Cases
Schoeller Allibert
- Webscraping
- Feature extraction
- Similarity scores
Schoeller Allibert is a significant producer of Returnable Transit Packaging (RTP), including plastic pallets and crates. Customers can either purchase these items directly or request custom designs. In the B2B market, it is crucial to equip the sales team with accurate data about competitor offerings.
Challenge
To gain insights into competing products, simply obtaining raw data from a web scraper is insufficient. This information needs to be organized and formatted according to Schoeller Allibert’s Product Information Management (PIM) system, and identification of rival products is essential. Since web scraping generates a sizable amount of unprocessed data, reducing manual work is vital. However, there can be considerable ambiguity in matching competing products, as key attributes (like “food-safe”) may be absent on competitors’ websites.
Solution
The project was executed in three stages. Initially, a web scraper was developed that could be adapted for most e-commerce sites with minimal adjustments. This tool was capable of identifying 20 characteristics of each product, including images.
In the second phase, the collected data was reformatted to align with Schoeller Allibert’s PIM system: feature names were adjusted, units removed, and dimensions provided in formats such as length x width x height were separated into three distinct features.
The third phase involved creating an AI system to recognize potential rival products. A similarity metric was developed based on the features of competing products alongside Schoeller Allibert’s products, ensuring that similar products received high scores while dissimilar ones scored low.
A product specialist assesses each product to determine which competing items are alike. A visual tool was created to simplify the identification of these products.
Result
The image above illustrates the interface used to select competing products. Here, products can be compared, and “verified matches” can be marked for future reference. By clicking on a match, users can delve deeper into the detailed features of the products. All information on the “verified matches” can be exported in an Excel format.
