Back when I first worked on eCommerce planning, forecasting was a guessing game. We stretched last year’s numbers forward, added a rough uplift for promotions, and hoped demand would follow the pattern.
The pain points were clear: forecasts built only on history, ignoring seasonal shifts; festive peaks like Christmas or Lunar New Year treated as random spikes; promotions distorting the baseline forecast; and hours spent in Excel, with results that still lacked confidence. And every time stock-outs or overstocks happened, the question returned: why wasn’t this forecasted?
Traditional options felt too technical or too rigid — heavy forecasting tools that needed specialist teams, consultants building models no one on the ground could maintain, and manual Excel work that couldn’t keep up with real-world changes.
Fast forward to now. When faced with thousands of SKUs that need a rolling 12-month forecast, I didn’t start with complex systems. I started with AI for guidance, Python for execution, and monthly-level data for manageability.
Step 1 — Structure the data
Instead of daily rows, I consolidate at the monthly level, which keeps the file light but still highlights patterns. My table looks like this: Month | SKU | Quantity Sold | Promotion Flag | Festive Flag. This way, each SKU has just 12 rows per year. Over three years, that’s 36 rows per SKU — simple enough to manage, but powerful enough to capture demand shifts.
Step 2 — Use AI as a coach
I don’t paste data into AI. Instead, I ask questions like: “What’s the best way to forecast monthly SKU sales over the next 12 months?” AI doesn’t crunch my numbers — it provides methods and code I can run locally.
Step 3 — Build forecasts in Python
With AI’s guidance, I run a simple Python script on my machine. It fits a seasonal model to each SKU’s monthly history and projects the next 12 months. The baseline forecast comes from past sales patterns. Festive periods are layered with an uplift factor. Promotion months get a separate adjustment based on past promo performance. This keeps the forecasts rolling, private and realistic — no need for large external tools.
The result
Before: guesswork-driven planning, hours lost in manual spreadsheets, festive peaks often missed or underplanned. Now: rolling 12-month SKU forecasts at monthly level, visibility into both promotions and festive demand, and forecasting that is private, simple and actionable.
The real learning
AI didn’t turn me into a demand planner. It made me a sharper eCommerce manager. AI guides the method, Python delivers the forecasts, and my business context fills the gaps. This isn’t about building complex statistical models. It’s about making forecasting reflect real retail cycles: complex to simple, slow to fast, guesswork to clarity, everyday to festive-ready.