Forecasting

Forecasts are planning instruments, not prophecies: compute the baseline, decompose the pattern, express ranges, and measure yourself honestly.

When to use

• Business planning (revenue, cost, headcount, capacity, demand).
• Operating forecasts and re-forecast cycles.
• Personal financial planning over time.

Method

1. Purpose & horizon: operational (weeks) vs strategic (years); forecast to the decision cadence — daily forecasts for daily decisions, not everything for everyone.
2. Baseline methods always: naive (last value), moving average, seasonal-naive — computed and shown; complexity must beat simplicity to earn its place.
3. Decompose: level/trend/seasonality; and prefer driver-based forecasts where drivers are known (units × price, loads × rates, headcount × costs) over blind extrapolation.
4. Ranges, not points: prediction intervals or low/base/high scenarios — a point forecast to leadership without a range is false precision.
5. Scenario & sensitivity: base/best/worst with trigger conditions (what signal tells you you're in the worst case) — feeds budgeting.
6. Accuracy tracking: measure per series (MAE/MAPE) and bias — consistently under-forecasting is a process bug, not bad luck; publish scorecards so trust is earned.
7. Revision hygiene: keep forecast-vs-actual history; document why revisions happened; never silently restate old forecasts ("we always said!" is a lie the spreadsheet tells).
8. Cadence: monthly rolling forecasts (12-month) typical; weekly for operations; annual for strategy — matched to the review rhythm.

Quality gates

• Baseline computed and compared; decomposition/drivers shown.
• Ranges + scenarios delivered, not bare points.
• Accuracy (incl. bias) tracked per series with a scorecard.
• Revision reasons documented; history preserved.

Red lines

• No hindsight edits; no forecasts without owners; no single numbers for uncertain futures without ranges.

Related

prediction · budgeting · revenue-management (forecast inputs).