The Scale Problem
Supply chain leaders often face a very practical challenge. Modern product portfolios can contain thousands of SKUs that require demand forecasts, supply planning decisions, and procurement or production contracts. A typical consumer products company may forecast around one thousand active SKUs while managing fifteen thousand or more items across production and procurement systems.
For the planners responsible for these decisions, the workload can feel overwhelming. Forecasts must be updated, supply plans reviewed, and inventory policies adjusted regularly. When a supply chain head expects planners to manage this volume manually, the question naturally arises. How can any team realistically handle this scale without becoming mechanical operators inside the planning system?
The answer rarely begins with forecasting models.
Behavior Over Identity
A practical starting point involves clustering SKUs based on how they behave rather than what they represent as products. Many organizations classify products according to categories, brands, or product families because those structures already exist in marketing and sales systems. While useful for commercial analysis, these categories often fail to capture the operational characteristics that influence supply chain decisions.
Instead, planners should focus on behavioral attributes. These attributes describe how demand and supply behave for each SKU rather than what the product physically represents. When planners cluster SKUs according to behavioral patterns, the resulting groups become far more useful for planning policies.
This approach transforms thousands of individual SKUs into a manageable set of operational patterns.
Defining The Feature Space
The clustering process begins by defining the attributes that best describe product behavior. These attributes form what data scientists often call a feature vector space. In simpler terms, it is a list of parameters that describe the product from an operational perspective.
Examples may include value, sales volume, demand volatility, seasonality, coefficient of variation, promotional activity, product age since launch, and criticality to customers. Additional attributes might include the target service level, target sales channels such as airports or hotels, application type such as pharmaceutical excipients or pain relief products, and the primary sales channel.
Commercial attributes such as retail price, cost, or product manager ownership can also appear in the list. The goal is not to create a mathematically perfect dataset but to describe the behavioral dimensions that influence planning decisions.
Once these attributes exist, clustering algorithms can begin identifying patterns.
Clustering The Portfolio
One widely used method for grouping such data is the k-means clustering algorithm. The technique examines the feature space and groups products into clusters whose behavioral attributes resemble each other. Each cluster then represents a set of products that behave similarly in the supply chain.
For example, one cluster might contain low value, high volume products with stable seasonal demand patterns. Another cluster might contain high value products with erratic demand and long replenishment lead times. A third cluster might represent new products with uncertain demand behavior and high promotional sensitivity.
Each cluster now represents a planning problem rather than a product category.
Planning Policies By Cluster
Once clusters exist, forecasting models and supply planning policies can be assigned according to behavioral characteristics. Products with stable demand, high volume, and short lead times may perform well with exponential smoothing forecasting models and continuous review inventory policies based on reorder points.
Clusters with volatile demand and long lead times may require different approaches. Safety stock levels might require frequent review, and planners may need more conservative replenishment policies to avoid stockouts. Promotional products may require specialized demand sensing methods or marketing coordination.
The key advantage of clustering is that planners no longer manage thousands of individual planning strategies. They manage a limited number of behavioral groups.
The Math Anxiety
Many supply chain managers hesitate when discussions turn toward clustering algorithms or machine learning techniques. These concepts often appear intimidating because they involve unfamiliar terminology or statistical methods not covered in traditional business education.
In reality the mathematics behind many clustering techniques is surprisingly straightforward. Tools available online can perform clustering calculations automatically once the data is prepared. Even basic calculators can produce reasonable clusters for operational use.
Understanding the conceptual logic behind clustering matters far more than mastering the underlying mathematics.
SAP And Clustering
Planning platforms such as SAP Integrated Business Planning provide powerful forecasting and supply planning capabilities, but they do not automatically perform SKU clustering inside the planning engine. Organizations typically perform clustering using analytical tools such as SAP Analytics Cloud or external analytical environments.
Once clusters exist, planners can assign those cluster classifications back into IBP as master data attributes. Forecasting models, safety stock rules, and planning policies can then reference those attributes when generating supply chain plans.
This separation ensures that the planning engine remains focused on operational decisions rather than data analysis.
Demand Analytics First
The broader lesson is simple but frequently overlooked. Demand analytics deserves as much attention as forecasting itself. Organizations often rush directly into forecasting models and inventory optimization policies without first understanding the behavioral structure of their product portfolios.
Without clustering, planners treat thousands of SKUs as independent planning problems. With clustering, those same SKUs become manageable groups governed by consistent planning policies.
For organizations managing large product portfolios, this distinction can transform the effectiveness of supply chain planning.
A Practical Conclusion
Clustering also provides an additional benefit. Once the SKU portfolio is grouped into behavioral clusters, organizations can estimate how many planners are realistically required to manage the planning workload. A planner responsible for two clusters may require a very different workload than one responsible for six clusters with volatile demand patterns.
In many cases the exercise provides clarity on staffing requirements as well as planning strategy.
Which leads to an amusing observation.
That insight alone could easily appear in a consulting presentation costing two hundred thousand dollars.
