Exhibitors & Products
Events & Speakers

"Every machine builder has an ERP. We even have a history of orders. Each order has information on the date of the order, confirmation by the supplier and delivery date. Many simply forget the data when the order arrives," Günther explains. But this data is very well maintained, he continues. So the Inform developers took that data – looking three years into the past – and used machine learning to first analyze supplier behavior and then develop a predictive model. The amount of data: Up to 50,000 orders per year over three years. That sounds simple. "The complexity of the data was low. But we had to find out which data was relevant for the order," Günther looks back. What data affects orders – when was the order placed (fall or winter), how valuable was the order (large orders went better), or which buyer was responsible.

Inform used a Random Forest approach for the evaluation. A Random Forest is a classification and regression procedure that consists of multiple uncorrelated decision trees. All decision trees have grown under a certain kind of randomization during the learning process. "After all, we wanted to improve the broad mass and not just one order," says Günther, explaining the use of the decision tree. With the help of Inform software, errors in estimating replenishment lead times can be reduced by up to 42 percent. "The customer now has up-to-date forecast dates. You can't lure anyone to the stand with that at the trade fair now. A table tennis robot has more of an effect," jokes the Inform manager. But, customers can now approach purchasing and analyze discrepancies faster. And: There is no need to review tens of thousands of materials. If these results are now used with a production planning tool, the user can make real optimizations.