Deep Learning in the Context of Inventory Valuation in the Pharmaceutical Industry

Artificial intelligence is currently one of the hot topics in business in general and thus also in the pharmaceutical industry. Due to the increasing digitalization of processes, the amount of data is growing daily. Data always also means a kind of "knowledge" that can be subsequently used by applying the right methods. With over 23,000 different stock positions, Teva ratiopharm offers a wide range of local inventory at the site in Ulm, Germany. Recurrently, from an accounting perspective, this inventory must be calculated to its value for the item on the balance sheet. With the amount of inventory, it is currently not possible for humans to value each item individually – there is a general average evaluation. Due to the mass of data, complex patterns and dependencies can be hidden in data that humans are not able to recognize with the eye. Consequently, the more abstract granularity of the valuation does not allow any conclusions to be drawn about inventory optimization. In this master thesis, we present the development and implementation of a prediction model, which is based on historical destruction data and is supposed to predict future destruction of every single stock position. By predicting the destruction of a stock item, its residual value can be easily derived. We only focus on the area of deep learning in order to be able to go into more detail about the methodologies and functionalities in this area. By using neural feedforward networks, the process of inventory valuation within Teva ratiopharm will be optimized and automated. In addition to the application of neural networks, rule-based manual process steps are automated using the robotic process automation approach. This work describes one of the first Teva ratiopharm internal projects towards artificial intelligence. The project should also include the development of knowledge and experience in this field and the networking of structures and people within the company for future digitalization projects. Furthermore, the topic artificial intelligence is still quite new at Teva ratiopharm and digital transformation has just started.
Therefore, a kind of trust in artificial intelligence and other approaches in this direction must first be established within the organization and among employees. An important component of trust describes the explainability of such models. We will take a closer look at three different
approaches in the area of explainable artificial intelligence and discuss the topic in the context
of the inventory valuation process.