Improving Product Relationships with Machine Learning
The Successful Collaboration between Intergamma and Squadra
Intergamma is the organization behind the DIY chains Gamma and Karwei. In the Netherlands and Belgium, Intergamma has been the undisputed number one in the DIY market with nearly 400 stores for years. Intergamma aimed to provide related products, such as variants, alternatives, products in the same family, accessories, and upsell opportunities, on its websites. However, manually establishing these product relationships was a time-consuming process. Squadra Machine Learning Company has made it possible to partially automate this process.
Challenge
Intergamma approached Squadra Machine Learning Company to facilitate product relationships based on variants. In many cases, all product attributes are the same except for one attribute where variation occurs. For example, think of a paint can that can vary in terms of gloss level, volume, color, or even a feature like anti-slip. To implement these variant relationships on the sales channel, Intergamma entered into a partnership with Squadra.
Intergamma had made various attempts to establish product relationships based on smart business rules from the product data. However, it was observed that whenever a feature was implemented, it had to be turned off after one or two days due to various issues. Intergamma learned two important things from this experience. Firstly, the product data was of insufficient consistent quality. For example, there were fields that were not filled in or filled in incorrectly, resulting in incorrect relationships between products. Secondly, determining which attributes to vary required significant product knowledge, as it depended on the assortment.
Hence, Intergamma approached Squadra to use machine learning to find relevant product variants that could be manually validated. The aim was to create high-quality relationships with minimal effort.
Solution
The project undertaken by Squadra consisted of two parts. It started with a proof-of-concept in which Intergamma provided Squadra with an Excel dump. Based on this data, Squadra managed to establish product relationships on the PowerRelate.ai platform. These relationships were then validated by Intergamma colleagues and imported back using Excel sheets. This proof-of-concept implementation resulted in improved product relationships for product detail pages in a total of four assortment groups and were A/B tested for Gamma Netherlands, Karwei Netherlands, and Karwei Belgium. The A/B test led to a transaction uplift of 4.6%, according to Anouk Renaud, Product Owner at Intergamma, which she described as a “fantastic score.”
Based on the A/B test results, the decision was made to apply PowerRelate.ai to the entire assortment. Currently, there are approximately 19,000 product relationships live for Gamma Netherlands, 16,000 for Gamma Belgium, and 15,000 for Karwei. It is estimated that around 75% of the total relationships have been established at this time.
Result
PowerRelate.ai features a user interface that allows an Intergamma user to enter the discovered product relationships, view related products, and approve them. The algorithm models these relationships using smart Natural Language Processing (NLP), automatically deriving the distinguishing attributes of the product (such as color, width, and length) from the product description. Fuzzy matching is also employed to detect similarities between values, such as gray/brown and brown/gray. Product images are provided if available, facilitating easy evaluation.
Challenges
Before collaborating with Squadra, Intergamma had worked with Google’s recommendation algorithm. In this algorithm, the product recommendations were entirely automated. However, this algorithm did not deliver the desired quality of recommendations because recommendations in the DIY segment require a high level of precision. For instance, there is a notable difference between recommendations from a clothing store and recommendations from a hardware store. If you purchase a pair of pants and the recommended matching T-shirt doesn’t suit your taste, it’s not a significant issue. However, if you buy a garden shed and the recommended matching underfloor has the dimensions of a different type of shed, it creates problematic situations.
At Intergamma, the relationships are much more intricate, which is why it is important to have the ability to adjust or validate them. Compared to the test Intergamma conducted with Google, PowerRelate.ai has been a positive surprise.
Collaboration
According to Intergamma, the collaboration has been extremely pleasant. “Squadra actively contributed and was proactive in providing various suggestions. They consistently went beyond the surface-level questions to understand our desires and what we wanted to achieve with this project. Furthermore, Squadra took the lead in setting up the platform. Overall, it was a great alignment,” said Anouk.
“Squadra actively contributed and was proactive in providing various suggestions.”