Predict AP Invoice Payment day with Machine Learning in Celonis
There are many areas where Machine Learning can be used in Celonis to improve business processes. An example is improving the on time payment rate in the Accounts Payable process by predicting when the AP invoices will be paid. The predicted payment day can be used to timely identify invoices that are likely to be paid late so that actions can be taken to ensure the invoice will still be paid on time.
The data from paid invoices in the ERP application (e.g. Oracle ERP Cloud) is used as input for the model that is going to predict when the invoice will be paid. All the information that is available about the invoice and the past invoices from the same supplier is used to generate the prediction model with Machine Learning. The data that is available includes the invoice amount, PO invoice or not, holds, number of lines, weekday the invoice is due, indication if certain activities are performed, sequence of specific activities, the number of days between the invoice date and the invoice entry date and the number of days from the payment terms. But also the information from previous invoices from the same supplier is available like the number of late paid invoices and the number of holds. All together more than 25 attributes are used to generate the prediction model. And additional attributes can be easily added when needed.
The data from the ERP application is loaded into Celonis IBC for process mining analysis. The invoices are assigned to five different payment categories: from paid on time until paid very late. In the machine learning workbench in Celonis a part of the data is used as training data set. The training data set is used to automatically generate the model with a machine learning algorithm that is predicting the payment category. A test data set is used to validate the generated prediction model.
For every new invoice that is entered in the ERP application the model is predicting the payment category. This gives an indication if it is likely that the invoice will be paid on time or not. The predictions are available in Celonis IBC analytics and in the Celonis Action Engine to take automatic actions.
By setting up the appropriate signals based on the prediction and other attributes of the invoice in the Action Engine it is ensured that there are signals send to users for those invoices that require direct attention.
With additional data coming available from new paid invoices the prediction model is trained further and improved versions of the model are implemented to further support the process to ensure timely payment of the invoices!