• Home
  • News
  • Fashion
  • AI in fashion: Interview with VP of Inspire and Engage at Zalando

AI in fashion: Interview with VP of Inspire and Engage at Zalando

By Huw Hughes


Scroll down to read more


Image: James Healey | Credit: Zalando

The role of Artificial Intelligence (AI) in the fashion industry, and the wider commerce sector, is growing rapidly. According to a January study from the Capgemini Research Institute, the technology - which Capgemini predicts could save retailers as much as 300 billion dollars (235 billion pounds) - was being used by over a quarter (28 percent) of retailers in 2018, a significant jump from 17 percent in 2017, and four percent in 2016.

The technology is now well and truly being used throughout the fashion industry’s supply chain - from sourcing to design, manufacturing to retail. Nike, H&M and River Island are just a few examples of the fashion companies who have recently invested (or invested further) in AI.

Another fashion giant investing in the technology is German e-tail giant Zalando. In October, the company launched its Algorithm Fashion Companion (AFC), an algorithm which uses machine learning to suggest outfits for its customers. James Healey is vice president of Inspire and Engage at Zalando, where he leads a multifunctional team of engineers, product managers, designers, and retail experts to develop new ways for shoppers to discover fashion.

FashionUnited spoke with Healey about the success of AFC since its launch in the Zalando website in October, future plans for developing the technology, and the potential AI holds within the fashion industry.

Could you explain what exactly the Algorithmic Fashion Companion (AFC) is and how it came to exist?

Zalando has done a great job in providing a wide collection of items, but we’ve realised based on customer feedback that a lot of our shoppers have two things in mind: One is that they are keen to explore and look for inspiration, proving that Zalando is not just a place for purchasing fashion, but a place for being inspired by it. The other is that, while we always try to make sure our product range is broad enough, shoppers can often feel overwhelmed by the 400,000 items that we sell. We are seeing an increase in engagement - shoppers are coming back more frequently, they’re looking for more ‘snackable’ content, but they’re having a hard time finding that with the amount of content we have.

We notice that when customers interact they’re not looking for individual products to drive inspiration, they’re thinking of products in the context of an entire outfit. They want to know more about styles and trends, and the occasions they might want to buy them for. So they don’t start off with the mindset of ‘I want a new pair of high heels’ or ‘I want a new dress’ - they think ‘I want something I can wear to a festival this weekend, or to a wedding’ without a real understanding of what exact article of clothing they are looking for.

Outfits are a really great way of filling that gap. But creating outfits that are manually created, that have a great sense of style, is something that was very hard for us to scale. That’s why we created AFC. Zalon, our curated styling service of 800 stylists, manually select outfits which we use to create a machine learning step that can dynamically create curated outfits at scale for millions of customers. The main benefit is that while some of those customers will go out and use the Zalon experience and work with real stylists, for many of them, they just want something easy where they can seamlessly swipe and quickly explore the outfits. That’s what we’re trying to do.

How are shoppers reacting to outfits compared to with single items on the Zalando website?

It’s great because we can really see that shoppers are engaged with them. We see that outfit recommendations are driving 40 percent larger basket sizes and twice as high conversion rates when compared to single items. It's because of that insight that we decided to invest in a scalable outfit solution, which we now know as AFC.

We started this with a long term vision in mind whereby the shopping experience wouldn't be so much about navigating through products but navigating through outfits and so these results are really confirming to us that we are going in the right direction, meaning we’re continuing to make a lot more investments in that area.

The outfits are manually created by our Zalon stylists, they have all been tagged so we actually know contextually each part of the outfit - so we know exactly what article they used for their shoe or bag. They also add dimensional facts such as what style the article is, what trend it is supporting and what type of occasion it can be used for.

These attributes then allow us to be able to say ‘well we know that these shoes will match with this dress’, not only from the recommendation but because of the style synergies that have been added as an enrichment based on our 2 million plus outfits.

What has the AI taught you about the way people shop? Have any demographics been particularly engaged with the feature?

Yes. If we look at gender, our male demographic have high engagement. If we delve deeper into it we see that our customers have varying degrees of style confidence and we’ve found that we have three main customer types: Those who are actively seeking a certain item and are really prone to seeking style advice. Others who are searching more on the inspiration side, so they might not be looking for a particular dress or pair of shoes, but for an idea of what people might be wearing at festivals this summer; they want to understand new trends. The last group - the sort of outlier to this - is the customer who already has a high level of style confidence. This customer really doesn't want to be recommended any outfits, or if they do it has to be very accurate, otherwise our credibility is hurt, so we have to be very considerate about how we recommend outfits to these customers.

For these really unique customers who aren’t particularly looking for fashion advice, we may push the AI style finder feature lower down the page, making it less likely that they find it, or we won’t show it at all, based on how well they reacted to it in the past.

Do you think there will come a time when you won’t need the stylists?

No, I think we will always need stylists. AI is an opportunity for us to scale but customers have very unique needs and they enjoy that experience of working with stylists to really find something that they feel helps them with their style confidence. We think that personal touch stylists offer won’t go away.

Are you considering branching the style recommendation service out to include beauty products too?

Absolutely. We consider beauty as a core part of fashion and part of an overall outfit. The difficult part there is that all of the training sets that we used with Zalon don’t actually have beauty products in them, so it is hard to create the level of sophisticated recommendations that we have for outfits. We are now working with our beauty team and our Zalon team to try to figure out how we create this data point that will help us train these algorithms.

Photo courtesy of Zalando

Artificial Intelligence
Machine Learning