Zalando relies on technology to aid fashionistas

Eric Bowman of German online apparel giant is betting on artificial intelligence to take fashion further

Eric Bowman is VP of engineering at online clothes and shoes retailer Zalando, which describes itself as Europe’s biggest fashion platform serving about 19 million active customers across 15 countries. Unlike most companies in the sector, Berlin-headquartered Zalando is outspoken about its use of technology and how critical this is to its operations which are based on custom-built logistics centres and pure-play e-commerce. The company even has a “bespoke development and cultural methodology” called Radical Agility (YouTube video) that it says helped to attract many of its tech department’s 1,350 staff. Zalando’s original backers included Germany’s Rocket Internet group, well known for refining existing formulas for internet success.

Eight-year-old Zalando has recently worked with Google on Project Muze, using machine learning and artificial intelligence (AI) to design apparel. It also developed  Bread&Butter by Zalando – a trade show turned virtual “trend show”. I swapped emails with Bowman to discuss the confluence of fashion and technology at Zalando which had almost €3 billion in annual revenue for 2015.


Q. On the face of it, fashion and AI are odd bedfellows. Did you always see the link or was this something that crept up on you?

A. The many areas that make up the broader scope of AI have been part of Zalando’s technological framework for some time now: machine learning, deep learning, algorithms, neural networks, on top of augmented and virtual reality (AR/VR), which are related disciplines. Zalando is utilizing the power of these approaches to improve its products for customers, on top of using intelligent systems in operational processes, such as warehouse logistics. While our own example shows that AI technology can span several domains, its application to fashion is exciting for technologists and fashionistas alike, particularly when looking at our recent collaboration with Google on Project Muze.

How did the relationship with Google come about and what struck you about Project Muze that made it so attractive?

The opportunity to showcase clothing designs and visualizations made by intelligent machines was something we had to be a part of, given our passion for fashion and technology. We want to connect people and fashion: this pilot is a pioneering experiment that allows consumers to be the muse for their individual 3D fashion designs by applying machine learning in the creative process, the ideal playground for fashion and technology to connect. We also had the perfect platform to reveal such a project, with our Bread & Butter showcase hosting the pilot in early September.

As users and creators of open source software, we felt it was important to work with another organization just as passionate about open source values. Zalando initiated the project; we take a very open approach to machine learning and were excited to help explore how this technology could benefit users in a creative field. Zalando and Google collaborated with creative studio Stinkdigital and professional fashion designers from [Zalando’s online private fashion brands group] zLabels, on top of using Google’s open source platforms, products, and technical support.


For several decades now people have talked about how neural networking, AI, fuzzy logic and so on will change the world. Do you tend to differentiate between the different technical approaches or just take them all for what they are worth under one broad umbrella?

Under the umbrella of AI, Zalando is currently concentrating on machine learning, which makes up part of the vast technological space that AI incorporates. We use machine learning to ultimately help us influence fashion journeys, fuelled by style preferences, trending Zalando styles, and a rich collection of data that customer behavior provides us with. All of these elements allow machines to recognize and categorize virtually any fashion item, which is then used to form complex article representations that reveal symmetries, structure, and trends from every corner of the industry. We can ultimately use machine learning and AI to revolutionize these fashion journeys, completely modifying the way we approach the acts of buying and browsing.

What is your biggest challenge with AI today? Maturity of the technology? Tools? Skills? Applying it to a vertical like apparel?

As we’re working with machine learning, one of the biggest challenges is ensuring that we’re collecting the right data points for machines to learn from, on top of developing a deep understanding of customer preferences. Machines learn adaptive, personalized recommendations from a lot of different data, which are primarily focused on past behaviors. Prior purchases and returns by customers are useful, but can also be misleading. Browsing behavior, such as what consumers look at, how long they look, and what they search for, can also provide helpful insights. In either case, these clues do a lot to help, but it’s not always the entire story: we’re also challenged when trying to turn what we’ve learned from this data into valuable learnings and forecasts for the future. Software can combine an understanding of what we know and understand about customers, but we’re constantly iterating and improving on the tools and processes to achieve the most well-crafted recommendation.


In some ways the fashion world appears (to this untutored eye) to be bound by certain rules and conventions. For example, retailers and media quickly falling into line after Paris catwalk models show off the latest season’s looks from leading design houses. Does that predictability make it a suitable environment for digital technology to have a big impact?

Digital technology has already had a major impact on the world of fashion, and the best example of this comes via the demand for instant gratification off the catwalk by customers, with Burberry and Tom Ford making their collections available immediately for those interested. This urge to instantly shop for trends pushes traditional fashion houses to keep up with the world’s newly normalized digital speed, creating the perfect leverage for mobile technology in fashion. These changes are already considerable in the eyes of the still somewhat conventional fashion industry.

How far can this all go? Do you ever see a scenario whereby fashion and garment design becomes automated to the nth degree or where swathes of activity like PR, marketing, advertising etc. could be disintermediated?

We believe that a certain level of harmony can be achieved when humans and machines work together, in parallel, in tandem. Especially in the context of fashion. However, a human level of personalization through software (via machine learning) is still something far off, and we’re completely okay with that. All the data, housed in all the computers in the world still can’t truly know and understand how to choose styles that customers will love, which is why the work of stylists, designers, and fashion houses remains essential in the equation. Even though software is currently unable to emulate the personal connections that brands and designers have with their clientele, it can still learn how to accurately identify articles customers may love, with stylists along for the ride and both sides benefiting from this synergy.


Will there ever be a major brand launch that is truly data-driven and almost autonomous?

Being predominantly data-driven is an endpoint that a lot of companies have already reached, and Zalando is no different here. Of course, there is definitely the possibility of major brands having autonomous and data-driven launches in future, but predicting how that might be achieved would be purely speculative on our part.


Also read:
Body scanning could stop retail clothing returns
Is visual search ready to revolutionise retail?
From Gulf War to Harrods – one man’s race to get wearables on track
Adidas a show in for 3D printing revolution