Calculating a forecast means using historical data to estimate future values, whether you’re projecting sales, revenue, demand, or expenses. The most common approach takes past performance, applies a mathematical method to identify patterns, and extends those patterns forward. The specific method you choose depends on how much data you have, whether your numbers follow seasonal patterns, and how precise you need the result to be.
Prepare Your Data First
Every forecast is only as good as the data behind it. Before running any calculations, you need a clean historical dataset that represents normal conditions. Start by plotting your numbers on a simple line chart. This visual step reveals trends, seasonal swings, and outliers (abnormal spikes or drops) that would distort your forecast.
Remove or adjust data points caused by one-time events like a major promotion, a supply shortage, or an accounting error. The goal is to isolate your “baseline history,” the pattern your numbers follow under typical circumstances. If you had a product launch that tripled sales for one month, leaving that spike in the data will inflate your forecast for the same month next year.
Missing data points need attention too. If a month is blank, you can fill it with the average of the surrounding periods or interpolate between them. For most business forecasts, monthly data works well because it smooths out the day-to-day noise you’d see in weekly or daily figures while still capturing seasonal shifts. Aim for at least two full years of monthly data so you can spot recurring patterns.
Simple Moving Average
A simple moving average (SMA) is the most straightforward forecasting method. You add up the values from a set number of recent periods and divide by the number of periods. The formula is:
Forecast = (A1 + A2 + A3 + … + AN) / N
Here, each “A” is the actual value for a period, and “N” is how many periods you’re averaging. If your last three months of sales were $40,000, $45,000, and $50,000, a three-month moving average forecast for next month would be ($40,000 + $45,000 + $50,000) / 3 = $45,000.
Choosing N is the key decision. A shorter window (three or four periods) reacts quickly to recent changes but can be jumpy. A longer window (six or twelve periods) produces a smoother forecast but responds slowly to genuine shifts in your business. If your numbers have been trending steadily upward, a long moving average will consistently underestimate the next period because it’s dragging in older, lower values.
Weighted Moving Average
A weighted moving average improves on the simple version by letting you assign more importance to recent periods. Instead of treating every period equally, you multiply each period’s value by a weight, add the results, and divide by the sum of the weights.
Forecast = (W1 × A1 + W2 × A2 + W3 × A3) / (W1 + W2 + W3)
For example, suppose you want a three-month weighted average and you assign weights of 3, 2, and 1 (most recent month gets the highest weight). With sales of $40,000, $45,000, and $50,000 (most recent last), the calculation is: (1 × $40,000 + 2 × $45,000 + 3 × $50,000) / (1 + 2 + 3) = $280,000 / 6 = $46,667. The forecast tilts toward the most recent month, which is useful when your business conditions are changing and older data is less relevant.
Exponential Smoothing
Exponential smoothing takes the weighted concept further by applying a single “smoothing constant,” typically called alpha (α), that controls how heavily the forecast leans on the most recent actual value versus the previous forecast. The formula is:
New Forecast = α × (Most Recent Actual) + (1 − α) × (Previous Forecast)
Alpha is a number between 0 and 1. A higher alpha (like 0.7 or 0.8) makes the forecast respond quickly to recent changes. A lower alpha (like 0.1 or 0.2) produces a more stable, slower-moving forecast. If last month’s actual sales were $50,000, your previous forecast was $44,000, and you set α at 0.3, your new forecast would be: 0.3 × $50,000 + 0.7 × $44,000 = $15,000 + $30,800 = $45,800.
This method is popular because it requires minimal data storage. You only need the last actual value and the last forecast to calculate the next one. It also naturally gives diminishing weight to older data without requiring you to manually set weights for every period.
Adjusting for Seasonality
If your business has predictable peaks and valleys (holiday rushes, summer slowdowns, back-to-school spikes), you need to layer a seasonal adjustment on top of your baseline forecast. Without it, your forecast will underestimate busy months and overestimate slow ones.
The process has three stages. First, calculate a centered moving average across a full year of data. This strips out the seasonal pattern and shows the underlying trend. Second, divide each period’s actual value by the moving average for that period. This ratio tells you how much higher or lower each month typically runs compared to the average. Third, average those ratios for each month across multiple years, then rescale them so they sum to exactly 12 (for monthly data) or 4 (for quarterly). The result is a seasonal index for each period.
For example, if December’s seasonal index is 1.35, that means December sales are typically 35% above the monthly average. If July’s index is 0.80, July runs about 20% below average.
To apply the index, generate a baseline forecast using one of the methods above, then multiply by the seasonal index for the target period. If your baseline forecast for next December is $60,000 and the seasonal index is 1.35, your seasonally adjusted forecast is $60,000 × 1.35 = $81,000.
Measuring Forecast Accuracy
Once you’ve produced forecasts, you need to check how close they land to actual results. Three metrics cover most situations.
Mean Absolute Error (MAE): For each period, subtract your forecast from the actual value and ignore the sign (take the absolute value). Then average those differences across all periods. If your forecasts missed by $2,000, $5,000, and $3,000 over three months, the MAE is ($2,000 + $5,000 + $3,000) / 3 = $3,333. This tells you the average size of your errors in the same units as your data (dollars, units sold, etc.). You may also see this called Mean Absolute Deviation (MAD), which is the same calculation.
Mean Squared Error (MSE): Same idea, but you square each error before averaging. Squaring penalizes large misses much more heavily than small ones. An MSE of 25,000,000 (in squared dollars) is harder to interpret on its own, but it’s useful for comparing two forecasting methods. The one with the lower MSE is doing a better job avoiding big errors.
Mean Absolute Percentage Error (MAPE): This expresses each error as a percentage of the actual value, then averages those percentages. If actual sales were $50,000 and your forecast was $47,000, the error for that period is $3,000 / $50,000 = 6%. MAPE is the most intuitive metric because a result like “our forecast is off by an average of 8%” is easy for anyone to understand. It also lets you compare accuracy across products or business units with very different sales volumes. One caution: MAPE breaks down when actual values are zero or very close to zero, since dividing by a tiny number produces misleadingly large percentages.
Choosing the Right Method
Your choice of forecasting method should match the characteristics of your data and the decisions you’re making with the forecast.
- Stable data with no strong trend: A simple moving average works well and is easy to explain to stakeholders. Use a three- to six-period window.
- Recent changes matter more than older history: A weighted moving average or exponential smoothing lets you emphasize the latest data. Experiment with different weights or alpha values and compare the MAPE of each version.
- Clear seasonal patterns: Calculate seasonal indices and apply them on top of your chosen base method. Skipping this step is the single biggest source of forecast error for businesses with cyclical demand.
- Long-term strategic planning: Consider combining quantitative results with qualitative judgment, such as factoring in a planned product launch or market expansion that historical data can’t capture.
Start with the simplest method that fits your data. Run it against a portion of your historical data you held back (this is called backtesting), measure the error, then try a more complex method and see if accuracy improves enough to justify the extra effort. A moving average you actually maintain and review each month will outperform a sophisticated model that sits unused in a spreadsheet.