Electrical wholesaler automates preliminary account assignment

Electrical wholesaler Sonepar in Germany wanted to automate internal processes in accounting and create organizational synergies. Around 5000 employees work for the company, which was founded in 1972. The solution-oriented electrical wholesaler for tradesmen and industrial supplies stocks around 500,000 articles from over 2000 brand manufacturers. 20 percent of these articles make up the core range, which is kept in stock at all times. 

Due to the rapid company growth, a very high volume of documents was created at Sonepar in a short time. The company has been using SAP in the areas of accounting and controlling since 2012. In the course of the recent S/4 transition, incoming invoice processing was also to be optimized. FIS took over the conversion of the accounting and controlling areas at Sonepar from ECC 6.0 to S/4 Hana with almost 500 million documents and a volume of 2.4 billion document lines. At the same time, the FIS Invoice Management solution was also introduced in order to optimize the invoicing process right from the start. Machine learning services for intelligent account assignment and approver determination round off the portfolio.

Sonepar decided in favor of the FIS solution because the processing of incoming invoices was to be made more efficient and, above all, digital. This is because the clerks process more than two million incoming invoices every year. Here, the application of FIS ensures a high degree of automation. In the area of preliminary account assignment, there is an optimal use case for artificial intelligence. For each document, the clerks have to enter data into the system to determine the company code, vendor, G/L account, cost center and the responsible approver.

Machine Learning (ML), as a subfield of AI, is a self-learning technology and can enable intelligent account assignment and approver determination in the context of accounts payable cost accounting. Both the management and the IT managers at Sonepar had a keen interest in implementing automatic preliminary account assignment of cost invoices using ML in a pilot project. Here, the company code, vendor and text layer are read from the invoices. Based on this information, the AI automatically selects the G/L account, cost center, internal order and approver.

To train the first models, learning data from 36,000 calculations was fed in. This is because the system had to learn how to proceed with each individual calculation. The implementation is done with Tensorflow, Google's neural network. "We trained the neural network with historical booking data from the past. For example, who posted what to which cost center, who was the first approver, and so on. So ultimately we automated this manual process using ML."says Daniel Stemig, Team Leader FIS EIM Consulting.

Based on this, the resulting model makes predictions for future invoices and automatically assigns them if the information is correctly assigned by the AI with a predefined probability. The user can intervene at any time if necessary. In Sonepar's case, an invoice is preallocated in a completely automated way only if the ML system achieves more than an 80 percent probability of the proposed account assignment objects. With the help of the application, different document types can be processed, such as cost invoices, but also goods invoices. Everyone was enthusiastic about the initial results of the pilot project. And the system is constantly learning.