Part 2 · Decision 4 · Weeks 8–9

"What should we make, and how much?"

Every operations decision downstream — how much to staff, how much to stock, when to expand — rests on a number you can't actually know: future demand. Forecasting is how you make that number defensible. S&OP is how you turn it into a plan everyone agrees on.

Moving Average Exponential Smoothing Seasonality S&OP

Why this decision matters

"Forecast" sounds like a math exercise; it's actually a coordination problem. Sales wants high numbers so they're not constrained. Operations wants low numbers so they're not stuck with stock. Finance wants the one that makes the quarter. S&OP is the monthly ritual where those three sit at the same table and agree on one number — and forecasting gives them the rigor to argue about it instead of guessing.

By the end of this topic you'll be able to

Choose the right forecasting method for a time series (level, trend, seasonal, intermittent). Calculate moving averages and exponential smoothing by hand, and tune the alpha to your data. Detect bias with a tracking signal. Run a chase-vs-level production plan. Explain the S&OP cycle and why misalignment between sales and operations is so expensive.

Materials

Key concepts to know
  • Naive forecast — tomorrow = today. The benchmark every model has to beat.
  • Moving average — average of last n periods. Smooths noise, lags trend.
  • Weighted moving average — recent periods weighted heavier.
  • Exponential smoothing — Ft+1 = α · At + (1 − α) · Ft. Alpha ≈ 0.1–0.3 for stable demand, higher for reactive.
  • Trend & seasonality — double exponential smoothing (Holt) for trend, Holt-Winters for both.
  • Forecast error — MAD, MAPE, MSE. Pick one and stick with it.
  • Tracking signal — cumulative error ÷ MAD; if >±4, the model is biased.
  • Chase vs. level production — chase changes capacity to match demand; level keeps capacity steady and buffers with inventory.
  • S&OP cycle — data → demand review → supply review → pre-S&OP → executive S&OP. Monthly.
Class notes & cheat sheets
Hands-on activity — Tim Hortons demand

Two weeks of work in one dataset: the Tim Hortons weekly sales file lets you build moving averages, fit exponential smoothing, detect seasonality, and propose a chase-vs-level production plan. The EireGreenWorks file is the S&OP companion.

Practice with games · Forecasting
Using AI on this decision

Modern AI is genuinely strong at time-series classification (level vs. trend vs. seasonal vs. intermittent) — paste a column and ask. It's reasonable at suggesting methods, weaker at picking alpha, and unreliable at point forecasts unless you make it write the formula and run it in a spreadsheet. The S&OP cycle itself is where AI shines: summarize the demand review for the supply meeting, draft the executive deck, surface assumptions worth challenging.

The AI Lab walks through forecasting the Tim Hortons dataset with an AI assistant.

Stay Ahead of the Curve

Subscribe to our bi-weekly newsletter for the latest insights on AI, data, and business strategy.