Demand forecasting

Demand forecasting is predicting how many of each item will sell, in each place and in each period, so that the stock, the staff and the prices are ready before the orders arrive. The forecast may be for one kind of bread in one shop tomorrow morning, or for every television a company will sell across India next October. Retailers, delivery apps, hotels and factories all depend on it. Goods take days or months to make and move, but customers expect them at once.

Here is how it works in general. The business starts with its past sales, recorded item by item and day by day. Data recorded like this, one reading for each period, is called time series data. A model looks for patterns in that history: the usual level of sales, the pattern within each week, and the seasons, such as the rush before Diwali or the drop during the monsoon. It then adds what is already known about the coming period: a festival, a planned discount, a cricket final or the weather forecast. The result is not one number but a range, for example "between 80 and 120 packets, most likely 100". A planner, or the software itself, then decides how much to order or make. When the real sales come in, they are compared with the forecast, and the model is corrected.

Why a range? Every forecast is wrong by some amount. A shop that stocks exactly the most likely number will run out on many days. So it keeps some extra stock, called safety stock, and the more uncertain the forecast, the more extra stock it keeps. Too little stock loses sales. Too much stock ties up money, and fresh food or fashion that does not sell goes to waste.

Consider a quick commerce app. It forecasts, hour by hour, how much milk and bread each small neighbourhood warehouse will sell, because a ten-minute delivery is possible only if the item is already on that shelf.

For instance, a hotel's revenue management system forecasts how many rooms will be wanted on each night of the coming months, and raises the price for the nights that are likely to sell out.

Imagine a clothing chain ordering kurtas months before a festival. Its merchandise planning software forecasts the sales of each store and each size.

How is demand forecasting different from supply-chain planning? The forecast says what customers are likely to buy. The plan decides what the business will do about it: how much to make, which parts to buy and where to keep the stock. The plan starts from the forecast, but it must also fit the factory's capacity and the money available. Dynamic pricing uses the forecast the other way round: instead of changing the stock to meet demand, it changes the price.