Procurement-Suite

Challenge

The procurement process in the automotive industry is a complex and multi-step task. Corporate purchasing supports departments in selecting suitable suppliers and achieving set budget goals. Preparing for tenders and negotiations can be very time-consuming. Our client faced the challenge of supporting their buyers with a data-driven tool in the procurement process and making corporate knowledge efficiently accessible. This was intended to automate parts of the procurement process, minimize the effort in finding relevant new suppliers, improve the discoverability of comparable offers, and optimize the price negotiation process.

Approach

Since the requirements in the various procurement steps differ significantly, specific machine learning models were developed and integrated into a web-based application. To identify relevant suppliers for a new tender, different data sources were combined in an NLP model, including internal information such as offer texts or past tenders and external sources.

To find offers based on keywords or by uploading a new offer text, two types of search methods were linked: a symbolic search, which allows for the efficient finding of exact keywords or item numbers, and a neural search, which maps and makes complex issues discoverable. Additionally, predicting a realistic yet ambitious target price was enabled, with the model incorporating not only internal factors but also external factors like macroeconomic data to reflect general price trends.

Furthermore, sellers are supported with arguments for the upcoming negotiation, derived from the model through explainable AI. The ETL process for data loading and model training is orchestrated by Airflow. The machine learning models were made available as microservices in a Kubernetes cluster through an automated deployment pipeline, enabling uncomplicated and highly efficient scaling.

Result

Our procurement suite makes the procurement process more efficient for our client, reducing preparation time and allowing buyers to focus on the most important tasks. Additionally, the negotiation process for low-value purchases, which was previously rarely negotiated, was automated, resulting in significant additional savings. Access to important information, which was previously difficult to find, has been significantly improved, allowing for the discovery of new and more suitable suppliers and the comparison of past offers with recent ones.