Forecast the next 1 to 6 weeks or months of demand for one product or category from your own history. Enter 4 to 24 past periods, choose a 3-period moving average, a weighted average or exponential smoothing, and add a seasonal uplift for a busy period. The tool shows every step of the arithmetic and compares the three methods' errors, so you can see which one would have tracked your history best before you print or download the forecast.
Make a demand forecast
- Name the product or category. Add the unit you count in, such as bags, cases or dollars of sales, so the results are labeled.
- Choose weeks or months, then enter 4 to 24 periods of past demand, oldest first. To copy a column from a spreadsheet, paste it into the paste box and select Replace history.
- Choose a forecast method. The weights and alpha under Method settings apply to their own methods and to the comparison table.
- Choose how many periods ahead to forecast, from 1 to 6. Enter a seasonal uplift for any period you expect to be busier or quieter, such as 15 for a promotion week or -10 for a slow month.
- Review the forecast table, the method comparison and the history table, then download text or CSV, or print.
Enter demand, not only what you sold. If a product ran out for part of a week, sales understate what customers wanted. Adjust that week, or leave it out and start the history after it.
The three forecast methods
None of these methods projects a trend or a seasonal pattern on its own. Each produces one base forecast, and every period ahead starts from it. Only the uplift you enter changes a single period.
Moving average forecast
Base forecast = (last 3 periods of demand) ÷ 3
Each period counts equally, so one unusual week moves the forecast by a third of its difference. The NIST/SEMATECH e-Handbook of Statistical Methods explains the single moving average and its limits.
Weighted moving average
Base forecast = (w1 × latest + w2 × one before + w3 × two before) ÷ (w1 + w2 + w3)
You choose the three weights, most recent first. They are divided by their total, so they do not need to add up to 1 or 100. The default of 3, 2 and 1 makes the latest period half of the forecast, so it reacts sooner when demand changes.
Exponential smoothing forecast
Next forecast = alpha × latest demand + (1 − alpha) × latest forecast
The first forecast is period 1's demand, and each later forecast blends the newest demand with the previous forecast. Alpha must be above 0 and no more than 1. A high alpha follows recent demand closely; a low alpha smooths out noise. This matches the single exponential smoothing formulation in the NIST handbook. The open textbook Forecasting: Principles and Practice explains why simple exponential smoothing gives the same forecast for every period ahead.
Check forecast accuracy with mean absolute error
For each past period, the tool works out what the method would have forecast using only the periods before it, then measures the gap to actual demand, ignoring whether it was over or under. Mean absolute error (MAE) is the average of those gaps, in your unit. Lower is better.
Periods 1 to 3 start each method, so all three are scored on period 4 onward. That keeps the comparison fair when you switch methods or change the weights or alpha. The same textbook describes MAE among other measures of forecast error. Treat a lower MAE as a guide, not proof: settings tuned to fit last quarter perfectly can fit next quarter worse.
Worked example: 12 weeks of coffee
A fictional shop sold these 250 g bags of house blend coffee over 12 weeks: 42, 38, 45, 40, 44, 47, 41, 46, 43, 48, 45 and 50. Scored on weeks 4 to 12, the three methods compare like this:
The moving average has the lowest error, so the sample uses it. Its base forecast is (48 + 45 + 50) ÷ 3 = 47.67 bags. Week 15 has a planned promotion with a 15% uplift: 47.67 × 1.15 = 54.82 bags. Weeks 13, 14 and 16 stay at 47.67, for a total of 197.82 bags over the next 4 weeks.
What the forecast cannot see
The forecast knows only the numbers you enter. It cannot see a price change, a new competitor, a product launch or a holiday unless you add an uplift for it. The forecast check under the preview flags two common problems: demand that rose or fell by 20% or more from the first 3 periods to the last 3, which these methods lag behind, and a history where a third or more of the periods had no demand at all. Both are hints; they never stop you printing or downloading.
If you enter dollars of sales instead of units, the same arithmetic works as a sales forecast calculator. To compare two finished periods instead, use the sales growth calculator.
Use the forecast for ordering
A forecast becomes useful when it changes an order. Divide a weekly forecast by your trading days to get the average units sold per day, then use that figure in the reorder point calculator, the safety stock calculator and the days of supply calculator. For the EOQ calculator, multiply a weekly forecast by 52 or a monthly one by 12 for annual demand, if the year is likely to look like the weeks you forecast. A large mean absolute error is a reason to hold more buffer stock.
Simple forecast vs demand planning software
This is a simple statistical forecast for one series at a time, not demand planning software. Dedicated demand forecasting tools can read sales history automatically, forecast many products at once, model seasonality and trends, and feed purchase suggestions. If you forecast a handful of key products by hand each week, this tool is often enough. If you forecast hundreds, a system built for it will save time.
Print or download the forecast
Print the forecast to keep beside your ordering notes, or download a text copy with the working or a CSV file for a spreadsheet. The CSV holds one row per history period and per forecast period, with the mean absolute error between them. Your draft stays in this browser tab through a refresh. Nothing is uploaded.
Frequently asked questions
Can I forecast demand without an account?
Yes. Enter your history and the forecast updates as you type. Nothing is saved to an account or sent anywhere, so download or print a copy to keep.
Which forecast method should I use?
Start with the method that has the lowest mean absolute error on your own history, then check it against what you know. A weighted average or a higher alpha suits demand that shifts often. A plain moving average or a lower alpha suits steady demand with random noise.
Why is every forecast period the same?
These methods estimate the current level of demand and carry it forward. They do not extend a trend, so each period ahead gets the same base forecast until you add an uplift.
How much history do I need?
At least 4 periods, so each method has one period to be scored on. Eight to 12 periods give a fairer comparison. Use the same period length throughout, and leave out periods affected by stockouts or closures.