Machine learning: fashion’s next revolution


Algorithms are hidden heroes of the modern world. Whether they are making sure you see your best friend’s social posts, recommending you new music or suggesting your next favourite film, they have become central to our lives.

However, we have only seen the start of algorithms’ impact. There will be no sector left untouched by this oncoming revolution and the fashion industry is no exception.

The technology is already making waves, playing a vital role in recommendation engines, which display items of clothing based on the user’s browsing history. And it will not be long before they play a far more complex role in the sector.

They’ll deliver accurate predictions about style and tastes to be able to identify items that consumers love. In other words, trend-driven humans and data-hungry machines will be living and working side-by-side.

>See also: Inside Yoox-Net-a-Porter’s vision for the future of luxury fashion

The perfect machine

For machine learning to truly deliver value, it will have to know individuals and their fashion taste. This does not mean just the clothes people wear – far from it. Instead, machines will need to comprehend inspirations – what people like, why they like it, the mediums that influence them, and who they admire.

The key here is data. For machines to ‘learn’, they need access to swathes of it. This data, correctly analysed, provides huge insight into consumers and their relationship with fashion.

When it comes to machine learning, there have been huge strides made in recent years. A decade ago, a computer being able to describe what was contained in an image was a pipedream. But now it can identify a particular brand of blouse in a fraction of a second.

To achieve this, computers are shown millions upon millions of images, allowing them to identify patterns. This knowledge is then linked to form a complex set of similarities, symmetries and structures, from which the machine can accurately state what’s in the picture and the shop it came from.

So far, so good, but things get much more complex when you insert preference into this. A great fashion recommendation is an inherently personal thing. It needs a human touch. This means that a machine cannot just understand the world of fashion – it needs to get an individual too. For example, it is easy to recommend a customer something popular by using purchase numbers, but it is not as simple to show a consumer something they like.

Understanding fashion vocabulary

As previously mentioned, the most important ingredient in machine learning is data. Yet, not all data has the same value and getting a machine to process it is far from simple.

Take purchasing history. How do you know if something has been bought as a gift? Or, precisely why was an item returned? Solving this requires another level of data comprehension.

For example, when returning an item, consumers often have to select a box explaining why. The same is true when purchasing something online, as the site usually asks if the item is a gift. These signifiers are vital in machine learning, as they assist the software in understanding the intent behind decisions, ornamenting the data.

Another layer of data revolves around browsing behaviour. By analysing a user’s journey through a site, a machine can connect patterns and garner insights into habits. This behaviour can be wide-ranging.

For example, by piecing together what people look at, the search terms they use and even where the mouse hovers, software can comprehend what they love and what they do not. This leads to what can be described as a fashion vocabulary – something that grows organically as fashion changes. And, importantly, machines can learn it.

This is where software is best placed to solve an eternal consumer problem. Most people struggle to describe the apparel they seek, but if a machine knows about consumers – what they love, hate and want – it can infer a great deal from something as simple as a search phrase.

Over time, the software will be able to predict the intent behind the search term, whether the customer is browsing or buying, and connect them with items they will love. To put it another way, computers will not just learn an overall fashion vocabulary, but a specific language for each person.

>See also: Next big thing: rise of the machine learners

This specific language will be created through data. When consumers struggle to describe the clothes they are looking for, machines will look towards other avenues for this information.

By analysing posts, likes and downloads, computers will be able to comprehend this personal expression of intent, while simultaneously creating a specific language for that consumer from constituent parts.

While true artificial intelligence – a technological mind capable of rational thought – is still some way off, software is becoming incredibly adept at recognising trends and painting them as part of a wider pattern.

The fashion industry is at the forefront of these developments, as brands strive to become closer and more useful to digitally native consumers.

To evolve and be truly effective in the coming years though, machine learning requires a human touch. There are still many parts of the fashion world that cannot be comprehended by machines, meaning people will be needed to maintain the link between fashion and individuality.


Sourced from Eric Bowman, VP engineering, Zalando

Avatar photo

Ben Rossi

Ben was Vitesse Media's editorial director, leading content creation and editorial strategy across all Vitesse products, including its market-leading B2B and consumer magazines, websites, research and...

Related Topics

Machine Learning