Identifying SKU Divergence Across RDCs Using Clustering

Why Traditional Classification Isn't Enough

If you're managing a regional or global supply chain, you're probably already using ABC and FMS classifications to optimize inventory. These methods are helpful, but they fall short when trying to answer questions like:

• Why is this SKU a top performer in one RDC but underperforming in others?

• How do I detect divergence without manually checking each region’s performance?

What Is Clustering and Why Use It?

Clustering is a machine learning technique that groups similar data points together — without requiring you to define the groups upfront. It’s perfect for situations where:

• You don’t know how many types of performance patterns exist.

• You want to detect anomalies or “outlier” SKUs.

• You want to find natural groupings of SKUs across RDCs based on their classification patterns.

How to Detect SKU Divergence with Clustering

Here’s how you can do it using Python and common data science tools like pandas and scikit-learn.

1.Collect ABC-FMS Classification Data

Prepare your matrix with ABC-FMS data per RDC:

2.Encode the ABC-FMS Data

Split and encode each classification:

from sklearn.preprocessing import OrdinalEncoder

import pandas as pd

# Sample data

df = pd.DataFrame({

'material': ['SKU1', 'SKU2', 'SKU3'],

'RDC1': ['A-F', 'C-S', 'A-F'],

'RDC2': ['A-F', 'C-S', 'C-M'],

'RDC3': ['B-F', 'C-M', 'B-S']

})

# Melt to long format

long_df = df.melt(id_vars='material', var_name='RDC', value_name='ABC_FMS')

long_df[['ABC', 'FMS']] = long_df['ABC_FMS'].str.split('-', expand=True)

# Encode ABC and FMS

encoder = OrdinalEncoder(categories=[['C', 'B', 'A'], ['S', 'M', 'F']])

long_df[['ABC_enc', 'FMS_enc']] = encoder.fit_transform(long_df[['ABC', 'FMS']])

# Pivot to get a feature vector per SKU (each RDC becomes two features)

pivot_df = long_df.pivot(index='material', columns='RDC', values=['ABC_enc', 'FMS_enc'])

pivot_df.columns = [f'{i}_{j}' for i, j in pivot_df.columns]

pivot_df.reset_index(inplace=True)

3.Apply Clustering (e.g., K-Means)

from sklearn.cluster import KMeans

X = pivot_df.drop(columns=['material'])

kmeans = KMeans(n_clusters=3, random_state=42)

pivot_df['Cluster'] = kmeans.fit_predict(X)

Now, each SKU is grouped based on how similarly it behaves across RDCs.

4.Identify Divergent SKUs

The trick: look for clusters with very few SKUs — these are your outliers or divergent items.

sku_counts = pivot_df['Cluster'].value_counts()

pivot_df['Cluster_Size'] = pivot_df['Cluster'].map(sku_counts)

# Flag small clusters as divergent

pivot_df['Flag'] = pivot_df['Cluster_Size'].apply(

lambda x: 'Divergent' if x <= 2 else 'Typical'

)

5.Visualization: Understand the Groupings

You can use PCA or t-SNE to reduce the SKU feature space and plot SKU clusters:

from sklearn.decomposition import PCA

import matplotlib.pyplot as plt

pca = PCA(n_components=2)

components = pca.fit_transform(X)

pivot_df['PC1'] = components[:, 0]

pivot_df['PC2'] = components[:, 1]

plt.figure(figsize=(8, 5))

for cluster in pivot_df['Cluster'].unique():

cluster_data = pivot_df[pivot_df['Cluster'] == cluster]

plt.scatter(cluster_data['PC1'], cluster_data['PC2'], label=f'Cluster {cluster}')

plt.title('SKU Clusters Based on ABC-FMS Patterns')

plt.xlabel('Principal Component 1')

plt.ylabel('Principal Component 2')

plt.legend()

plt.grid(True)

plt.tight_layout()

plt.show()

Business Value

This approach enables:

• Smarter inventory segmentation across RDCs

• SKU-specific replenishment policies instead of global rules

• Early detection of behavioral changes across geographies

Final Thoughts

In an era of demand-driven and decentralized inventory, static ABC-FMS classifications are no longer enough.

Clustering gives you the power to detect divergence automatically, helping you drive more localized, intelligent inventory decisions.