Cebeo, a leading Belgian distributor of electronics to professional installers and companies and part of the larger French Sonepar group started a partnership with Faktion to leverage AI to optimize its internal processes. As an organization with an extensive and dynamic catalogue of technical products, data quality is essential for reporting, product information management (PIM), catalogues, and e-commerce.
To enhance the quality of Cebeo’s product data, Faktion developed an AI system together with Cebeo for automated ETIM search and ETIM classification.
To enhance the quality of Cebeo’s product data, Faktion developed an AI system together with Cebeo for automated ETIM search and ETIM classification.
Manual, Time-Consuming Data Classification Tasks
Cebeo uses 3,281 of the 5,481 different ETIM (V8) classes for its 1.9 million products, of which 750,000 are active items with sales history. The main challenge faced by Cebeo is the manual, tedious, and time-consuming nature of the product classification task. Product managers have to classify technical products based on limited descriptive information, leading to potential inaccuracies and inconsistencies. Furthermore, confusion arises among some products and ETIM classes, as they sometimes have different representations, resulting in incorrect labeling. For example, the term "Differentieelschakelaar" can refer to two distinct Cebeo product groups. In the last two years, Cebeo's product managers have been performing 400 ETIM lookups per month, spending a minimum of 10 minutes per product. Translating this to the total amount of products in Cebeo's inventory, the manual workload is enormous, and we cannot keep up with the pace of new product introductions.
The classification task was very difficult to do manually because of issues regarding data quality. The main challenge Cebeo faced was in their categorical fields, notably the Supplier, Brand, and ParentClass fields. There were inconsistencies, cross-lingual redundancies, and a lack of formal relationships in these fields. Moreover, sparse data issues such as infrequent suppliers and parent classes could lead to model bias. Balancing the dataset with additional instances for these sparse classes would help achieve better model results. These issues significantly impacted both search and classification accuracy for ETIM class selection.
The long process of product classification and the data quality issues impact the business negatively. Reports are less reliable, managing the products becomes a difficult task and the quality of their website suffers from it. Manually managing this data is not scalable over time, and hinders Cebeo from further expanding its business.
To address these challenges, Cebeo needs an effective and reliable solution that automates the task of ETIM code classification with high accuracy, minimizes manual workload, and continuously improves the performance of the models over time.
Cebeo and Faktion's Proof of Concept for Data Quality Automation
Cebeo and Faktion initially engaged in a Proof of Concept (POC) to investigate the possibility of automating ETIM classification using Faktion's Intelligent Data Quality Optimization (IDQO) The focus of the POC is on the main ETIM category. The vision however was to build a productized AI application that would also be capable of predicting the underlying ETIM features in a subsequent development phase. Faktion received a sample of 54,000 products to test its solution and demonstrate its effectiveness in automating product data classification.
Faktion's team identified specific issues, like the overlap of information due to sector and jurisdiction specifics, redundancy in supplier names across languages, and the presence of categorical information within brand names. Faktion provided actionable recommendations, such as introducing a structured taxonomy and a standardized language approach. We also proposed leveraging advanced AI solutions, like expanding the ETIM Search API for categorical fields and using the Bing Search API to improve data identification and consistency. Moreover, the API is also integrated as an Excel add-in, so Cebeo’s product managers could instantly use the ETIMFinder as a macro in Excel.
At the end of the initial phase, Faktion's machine learning tools and models, in a combined setup of unsupervised similarity search and supervised classification could automatically match and classify ETIM codes for each new article at an accuracy of more than 97%.
By reducing substantial workloads of manual effort by Cebeo's domain experts, this AI-driven approach enhances data quality and accuracy, enabling Cebeo to have detailed business reporting and to better serve their customers, with a substantial bottom line improvement.
Resulting in Enhanced Product Data Management
Through their collaboration with Faktion, Cebeo's product managers can now focus on other essential tasks. The improved data quality and product searchability have led to a more positive customer experience, setting new industry standards in electrotechnical product data management.
Addressing New Data Management Challenges
As Cebeo and Faktion continue to build on the success of their partnership, we plan to explore new opportunities for improving data management. Following the successful classification of the main category, the solution will subsequently be used to predict the underlying ETIM product features.
By leveraging Faktion's expertise in AI and machine learning, the broader aim is to tackle other data-related challenges outside the product data domain as well, ensuring that Cebeo remains at the forefront of innovation and continues to deliver exceptional customer experiences.
