Every business owner forecasts. Most do it by gut: "summer's usually slow, December's big, I think next year's up a bit." The gut is often decent, because it's been watching the business for years. But the same history the gut runs on is sitting in your invoices and your point of sale as actual numbers, and the math to turn it into a real forecast is high-school level. No machine learning, no data scientist, no expensive platform. Here's what forecasting from your own data looks like when it's kept simple and honest.
Trend and seasonality: the two things your history is made of
Take any business's monthly revenue and plot it as a line chart. Two patterns account for most of what you'll see.
Trend is the long direction: growing, shrinking, or flat, once you squint past the wiggle. Excel and Google Sheets will draw it for you: right-click the chart series and add a trendline. A linear trendline through 24 months of revenue is already a crude forecast, because you can extend it forward and read next quarter off the line.
Seasonality is the wiggle that repeats. An HVAC shop peaks in summer, retail peaks in Q4, B2B services crater every December. To see yours in numbers, compute each month's average share of annual revenue across a couple of years. If July reliably does 11% of the year and February does 6%, that's not noise, that's a pattern you can plan against. A gut says "summer's busy." The data says July runs 35% above baseline, which is a number you can staff and stock to.
The honest recipe for a simple forecast is exactly those two pieces: extend the trend, then reshape it with the seasonal pattern. Estimate next year's total from the trend, spread it across months using each month's historical share. That two-step forecast, done in an afternoon in a spreadsheet, beats gut feel and beats doing nothing, and it's transparent enough that you can explain every number in it to your banker.
Honest error bars, or: a forecast is a range
Here's the part most forecasts skip, and it's the part that makes them trustworthy. A single number ("next quarter will be $412,350") is false precision. Nobody knows that. The honest version is a range: "we expect $380k to $445k, centered around $410k."
You don't need statistics coursework to get a usable range. Backtest yourself: apply your forecasting method to the past, as if you were standing a year ago, and see how far off it would have been. If your method historically misses by around 8%, then your forecast deserves a band of roughly plus or minus 8%. Spreadsheet forecast functions (Excel's FORECAST.ETS family, for one) will even produce confidence intervals for you, which is the same idea done more formally.
Ranges aren't hedging, they're operational information. If the bottom of the range still covers payroll and rent, you can breathe. If the bottom of the range doesn't, you know that now, in time to do something, instead of in the month it happens. A forecast's job is not to predict the future. Its job is to tell you which futures to prepare for.
Simple methods before ML, every time
It's 2026 and every tool wants to sell you AI forecasting. Our position, from doing this work: a small business should exhaust the simple methods first, and most never need to go past them.
- Simple wins on data size. Machine learning models are hungry. Three years of monthly figures is 36 data points, which is a rounding error to an ML model and plenty for trend-plus-seasonality.
- Simple is explainable. When your forecast says Q3 is soft, the trendline-and-seasonality version lets you point at exactly why. A black-box model gives you a number and a shrug. You can't make staffing decisions on a shrug.
- Simple fails loudly. When a simple forecast goes wrong, you can find the broken assumption and fix it. When a complex one goes wrong, you mostly just stop trusting it.
- The ceiling is the data, not the math. Fancier math squeezes a few percent more accuracy out of the same history. Cleaner, longer history improves any method a lot more. Spend there first.
ML earns its keep at a different scale: forecasting demand for thousands of individual products, or when outside signals like weather genuinely drive your sales. If that's you, you'll know, because the simple method will be visibly leaving money on the table.
What 12+ months of clean history unlocks
The gate to all of this is boring: twelve or more consecutive months of consistent numbers. Revenue by month, at minimum. Revenue by month by product line or location is better. "Clean" means the categories mean the same thing across the whole period and there are no gaps or mystery lumps ("that spike is when we changed systems and double-counted a quarter").
Twelve months gets you one full seasonal cycle, which is the minimum to see the pattern. Twenty-four months is where it gets good, because you can confirm the pattern repeats and separate it from the trend. If your history is a mess, cleaning it back two years is usually a recoverable job, and it's step one of every forecasting project we do.
You'll know your forecasting is working when three things are true: you write the forecast down before the period starts, you compare it to actuals after, and the range contains reality most of the time. Miss those and it's astrology with a spreadsheet. Hit them and you've got an instrument you can plan hiring, inventory, and cash around. If your sales history is trapped in a system you can't report from, that's the unglamorous first step, and it's the kind of thing we dig out for people regularly.
Stuck on this, or want it done for you? That's the job.
Email us →