Machine learning is having a big impact on fashion retail.

We recently caught up with senior data scientist Joe Berry at retail tech company Edited and asked him about trends in this area, as well as how the Edited product uses machine learning to help set retail strategy.

(N.B. If you're interested in marketing applications of AI, Econsultancy's Supercharged conference takes place in London on May 1, 2018 and is chocked full of case studies and advice on how to build out your data science capability. Speakers come from Ikea, Danske Bank, Just Eat, Age UK, RBS and more)

Econsultancy: How does your product use machine learning? Is it chiefly for product categorisation?

Joe Berry: As the retail industry is highly visual, product categorisation is one of the most critical aspects where machine learning is used. How a retailer describes a product that customers consider to be similar can vary wildly, creating a lot of inconsistencies that make it hard for businesses to analyse information.

Edited builds systems that review millions of individual items every day to accurately and consistently categorise them. To perfectly categorise a garment, we look at more than just the words used to describe an item (text recognition). We need our machines to process and understand images as well as text. This entails knowing which parts of the picture are the model, identifying the background and differentiating it from the garment being retailed.

These tasks are often complex, as they often require separating a long-sleeved polo shirt from a short sleeved polo shirt, isolating a belt worn over jeans, or knowing what in the database was technical sportswear, versus athleisure, for example. 

Standardising the data in this way is transforming the industry as for the first time, retailers can run a direct comparison of their product assortment alongside every one of their competitors’ merchandise.


E: Will machines ever make pricing and merchandising decisions autonomously? 

JB: The Edited product is about using machine learning to make better decisions in their retail strategies - and this includes approaches around pricing, assortments, merchandising and other specific insights. Machine learning represents a reliable way of categorising data and spotting patterns in data without a risk of making biased decisions. The more data a company can tap into, the better it can understand patterns based on past performance and trends. 

However, in order for machines to fully replace humans, computers would have to be fed information such as margins and inventory strategy, which are not only complex but also highly specialised making it difficult to generalise. The approach we use at Edited is to ensure that retailers have access to the world’s available data organized in a way where they can make strategic decisions based on variables suitable for their business.

E: What are the significant current trends in fashion ecommerce UX?

JB: When a customer enters a physical store, there are clear ways that retailers can maximize the shopper experience to influence a sale. 

For online stores, the customer experience is centered around ease-of-use and convenience. E-commerce can optimise conversion by adding image-based classifications to extract information from pictures, or make product searches much easier.

ASOS, for example, has added an additional product categorisation in a more colloquial way, which better reflects how a shopper might refer to an item. By listing jeans in special category as “high waisted”, “ripped”, “cropped and ankle”, ASOS makes it easier for customer to find precisely what they want without having to browse hundreds of products with otherwise very basic categorisation.

asos search

ASOS colloquial categorisation

Another great UX example used across multiple retailers is adding customer-styled images next to the product image. ModCloth, for example, has an “Explore & Shop Outfit Photos” section where customers can see the product fit on other customers, which aims to boost purchases and reduce the rate of customer returns or exchanges.

modcloth ugc

ModCloth user-generated outfit photos on product pages

E: Do you think visual search will have a big impact on the industry?

JB: Using machine learning to understand images will always be a key functionality within retail. The vast majority of decisions within the industry are made with some form of visual input. Whether that’s a shopper deciding which dress to buy, or a buyer detecting the latest trends from the high street, we all rely on visual input when it comes to retail. At Edited, part of our focus is to create systems that comprehend visual aspects of the industry in a way that is synonymous with human expectation.

Related reading:

Ben Davis

Published 19 September, 2017 by Ben Davis @ Econsultancy

Ben Davis is Editor at Econsultancy. He lives in Manchester, England. You can contact him at, follow at @herrhuld or connect via LinkedIn.

1219 more posts from this author

You might be interested in

Comments (7)

Save or Cancel
Pete Austin

Pete Austin, Founder and GDPR Geek at Fresh Relevance

I'm really puzzled by the UI for this in the ASOS example. Look at the contents of the category drop-down (highlit in red). Are the categories deliberately non-standard?
* No punctuation, e.g. not "men's ripped skinny jeans"?
* A similar category is expressed very differently, "black ripped skinny jeans men"?
* And the ordering looks random - not alphabetic, nor by type or gender

Would have been simple to fix this by data cleaning, so did they A:B test it and find an "artisan" approach to categories either didn't matter or performed better?

9 months ago

Ben Davis

Ben Davis, Editor at EconsultancyStaff


Good question. I've tweeted your comment at Ben Chamberlain, ASOS's chief data scientist. Watch this space.

9 months ago


Ben Chamberlain, Senior data scientist at

Hi Ben, thanks for the tweet and @Pete, very interesting question! My job title at ASOS is senior data scientist, not chief data scientist and I don't work on search, but I will speak to the team that did this and get back to you.

9 months ago

Ben Davis

Ben Davis, Editor at EconsultancyStaff

Thanks Ben. Apologies for promoting you.

9 months ago

David Ironside

David Ironside, Senior Product Manager at ASOS

Hi everyone. The options within the illustrated drop down menu aren't categories which we (the business) have defined, rather these are sourced from our search traffic and aggregated by frequency. It's the customers voice, ranked by demand. Nice and simple. :)

9 months ago

Ben Davis

Ben Davis, Editor at EconsultancyStaff

@David Thanks very much. Makes perfect sense.

9 months ago

Angel Maldonado

Angel Maldonado, Founder at EmpathyBroker

Just came across this piece Ben while doing some research on AI applicability to Merchandising.

We tend to think that WHAT the order of results is what matters in merchandising, however and in my view, is not what the order is but HOW this order is perceived (hence I love the UX sample by JB where customer styled images are utilised).

The way the user (the subjects) sees results, their timing, their appearance, the aesthetics and design that wraps them up is ultimately what creates value.

It’s impossible to understand the most successful brands without being amazed by their style and aesthetics, values which haven’t yet realised their potential on the online world and values for which AI is of no particular use.

Here some limitations we have encountered so far in AI driven sorting and merchandising:

1. “AI” is based on the data sets used to build it’s training models.

2. Training models are snapshots of momentary and perceptible data.

3. “AI” makes decisions from a fixed set of inputs (a fixed point of view).

If we see an online store a set of inter-connected gears, surely AI works. But is there something else to commerce than logic?

Attribution, etc sees users as objects (as opposed to subjects), who make decisions linearly, causally as in a domino effect, and its from those metrics from where AI feeds its learning.

I wonder if we can create absolutely irresistible commerce experiences by seeing subjects as objects.

AI works for stores that are seen as machines, but no extraordinary store experience was ever reasoned logically but rather emotionally into existence.

3 months ago

Save or Cancel

Enjoying this article?

Get more just like this, delivered to your inbox.

Keep up to date with the latest analysis, inspiration and learning from the Econsultancy blog with our free Digital Pulse newsletter. You will receive a hand-picked digest of the latest and greatest articles, as well as snippets of new market data, best practice guides and trends research.