Most business revenue forecasts are last year's number plus a hopeful percentage. A statistical forecast is not much harder to build, is far more defensible, and - crucially - comes with a range rather than a single figure you will inevitably miss.
Step 1: Get clean monthly history
You need at least 24 months of revenue by month, ideally 36 so you can see seasonality repeat. Use consistent definitions throughout - if you changed how you recognise revenue in month 14, fix it or note it. Garbage history produces garbage forecasts, and no method rescues that.
Step 2: Decompose the series
Every revenue series is made of three things:
- Trend - the underlying direction
- Seasonality - the repeating monthly or quarterly pattern
- Noise - everything else
Plot the data. You will usually see all three by eye. Calculate a 12-month moving average to strip out seasonality and reveal the trend, then compare each month against that average to estimate its seasonal factor.
Step 3: Project the trend
Fit a simple linear trend to the deseasonalised data - a spreadsheet trendline is genuinely adequate for most businesses. Extend it forward for your forecast period. Resist the urge to fit a curve that hugs your history perfectly; that is over-fitting, and it forecasts badly.
Step 4: Reapply seasonality
Multiply each projected month by its seasonal factor. You now have a monthly forecast that reflects both direction and pattern - already substantially better than most business plans.
Step 5: Add the range, which is the whole point
Look at how far your model's fitted values missed actual history each month. That spread is your realistic uncertainty. Express the forecast as a range: "£182,000 next quarter, most likely between £165,000 and £199,000."
This is the step people skip and the reason forecasts lose credibility. A single number is always wrong. A range with a stated confidence level is a planning tool - it tells you what to prepare for if things go badly, which is exactly what a board needs.
Step 6: Adjust for what the data cannot know
Statistics extrapolates the past. It does not know about the contract you signed last week, the competitor entering next month or the price rise you are planning. Apply judgement adjustments on top of the statistical baseline, and document each one separately so you can see later whether your judgement or your model was at fault.
Step 7: Track your forecast error
Record forecast versus actual every month. After six months you will know your typical error, which makes every subsequent forecast more honest. Businesses that do this consistently develop a genuine feel for their own predictability.
What to avoid
Forecasting from too little history, projecting an unsustainable growth rate indefinitely, ignoring seasonality, and presenting a point estimate with no range. Also beware forecasting revenue without forecasting the cash flow that follows it - profitable businesses fail on timing.
The Statistics for Business course at London School of Business UK covers time series and forecasting applied to business data. Enquire today.