What a forecast is for
A small producer, whether a vegetable grower, a bakery, a candle maker or a brewer, has to decide in advance how much to make. Make too little and customers leave empty-handed. Make too much and the surplus spoils, sits in storage or gets sold at a discount. A demand forecast is an estimate of how much customers are likely to buy over a coming period, built mostly from what they bought before.
A forecast is not a promise. It is a way to turn a gut feeling into something written down that can be checked later.
Start with the sales history
The raw material is a record of what sold and when. Even a spreadsheet with one row per sale is enough to start. Useful details include:
- The date of each sale, or at least the week.
- The product and quantity.
- The sales channel, such as a farmers’ market, wholesale customers or an online store.
- Notes on unusual days, such as bad weather, a holiday or a stockout.
That last point matters. If you ran out of an item by noon, the sales figure understates demand, because customers who arrived later could not buy. Recording stockouts lets you treat those days differently.
Patterns worth looking for
Past sales often show a few regular patterns.
- Seasonality. Some products sell more at certain times of year, such as soup in winter or salad greens in summer.
- Weekly rhythm. Weekend markets may be busier than weekdays.
- Trend. Sales may be slowly rising or falling over many months.
- Events. Holidays, local festivals or promotions can cause spikes.
Simply plotting weekly sales on a chart often reveals these patterns faster than any software does. Charting tools described under data visualization can help here.
A simple worked example
The numbers below are hypothetical. Suppose a small bakery sold the following loaves on the last four Saturdays: 80, 90, 85 and 85.
The average is (80 + 90 + 85 + 85) divided by 4, which is 340 divided by 4, or 85 loaves. A simple forecast for next Saturday is therefore about 85 loaves.
Now suppose a local festival is happening next Saturday, and the bakery remembers that during last year’s festival sales were roughly one fifth higher than a normal Saturday. One fifth of 85 is 17, so the adjusted estimate is 85 plus 17, or 102 loaves. The producer may bake somewhere near 100, and decide how much extra to risk based on how easily leftovers can be sold, frozen or donated.
Notice that the adjustment came from human knowledge of the calendar. Many software tools do the same kind of arithmetic with more data and more patterns, but the logic is similar: start from a baseline, then adjust for known factors.
Where software and machine learning come in
Forecasting features in inventory, point-of-sale and farm-planning software generally work in one of a few ways:
- Moving averages like the one above.
- Methods that weight recent data more heavily.
- Models that find seasonal patterns and combine them with outside data such as weather.
- More complex machine learning models, which can handle many products and factors at once.
For a producer with a handful of products and a few years of sales at most, a simple method is often as useful as a complex one, and it is easier to understand. Complex models need more data than a small business may have, and they can produce confident-looking numbers from thin evidence. Retailers face a similar issue with inventory forecasting, and the same caution applies.
Common ways forecasts go wrong
- Too little history. One season of data cannot show a seasonal pattern.
- Missing context. The tool does not know about a new competitor, a road closure or a social media mention.
- Untracked stockouts. Sales records reflect supply, not demand.
- Perishable items. The cost of overproducing a product that spoils differs from the cost for one that keeps. The best quantity is not always the average.
- Ignoring uncertainty. A forecast is better thought of as a range. If recent Saturdays varied from 80 to 90, plan for some variation.
Using a forecast responsibly
Compare each forecast with what actually happened, and keep a note of the gap. If the estimates are repeatedly high or low, adjust the method. Treat the forecast as one input beside your own knowledge of customers and suppliers. Start with one or two products rather than the whole range, and judge success by fewer leftovers and fewer empty shelves rather than by how advanced the tool is. Outcomes depend on the market, and no forecasting method can remove the uncertainty of demand.