A Fashion-Savvy Algorithm

Zalando uses machine learning to create individual outfits with the Algorithmic Fashion Companion

Can a computer understand fashion? Know what makes a great outfit? A new Zalando product shows that this is possible. We’ve just launched the Algorithmic Fashion Companion (AFC) to inspire customers with individual outfit recommendations: an algorithm with styling expertise that suggests outfits matching the most recent items customers have bought.

AFC Men Screenshot EN
Approximately 50 percent of outfits by both humans and the AFC were considered “good outfits."

The Algorithmic Fashion Companion acquired its taste through machine learning. Developers at Zalando created the algorithm to be able to identify items of clothing and then put these together into outfits. The algorithm teaches itself what makes a good outfit. All it needs is training data that tells it what makes a good outfit. As training data, Zalando used more than 200,000 outfits created by stylists, primarily from our curated shopping service Zalon.

“Customers often tell us that they find it difficult to combine items and that they appreciate getting inspiration and fashion advice,” says AFC Product Manager, Marta Skassa. “A previous analysis from other outfit-based services we offer is that customers who interact with outfits have a higher conversion rate and larger basket sizes. We can now offer this with unlimited, free outfit suggestions based on an item that they either already own, or have already expressed interest in."

The algorithm has already proven it’s fashionability. “We did surveys containing outfits created by stylists and outfits created by the AFC to see if they would resonate with customers. Even our team was really surprised by the outcome,” adds Marta. Survey results revealed that approximately 50 percent of outfits by both humans and the AFC were considered “good outfits.”

“I think this proves that algorithms are really good at replicating things they have learned. Out of all the training data, the algorithm is able to create something that people generally like. Human stylists can set trends by exploring new ideas; algorithms can only follow trends,” Marta concludes.

The AFC creates outfits around so called “anchor items”; articles customers have purchased or added to their wishlist recently. The AFC then recommends additional items that match the anchor item. After the initial tests, the AFC is now ready to be rolled out. It will be available for male and female customers in all 17 Zalando markets this week.

Explore the AFC in the video below.


Related Content