Forecasting fundamentals

A budget is the plan management commits to for a period, usually set once a year. A forecast is the current best estimate of what will actually happen, updated as information arrives. A target is an aspiration. Mixing them up is the commonest cause of bad forecasts: a forecast that is really a target is optimistic by design.

A driver-based forecast models the few operational quantities that actually cause the financial results — the drivers — and derives the figures from them. Greenleaf Kitchen sells meal-kit subscriptions. Its revenue isn't forecast as "last year plus 8%"; it's built from:

• Opening subscribers — a fact, from the subscriber system.
• Churn rate — the share of subscribers who cancel each month (5% of the opening base).
• New subscribers — sign-ups per month. Greenleaf uses a three-month moving average of recent sign-ups: (280 + 310 + 370) ÷ 3 = 320.
• Revenue per subscriber — 160 a month (four boxes at 40).
• Cost behaviour — variable costs at 60% of revenue; fixed costs of 110,000 a month.

Because each output traces to a driver, managers can see why the number moves, update one assumption without rebuilding the model, and run sensitivity analysis: change one driver and watch the result. Drivers that move the answer most deserve the most scrutiny and the most frequent updating.

Good forecasting practice: state every assumption with its source; separate facts from judgments; prefer simple methods you can explain (moving averages, trends adjusted for known seasonality) over complexity no one can check; and roll the forecast forward each month so it always looks a fixed distance ahead (a rolling forecast).