Processing documents automatically

The machine manufacturer Muller Martini is also confronted with these challenges. Procurement officer Felix Willers explains: "Checking and repetitively transferring data from an order confirmation has become much more time-consuming. For example, because of the Ukraine war and COVID, there are now significantly more supply bottlenecks, delivery delays, and price adjustments. We have to adjust these in our orders, as well as in our master data, in SAP, to inform our internal and external customers in time and keep our data in the system up to date."

Database structure and AI developed in-house.

Software that promises to automatically capture documents such as invoices is not new. So far, these systems mostly work based on templates and rules. However, this approach often works poorly. New technologies promise a remedy. Tim Beck, co-founder of BLP, explains: "Thanks to neural networks, we can read new documents without any templates or rules. Especially for items listed in tables, this is a crucial breakthrough."

However, just because an AI understands what is written on the document does not mean that the process can also be automated. Beck continues: "The information on the documents often looks completely different than in the ERP. But an order confirmation for 100 CHF has to match an order for 110 CHF with a 10 CHF discount." To solve the problem, BLP has developed more neural networks that can handle this translation from document information to ERP data. BLP Digital has written its own database structure for this purpose. All data is first extracted from the ERP systems and fed into the BLP structure. This brings a considerable advantage: the AI can be trained for all customers in the same way, which delivers significantly better results because the AI has more data to work with.

The BLP software is connected to the customer's ERP system. Incoming documents from customers and suppliers are read directly by the AI and compared with the ERP data. The document is automatically completed if everything is correct within set tolerance limits. The rest has to be done by employees, for whom personalised dashboards and workflows are available in BLP's browser solution. On balance, companies save an enormous amount of time, reduce sources of error, and counteract the shortage of skilled workers. Beck clarifies: "Time savings must always be measured over the entire process, i.e., as a combination of full automation rate and accelerated decision-making by employees. The mere recognition rate falls short here."

Significant progress thanks to ETH development

The recently completed Innosuisse project with ETH was initiated to further improve document reading. Researchers from the Information Management Group at ETH Zurich developed an algorithm that uses all OCR (optical character recognition) cells of a document as input, thereby better understanding document structures and thus delivering better results for more complex contexts. For example, table rows that are spread over two or more pages can be better identified, or complex price conditions within a table row can be better interpreted. This reduces the manual work to the checking of discrepancies, which still have to be sorted out by the respective departments.

Felix Willers from Muller Martini also benefits from these improvements: "After going "live" with BLP in May 2023, we currently estimate our time savings to be around 40-45% in day-to-day business. The goal is to save 70-80% of the time needed to check and transfer order confirmations and maintain master data. This helps us to focus on value-adding activities". The software helps to continuously improve master data. This reduces the number of discrepancies, which will lead to the desired efficiency goal.

Pioneers in the automation of ERP processes

Thanks to the new technology, BLP Digital has once again significantly improved its product, which is confirmed by a growing customer base. This ranges from classic Swiss SMEs to the largest companies in this country. Together with these customers, BLP is continuously developing its product and is happy to welcome further customers.