Five ways AI is working for retailers

From producing the perfect clone of a Kardashian’s latest Instagram post to racing to restock a dress worn by Kate or Meghan, speed to market is part of what gives retailers a competitive edge.

A WGSN report last year found that fast fashion retailers are popular with consumers because they get products to market more quickly.

Artificial intelligence and machine learning are helping more retailers achieve this by transforming traditional processes in product development. Brands are beginning to use these tools to understand what products and styles will prove the most popular with their customers, and in the very the first stages of the creation of a collection, AI can pull in imagery through targeted scanning of the web and social media, as well as pinpoint fashion trends to guide and inspire designers.

Arti Zeighami, H&M group’s head of advanced analytics and AI, sees it as a key support to the design process: “Fashion forecasting is still at the early stages but it has shown really great results. We’re helping our designers and our merchandisers to be sharper in their decision making.”

Fashion forecasting is still at the early stages but it has shown really great results. We’re helping our designers and our merchandisers to be sharper in their decision making

Arti Zeighami, H&M group

Doug Stephens, founder of trend forecaster Retail Prophet, says human intuition can slow things down: “We’ve been relying for the most part of human intuition to sort through the data and understand which products are most likely to succeed, and a lot of that has been responsible for the time it takes to get products to market.

“Now we’re on the brink of breakthroughs that will enable AI programs to analyse myriad data and find hidden correlations in that data in a fraction of a second of the time.”

Another way to make development processes more efficient is to estimate a product’s cost and price positioning, through data and image analysis.

Emilio Cogliati, senior manager at Accenture, explains: “[Retailers] can easily estimate the costs just after the initial creative sketches with an error margin of 5% to 10%. Taking more accurate decisions gives rise to a better assortment in the collection, and an increase in sell out during the full-price period.”

Navigating the complexities of getting products on to the right channels can be another time consuming aspect of the process, which AI is starting to tackle. Drew Smith, vice-president of product strategy at retail technology company Volo says: “We’ve seen brands spend a lot of time trying to guess which channels to go on and, then, once they’ve made that decision, being able to get products out to those channels is really difficult.” AI speeds up the process because it tracks and understands the requirements demanded by different channels.

Smith adds: “If you take Amazon, Instagram or Google, they are constantly updating their requirements.

“You have to be able to monitor how a consumer is searching on those different platforms, but then also understand the necessary data from a channel perspective. With AI you can do this in real time and allow it to adapt your product information in real time to fit.”

Increased automation speeds up processes across the board

Sally-Anne Newson, Shop Direct

Getting the right products in front of customers as quickly as possible requires analysis of product attributes, such as size, shape and colour. Shop Direct, the owner of Littlewoods and Very, enhances its product data using AI-powered computer vision technology that recognises characteristics from images, such as floral patterns.

“After seeing enough of such images, the algorithms can detect these attributes on new items, quickly enriching our product data,” says Sally-Anne Newson, director of customer experience and digital product at Shop Direct, adding: “Increased automation speeds up processes across the board.”

And for those retailers who have always excelled at speed, relevancy is the aim, H&M’s Zeighami notes. “Speed is not a problem, we can create stuff very fast and get it to market. You want to make sure you have the right product for the right customer in the right place at the right time,” he says.

The number of channels that shoppers use to find fashion is growing.

From Instagram to image recognition to voice-enabled speakers, newer channels are connecting brands more closely with consumers, and having a presence on these channels is an increasingly necessary part of making sure products reach the right people.

The challenge is that each has its own continuously updated SEO requirements, which makes this task more complex. For some brands, AI is not only monitoring these SEO rules, it offers a deep understanding of the different ways consumers search for products. Natural language processing (NPL) analyses the vast number of terms people use to describe what they are looking for.

“If you understand the product very well, and you have good attribution of the product, then you can create taxonomy around that product,” says Arti Zeighami, head of advanced analytics and AI at H&M group. “Different people use different words for the same product. This is what we are looking at with natural language processing.”

This holistic approach to search can give retailers a competitive edge.

An AI system can provide a solution that is proactive, telling an ecommerce team exactly what they should do

Drew Smith, Volo

“The challenge [with search] is that it’s a lot of reactive work, whereas an AI system can provide a solution that is proactive, which is telling an ecommerce team exactly what they should do and how to set it up,” says Drew Smith, vice-president of product strategy at technology firm Volo.

Like many other fashion companies, N Brown Group is deploying machine learning to increase the relevancy of its search results.

Gareth Powell, director of data science, says: “We have developed a search-relevancy score using machine learning to quantify the relevancy of our own search results, which we are then using to further optimise those results, so that we present customers with the most appropriate products to their search terms, and help them find the products they want quicker. We’re currently working on improving this so it can be self-learning.”

Image and voice search are transforming how people find and discover fashion. Brands are beginning to experiment with voice, as more consumers embrace voice-activated devices and digital assistants such as Amazon Alexa and Google Assistant. H&M, which offers voice search through H&M Home Stylist, is among them.

“By giving a voice to our brand, we will enhance the customer experience and be able to speak with our customers whenever they want,” says Camilla Henriksson, head of marketing and communications at H&M Home.

