D3S team developped an Artifical Intelligence leveraging mass webscrapping to reduce General Procurement spend. It was applied to a leader in the aeronautic industry during the COVID crisis. The digital solution allowed to classify goods and services in robust categories and detect unecessary purchase requests. Short term benefits were demonstrated.
DATA & STACK
Our algorithms analyzed x millions purchase orders and millions of Amazon product descriptions through webscrapping bots.
The technical stack included : Python, Scrappy, SKL, Numpy, PostgreSQL and QlikSense (requested by our customer).
CONTEXT & APPROACH
There is a high General Procurement cost base with very heterogeneous goods & activities in the industry : X Bn€ yearly spend baseline accross x0 000 suppliers and >1 million PO lines per year. Classifying all procured goods and services is challenging as they described with short free text. As a consequence, the dynamic financial baseline and savings control loop are not robust enough to trigger targeted actions.
COVID-19 crisis requiring drastic cash containment measures. Optimizing General Procurement spend was critical.
The first step was to classify all goods and services in small categories (eg. for consumables, identify hammers, screw drivers, metrology tools, chemical products, papers and pen… among many other categories). To achieve this results, we trained an algorithm on a data set build from a mass scrapping of popular market places in several languages. this allowed to establish a link between text description and the market place category. The algorithm was then applied to the text describing the purchased goods.
An even deeper caracterisation was achieved on machining cutting tools, analysing pdf and drawing to extract dimensional information and key features. this allowed to standardize procurement of these tools across plants.
We then trained a bot to identify suspicious and non critical purchase request. A Committee was then confirming the algorithm proposal to block the request and reduce spend during the crisis period. Confirmations were used to continuously improve the algorithm.
Tbe solution delivered contributed to build a robust baseline allowing savings control loops. Our Artificial Intelligence algorithms and the governance setup during the COVID Crisis contributed to 1 billion euros cash reduction. The results were frequently reported to the Executive Committee.
All developments were transfered to internal team and a dedicated data scientist was hired to take over.