Škoda Auto is the largest company in the Czech Republic as one of the 12 brands of Volkswagen Group, the largest car manufacturer in the world. Škoda produces 900,000 cars annually, with sales of nearly 460 million crowns, and resulting profits of more than 46 billion crowns. It employs 37,000 people in 3 production plants.
Such a large company naturally has a huge invstment in warehouse operations. Besides production, they have 20 warehouses stocking spare parts, tools, maintenance and overhead materials with a total value of 2.5 billion crowns for 250,000 items.
However, such a large warehouse system creates problems, and Skoda Auto needed to identify duplicate items plus find the optimal inventory so that they would not sink money in the warehouses on idle stock, run into storage capacity limits, or other common problems.
However, taking on such a task is not easy -their internal capacity to address these issues was not sufficient and they did not have enough experience, so they were looking for someone to help them tackle this task.
Skoda Auto is a very progressive company and they thought they could find the right team through a hackathon - a competition where they submitted a problem brief plus sample data and saw who among competing teams could do the task best and deliver the best results.
Many analyst firms participated in the hackathon, and no one wanted to miss out on the opportunity to prove themselves to such an important company. In this case Revolt BI easily won the contest, uncovering 6,000 duplicate spare parts (2.4%) during the 48 hours of the hackathon, and developed an MVP (Minimum Viable Product) of a web application that allows employees to work with the duplicates from then on out.
Revolt BI's resounding success convinced Skoda Auto to enter into a follow-on cooperation agreement, which is still ongoing and expanding.
This was not our only successful hackathon at Škoda Auto - the very next year we followed up by winning a hackathon focused on financial variance analysis.
(Revolt BI takes Diploma for winning the Skoda Smart Warehouse Hackathon)
The basis of the solution for the actual identification of duplicates was an NLP analysis of labels and classification based on text similarity and other parameters. Items are deduplicated using a scoring algorithm that precomputes material similarity into the database, while new material requests can be compared using the REST API.
According to other requirements, we have added an additional scoring that works off of turnaround time and aims to get rid of items that are "non-turnaround", leaving only the minimum necessary amount for the following period.
In doing so, we of course evaluate the current stock data, the consumption by month over the last few years, and classify items according to how often and for how long they are consumed.
The aim is therefore to calculate the duration of their remaining stock, plus what items they can get rid of so as not to tie up funds and warehouse space
Quantification is then divided into pieces by item type, meters, kilograms, and then of course by price.
In addition to the calculations themselves, we also developed an independent web application with our partner company Cookielab, which we integrated as part of the internal system for Škoda Auto logistics staff. It is adapted in terms of design and logic to the internal application.
Škoda Auto is extremely satisfied with the result, but of course our cooperation doesn't end there - we continue to fine-tune individual parameters and add more features & functionality.
Thanks to the additional integration of the application, there was no need to train employees, which was confirmed immediately after deployment, when everyone started using it naturally and intuitively, praisiing its benefits.
In addition to deduplication of current stock items, the stock purchasing process has also been improved and streamlined, with procurement staff being automatically alerted to the existence of the same or similar items in any of the warehouses during any purchase.
The system also trains and learns from item movements over the course of the year, and can thus suggest to purchasing staff the optimum order size to meet future stocking demand.
All of this, of course, brings additional savings by avoiding the purchase of surplus stock items or running out of space.