Over 1.1 million incoming invoices annually, more than 70 employees involved
Automation of invoice receipt, increase in quality and relief for employees
Return on investment of over 300% in the first year after the rollout
Artificial intelligence in invoice verification
Intelligent algorithms as an accounting aid at B. Braun Melsungen AG
B. Braun is one of the world’s leading manufacturers of medical technology and pharmaceutical products and services. With the expertise of its more than 60,000 employees, B. Braun makes an important contribution to medical progress. Internally, B. Braun is also driving improvements in processes – for example, with the key technology AI in the area of invoice receipt.
Every year, B. Braun’s accounts payable department receives over 1.1 million invoices, which are digitalized and stored using automatic text recognition (Optical Character Recognition (OCR)). B. Braun’s more than 70 financial experts check the invoices and post them to an account and a cost center. A time-consuming manual process.
B. Braun would like to automate the verification and classification of invoices as well as account allocation in order to provide the responsible employees with relief from a time-consuming routine task. With AI as an accounting aid, processes should become more efficient and employees should have more freedom for value-adding work.
In a sensitive business area with high regulatory requirements, such as invoice receipt, B. Braun relies on the data science experts from eoda, a service provider that understands the requirements and can scale the AI system from proof-of-concept to productive use in the business-critical process.
Based on the historical invoice allocations and recordings, eoda developed the machine learning algorithms that automate the verification and account allocation of the digitalized incoming invoices.
Since great potential of AI lies in the consistent combination with human expertise, the system actively involves the responsible B. Braun financial experts. The AI system determines a so-called confidence value for each invoice checked. If this confidence value is above a defined threshold, this means that the algorithms are secure in having correctly assigned this invoice and that no human intervention is required.
If the confidence value is low or the invoice amount is particularly high, B. Braun’s financial experts intervene. Through feedback from experts, the algorithms are trained and further improved. The expertise and experience of eoda’s data experts enabled more than a third of the invoice checks to be completed via the workflow automations after a very short time.
In addition, eoda trained those responsible at B. Braun to handle the AI system and to be able to easily adapt it to their requirements. This makes the algorithm comprehensible for those involved and sustainably increases acceptance.
With the support of eoda and the use of artificial intelligence, it was possible to automate the invoice receipt process to a greater extent, create more time for other tasks, and improve quality.
“Working through AI saves our employees’ resources. The project with eoda is a big step towards more productivity. The hit rate of the AI is fortunately high and the users are very satisfied with the results. In the first year after the rollout, we can generate an ROI of more than 300% with this project.”
Andreas Amrein | Director Smart Automation Services & Accounts Payable | B. Braun
“B. Braun has recognized the enormous potential of AI and wants to use it to further strengthen its competitiveness. With the successful integration of AI into the accounts payable processes, we were able to jointly implement the first productive AI project in the entire company. We are grateful to B. Braun for the trust they have shown in us.”
Heiko Miertzsch | Co Founder & CEO | eoda
Information on technical and methodological implementation
For optimal usability, eoda seamlessly integrated the algorithms into B. Braun’s existing system landscape around Microsoft Azure Databricks. The algorithms and their calculation were implemented by eoda in R and Spark to ensure high performance of the analyses even with this large database. The methodological basis of the automated classification of incoming invoices is ensemble modeling using random forests, boosted trees and neural networks, as well as other methods.
*Image source B. Braun Melsungen AG