Evonik Enterprise Data Management was looking for a new, integrated, and future ready platform to support the data management processes across multiple data assets and value cases. It just needed the right solution and implementation team to put it in place. Evonik products make tires fuel-efficient, mattresses more elastic, medications more effective, and animal feeds healthier. That’s what specialty chemicals are all about. And when it comes to specialty chemicals, Evonik is one of the leading specialty companies in the world. Working with Camelot Innovative Technologies Lab (Camelot ITLab) to leverage the SAP® Data Intelligence platform, Evonik Enterprise Data Management was able to:
- Take advantage of a strategic, innovative data management co-development project with Camelot ITLab and SAP
- Ease support of specification extraction from unstructured data within the purchasing process by integrating domain experts as vital part within AI applications
- Develop and deploy machine learning models in less than three months in the field of packing material creation through SAP Data Intelligence deployment
- Manage the increasing complexity of different data sources to lay the foundation for future data alignment
- Improve the manual information extraction overhead by up to 30% while systems continuously converge to its best possible state
“With SAP Data Intelligence platform and Camelot’s industry and technology expertise, we launched the next generation of enterprise data management. We now have an open platform that lets us efficiently integrate and harmonize different data sources across structured and un-structured data types.” Frank Schmalle, Head of Enterprise Data Management, Global IT, Evonik Industries AG
The first step within the journey was a PoV focusing on smart information extraction
Together with Evonik, our Camelot experts are looking for ways to streamline the manual work that comes up when extracting packaging material information. As of today, the process relies on experts going through every package received: they must either read a PDF document or an image that has all the information like size, volume, name etc. Our aim was to minimize the time and effort it takes for the experts to store valid master data by introducing an AI-enabled extractor and an easy to use annotation tool. Thus, providing the first level of semi-automation for a valid material master data process. 
For this, we implemented a so-called expert in the loop system where the domain expert is in the center of the process. The goal is to learn continuously from the expert input (annotation process) while allowing a more convenient staging towards the master data systems (UX layer). The machine learning framework is operating in two stages which are based on classical Optical Character Recognition (OCR) extraction and a specialized Conditional Random Fields (CRF) machine learning algorithm which takes the already annotated documents into account. The realization and deployment of the project was done on the SAP Data Intelligence Platform where the pipeline is now running in a well-managed environment. This was the overall pre-requisite of the project: how to integrate and manage machine learning applications in the complex world of Evonik’s Enterprise IT. For benefits of the SAP Data Intelligence Platform see our blog post SAP’s Machine Learning and Data Science Platform welcomed by Camelot. 
We would like to thank Frank Kienle for his valuable contribution to this article.
