targets across very different business contexts.
• Prioritize robustness over complexity. Use methods that handle noise, structural breaks, and process-driven distortions reliably.
  

In Conclusion

Great forecasts don’t just come from smarter math, they come from smarter understanding.
By focusing on data quality, structure, and interpretability, manufacturing leaders can make forecasting a genuine decision asset, not just a reporting exercise.

In a world of volatility and complexity, the first step toward better predictions isn’t adding more algorithms, it’s seeing your data clearly.

At Quantics, we combine deep data science expertise with extensive experience in enterprise manufacturing to help teams understand, harmonize, and prepare data across complex ERP landscapes and systems, turning fragmented data into a reliable foundation for forecasting.

Disclaimer

In this post, we share highlights from the book “Demand Forecasting for Executives and Professionals” by Stephan Kolassa, Bahman Rostami-Tabar, and Enno Siemsen, together with our own reflections on how these ideas apply to today’s manufacturing supply chains. This is not a replacement for the book, but rather a guide to spark thought and discussion. For a deeper dive, we encourage you to explore the book itself - available here.