Many retailers, including Asos, H&M and Marks & Spencer, offer image-recognition tools that allow consumers to upload a photo of an item and receive suggestions of matching shoppable products. Continual improvements to the technology look set to transform the search experience further.

Kelsie Marian, principal research analyst at Gartner, says: “New applications can use pixel-by-pixel analysis to identify multiple items represented in an image, thus expanding the accuracy of the search and the corresponding results.” This will enable consumers to shop for all items in an image, such as the earrings or shoes that have been paired with a dress

Virtual stylists offer personalised product recommendations that match a consumer’s taste and style, as well as taking search a step further. StyleScript’s AI Fashion Stylist and Amazon’s virtual fashion assistant, Echo Look, not only recommend items to buy, but analyse the outfit users are wearing.

As voice and visual search converge in the future, consumers will be able to interact with their own voice-activated personal stylists that are armed with a deep understanding of the user, and only suggest fashion items they will like. Increased adoption of this technology may only be a few years away.

Emilio Cogliati, senior manager at Accenture, says: “In two to three years’ time, instead of scrolling through boring web pages, you will be able to interact with someone who knows you and makes interesting recommendations.”

One crucial difference between artificial intelligence and traditional report-based systems is AI’s ability to come up with what should happen next. 

It can also find answers to questions that would take humans a huge amount of effort.

“With products such as Google Analytics, you have a lot of data and reporting going on, but to actually get to the crux of what you need to do is really difficult on those systems,” says Drew Smith, vice-president of product strategy at tech firm Volo. “The beauty of AI is that you can constantly monitor millions of data points in real time and instantly provide users with a ‘this is what you should do’ recommendation around SEO or keywords or market trends, even down to when they should promote a product.”

A rapidly growing number of retailers are adopting AI to make increasingly personalised recommendations. A global study last year from the Capgemini Research Institute found that 28% of retailers are deploying AI, up from 17% in 2017 and just 4% in 2016.

We use unsupervised learning in this space to highlight customers and trends that aren’t apparent off the back of traditional reporting

Gareth Powell, N Brown Group

The rapid adoption of AI is not surprising considering the improvements it can potentially make throughout a retailer’s operations – the Capgemini study estimates AI could offer as much as $340bn (£262bn) in cost-saving opportunities for retail companies.

Gareth Powell, director of data science at N Brown Group, says machine learning algorithms have enabled the brand to connect products with customers in a way that would not be possible using the reporting systems of the past: “We use unsupervised learning in this space to highlight customers and trends that aren’t apparent off the back of traditional reporting.”

For example, he adds, the retailer is now more easily able to pick out the most popular product in a group, which it can then present to new customers to persuade them to purchase. “We’ve developed network product maps to identify naturally occurring clusters of products, enabling us to curate a list of ‘hero products’ that individually are representative of the cluster of products they sit with, and together span a wide range of major product clusters. This hero product set is then used to communicate our portfolio of products to customers.”

However, for all of its capabilities, AI is a support for human-powered strategic thinking, not a replacement.

”AI is just a tool. It’s data we’re utilising to be sharper in our decision making,” says Zeighami at H&M. “AI amplifies our existing intelligence.”

From shelf-stacking robots to merchandising planning, AI is gradually taking over time-consuming menial jobs.

It is also tackling highly complex data-centered tasks that cannot be executed efficiently by humans. This can give more time staff to focus on the more creative aspects of their jobs, as well as more information to work with.

Arti Zeighami, H&M Group’s head of advanced analytics and AI, says the technology allows staff to instantly calculate outcomes, instead of spending time trying to estimate them: “Quantifying how many pieces of a certain garment we should buy for certain markets can require a lot of decision making, and takes lots of hours to do. Now, we’re introducing AI to that process, which opens up time for merchandisers and buyers to do other stuff and focus on what they’re good at. They came to H&M to work with fashion – not to sit with Excel sheets.”

Buyers and merchandisers came to H&M to work with fashion – not to sit with Excel sheets.

Arti Zeighami, head of advanced analytics and AI at H&M Group

AI has also transformed processes around product data management. Each fashion item can have hundreds of product attributes, spanning everything from colour and fabric to the country in which it was manufactured. The technology can manage inventory across ecommerce channels and in store, getting products to customers more efficiently. N Brown Group is among those turning to AI to help address this challenge.

“We have been investigating how we could use AI methods to carry out image processing to reduce the need for extensive product attribution by inferring as much as possible directly from the product image,” says Gareth Powell, director of data science. “As with all retailers, it’s crucial we buy in the right volume of stock, yet much of the time we are buying lines that we haven’t stocked before. Traditionally it’s taken in-house expertise to estimate how much stock to buy.

“We’ve now developed AI models which predict how many items we’ll sell, by size. This works by creating clusters of ‘similar’ items we’ve sold in the past and using the historic sales data of these products, by size, to forecast future sales of a new (but similar) line. The ability to create clusters of similar products is heavily reliant on having complete and consistent product details.” He adds the team is using image recognition to automatically generate product details. “This is more consistent, more complete and more comprehensive than what we could ever achieve through manual entry.”

