Client behaviour sometimes follows regular patterns in time. For example many businesses may purchase consumables regularly (e.g. once a week or once a month) but this buying pattern varies from business to business based on their own usage patterns. If we consider a company, (Company A) which uses soap in their washrooms then they will probably use soap at a different rate than another company (Company B).
Facilities management companies are often responsible for the management of premises, including consumables such as soap or hand towels. In the example of soap this would generally be re-order when on-site staff notice supplies are low. This is, however, inefficient as deliveries are driven by on demand orders which makes it difficult to optimise delivery costs. For example a delivery may be made to a premises on a Monday only for an order to be placed by another premises on the same street on a Tuesday. If the supplier could predict the second premises was also in need of re-supply then a trip could be saved.
Our solution takes order or delivery data and attempts to recognise repeat orders. This is done by forecasting the usage rate based on the the quantity and time period between orders for each individual customer and customer site. The algorithm then predicts when the customer site will run out of consumables.
These predictions are then fed to another optimisation algorithm which plans the best re-supply strategy based on when and where re-supply is required.
The implementation uses clustering techniques as well as genetic algorithms.