How to Forecast Sales for a Small Business, Without a Data Scientist

Start with a baseline you have to beat

The simplest forecast says next month will look like last month. For a seasonal business, a better simple version says next month will look like the same month last year.

Keep that baseline, and hold every fancier method to beating it. A forecasting method that can't outperform "same as last month" is adding complexity and no accuracy.

Smoothing out the noise

Monthly sales bounce around for reasons that won't repeat: a big one-off order, a rainy week. Averaging the last few months smooths some of that out, but a plain average treats a month from a year ago the same as last month.

Exponential smoothing weights recent months more. Each new forecast is a blend: a share of the latest actual sales plus the rest of the old forecast. With a blending share of 0.3, a forecast of 1,000 that met actual sales of 1,120 becomes 0.3 × 1,120 + 0.7 × 1,000 = 1,036.

That share is the one setting to choose. A high value reacts quickly to change and also chases every random bump. A low value stays steady and is slow to notice a real shift. Try a few against your own history and keep the one whose past forecasts would have been closest.

Seasons and trends

If December usually sells 1.6 times an average month, a smoothed forecast of 1,036 for an average month means about 1,658 for December. That 1.6 is a seasonal index, and you can work one out for each month by dividing its sales by the average month across a year, ideally averaged over two or more years.

The cleanest method strips the seasonality out first: divide each month's sales by its index, smooth the adjusted series, then multiply the forecast by the index of the month you're predicting. Otherwise the smoothing mistakes every December for a boom and every January for a slump.

Plain smoothing also lags behind a steady trend. If sales grow every month, the forecast will run low month after month. Methods that track a trend alongside the level fix that, and so does adding the recent growth rate by hand, as long as you check whether the trend is still there.

Measure how wrong you are, and in which direction

Record each month's forecast and compare it with what happened. The average size of the misses, ignoring whether they were high or low, tells you how accurate the forecast is. The average as a percentage of sales makes it comparable across products: misses of 50 units in a month that sold 1,000, 30 in one that sold 800 and 90 in one that sold 1,200 are errors of 5%, 3.8% and 7.5%, an average of about 5.4%.

Direction matters as much as size. Add up the errors with their signs and divide by the average miss, and you have a tracking signal. If it drifts far from zero, a common rule of thumb flags it beyond about 4, the forecast is consistently high or consistently low, which usually means something changed that the method hasn't caught.

Forecast a range, at the level that holds together

A single number hides how uncertain it is. Forecast a likely value and a range around it, and use the range for decisions: order enough stock for the high end when running out is expensive, and staff for the middle.

Totals are easier to predict than details. Forecasting total monthly sales and then splitting it by the usual product mix is often more accurate than forecasting every product separately, because individual items swing more than the whole.

For items that expire or go out of date, the right order balances the cost of having too many against the cost of having too few. That problem has its own neat solution, known as the newsvendor model, and a calculator for it is linked below.

Adjust with judgment, and write down why

You'll know things the history doesn't: a promotion next month, a new competitor, a large customer's contract ending. Adjust the forecast for them, and note the reason next to the number.

When the month closes, check whether the adjustment made the forecast better or worse. Over a year that record shows which of your instincts are worth trusting, which is valuable in its own right, since demand forecasting errors feed straight into inventory and staffing.

Key takeaways

• Keep a simple baseline, such as last month or the same month last year, and make every method beat it.
• Exponential smoothing blends the latest sales with the old forecast; the blending share trades responsiveness against stability.
• Remove seasonality before smoothing and multiply it back in afterward, and watch for trends that smoothing lags behind.
• Track both the size and the direction of errors; a forecast that's consistently high or low needs its method fixed, and a wider range won't do that.

Frequently asked questions

How far ahead should a small business forecast?

At least as far as your longest commitment: the lead time on your stock, the notice period for hiring, the term of a lease you're considering. Monthly forecasts for the next 12 months, updated each month, suit most small businesses.

What if I don't have much sales history?

Start with what you have, use comparable products or local competitors as rough guides, and widen the range you plan for. Forecasts from short histories improve quickly once you start recording actual results against them.

Which forecasting method is best?

The one that has been most accurate on your own data. Test a few simple methods against your past sales, compare their errors, and keep the winner, then check periodically whether it still wins.