As retailers overhaul outdated process to save costs and increase efficiency, it is unavoidable that staff roles will evolve. The British Retail Consortium estimates that over the next 20 years, 60% of jobs in retail could be lost to automation.

“Any task that requires a basic analysis of data is going to fall into the capabilities of AI,” says Doug Stephens, founder of industry forecaster Retail Prophet. “A lot of the rudimentary product knowledge that retail associates have been responsible for and a lot of data-driven tasks, such as inventory management that floor employees are responsible for are going away.”

Last year, Zalando cut around 250 marketing and communications roles, and replaced staff with algorithms and artificial intelligence. Uniqlo transformed a warehouse in Tokyo with an automated system that resulted in 90% of the staff being replaced by robot technology that does everything from moving crates on to conveyor belts to identifying products by scanning electronic tags. This also enables the warehouse to operate around the clock.

For things like knowing when the seasons are changing, humans are really important

Rob Feldmann, Brand Alley

But, despite the advance of AI, experts say that the human touch will remain vital to operations in the fashion business. Key staff, such as merchandisers and buyers, will be supported by AI systems, but they will not be replaced, Emilio Cogliati, senior manager at Accenture, believes: “AI can do the basics of a merchandising and assortment plan, but the review of the plan should still be revised by humans, with their creativity and talent.”

Brand Alley chief executive, Rob Feldmann echoes this sentiment: “At the moment I trust my merchandising teams more than I trust the technology. For things like knowing when the seasons are changing, humans are really important.” Retailers will, however, increasingly seek to hire people with different skillsets as roles demand more lateral thinking.

Stephens says: “[Retailers] are going to be looking for people that are much more dynamic and creative, and lateral problem solvers.”

Fashion trends are usually picked up on by forecasting experts, who use intuition and experience to predict what will be big in a year’s time.

AI has accelerated the business of fashion forecasting and trend spotting with its ability to scan social media and the web, and analyse data such as the colour, size, style and pattern of clothes.

Yoox Net-a-Porter Group has given an idea of how this could inform the design process. The clothes created for its first own-label brand, 8 by Yoox, are designed with input from in-house AI tools that analyse social media, and pick up the most buzzed-about trends and styles.

Luxury fashion brands Christian Dior and Louis Vuitton have enlisted start-up Heuritech’s AI-powered mood boards for product development and merchandise planning. Paul Smith, meanwhile, has used Google’s Art Palette to help with colour inspiration. He said of the project last year: ”We would start with working with the colours that I’ve selected for that season, then through the app look at those colours and see what gets thrown up. That can be so influential on how we design.”

H&M has also begun to use AI-powered fashion forecasting. As well as being a source of inspiration, the insights it provides empowers designers, argues Arti Zeighami, head of advanced analytics and AI at H&M group.

“AI amplifies our existing intelligence,” he says. “It can help designers be more sure of their gut feelings and make the right decisions.”

Brands are also developing a deeper understanding of customers both online and in store. H&M uses AI to offer a tailored product mix in its individual stores around the world.

River Island is doing something similar. “A lot of what we’re doing now in our roadmap is building out a strong single view of customer and capturing interactions in the store,” says chief information officer Doug Gardner. “Right now the only data you’ve got is signals you’re getting off the website in terms of intention. It’s not until you start to know and capture interactions within stores that AI really kicks in, as you have more signals about what people are doing online and store.”

“Customers are different, depending on where they’re from,” says Zeighami. “If you have a local store in a city where a large amount of inhabitants are university students, that store will look different to a store in a posher area. It’s all about getting closer to the customer.”

Internal data on popular products can also be helpful. At US brand Stitchfix, for example, AI is used to pick clothes for customers based on their preferences, style trends, and which products are popular with people of the same age and demographic. Meanwhile, Net-a-Porter is testing technology that scans customer data for information on people’s upcoming trips and events, and uses that to tailor its recommendations.

N Brown Group uses AI to calculate the right stock levels and correct sizing for its product lines.

“As specialists in fit, we’ve built models to predict which size would best suit a customer for a given product,” says Gareth Powell, the group’s director of data science. “To trade smarter, we know having the right level of stock and sizing is critical, so we’ve developed AI models for our product teams to optimise buying quantities in future.”

Having the right level of stock and sizing is critical, so we’ve developed AI models for our product teams to optimise buying quantities in future

Gareth Powell, N Brown

Factors that affect future sales and performance, such as when to discount products and when to restock them, can be calculated with much more efficiency by AI tools.

“Most [fashion retailers] on Monday morning go through a long report trying to understand what the next outcome is, when AI allows the ability for the user to remove all that hard work and just say this is what we’re seeing, this is what you now need to do and action that instantly,” says Drew Smith, vice president of product strategy at technology firm Volo. “The power to be proactive and on top of consumer behavior changes ensures you’re future-proofed with an understanding of what your customer is buying and what they care about.”

The technology will continue to move forecasting beyond human capabilities – rather than responding to trends, brands will soon be able to see them before they are visible to the naked eye.

It is clear that AI is already infiltrating different parts of the fashion business; in the next couple of years, this is likely to accelerate, shift and provide further game-changing options for retailers keen to evolve.