How Code-Driven Demand Modeling Eliminated My Inventory Cannibalization Woes

Have you ever noticed two of your product variants cannibalizing each other’s sales without you realizing it? I used to brush these dips off as random customer behavior, assuming my gut instinct about which variants to stock was on point—until I realized I was leaving money on the table.

Instead of guessing, I wrote a simple Python script to pull sales data directly from my e-commerce platform’s API, then used Pandas to clean and analyze six months of transaction records. The script tagged each SKU by key attributes: color, size, material, and price point. When I ran a correlation analysis, it revealed a 72% overlap in sales between two nearly identical blue hoodie variants—one priced $5 higher with a minor fabric difference that customers barely noticed.

Concrete Tactics I Used

• Scripted a weekly check to flag SKUs with a sales correlation score above 70%, indicating high cannibalization risk.
• Integrated margin data into the analysis to prioritize phasing out the variant with an 11% lower gross margin, rather than just the lower-selling one.
• Used a spreadsheet tool to simulate the impact of discontinuing the underperforming variant, projecting a 15% lift in sales for the remaining SKU without losing total customer volume.
• Added a filter to exclude limited-edition variants, which often drive brand engagement even if they overlap with core products.

The results speak for themselves: after phasing out the low-margin variant, I cut holding costs by 18% for that product line and increased overall gross margins by 12%. Intuition would have never caught the subtle overlap between those two hoodies, but code and data didn’t leave room for guesswork. For me, that’s the difference between running a business and letting it run you.