The Norwegian newspaper Dagbladet is sold daily at thousands of salespoints. It is difficult in advance to determine the exact number of newspapers that will be sold at each salespoint. Each salespoint is therefore supplied with a few more copies that it is expected to sell. This can lead to surplus copies at the end of the day, yielding large costs for handling the returns. On the other hand at some of the salespoints Dagbladet will be sold out, leading to lost potential sales. Accurate predictions of the sale at each salespoint will reduce both the cost of returns and the loss of potential sales.

Our solutions are built on statistical models describing the sale at each salespoint as a function of time. The expected sales are modelled as a function of day of the week, time of the year, holidays (Christmas, Easter, etc.) together with a trend over time. Important statistical techniques used are time series analysis, multivariate and nonlinear regression including generalised additive models.

Optimal distribution as described here can be generalised to other types of products with a limited shelflife, e.g. fresh bread and milk.