How machine learning is transforming retail both online and offline

Posted on: 22/10/2019

According to a 2018 study by Juniper Research, spending by retailers on artificial intelligence tools will grow almost fourfold between 2018 and 2022, reaching $7.3bn per annum – up from an estimated $2bn in 2018.

By 2022, the research found that retailers will be investing in AI tools to improve everything from customer service to sentiment analytics, automated marketing and demand forecasting.

Ten years ago, using artificial intelligence to optimise retail processes, power product recommendations and improve the customer experience was the sole province of cutting-edge ecommerce powerhouses like Amazon. But the second half of the 2010s has seen a huge acceleration in the capability and availability of artificial intelligence and its subsets – machine learning and deep learning – and a veritable explosion of potential applications in retail, both ecommerce and bricks and mortar.

Machine learning (a subset of AI that refers to the use of algorithms to analyse data and learn from it) in particular has seen a wide variety of applications in retail as retailers use it to streamline processes, optimise pricing, forecast demand, surface products in an intelligent way, and provide a tailored, personalised experience to the shopper.

From visual search to computer vision, natural language processing to predictive modelling, machine learning underpins all kinds of innovations that are levelling the playing field by giving retailers of all sizes access to the same tools as behemoths like Amazon – and allowing them to develop cutting-edge online and in-store experiences.

This in-depth briefing will look closely at a number of different applications for machine learning in retail, accompanied by examples of how retail brands are putting them into practice and how they translate to improvements in sales, processes, customer engagement, and the customer journey. It will examine both ecommerce and bricks-and-mortar retail, noting the differences in how machine learning is used in digital versus offline environments, before finally considering how this usage might evolve in the future.

How are retailers applying machine learning in ecommerce?

Machine learning is key to many of the current trends we are seeing in ecommerce, particularly regarding product discovery and personalisation, powering innovations like dynamic personalisation, visual search and intelligent assistants.

It also plays a more invisible – but nonetheless key – role behind the scenes through forecasting, optimisation and analysis, saving retailers valuable time and energy and aiding them in decision-making.

Product discovery

Recommendation engines

If you’ve ever been tempted into buying something featured in a section labelled ‘Frequently bought together’ or ‘People who bought this also bought…’ – you’ve encountered a product recommendation engine.

Product recommendation engines are typically powered by sales or product data (for example, most popular products, discounted products, or similar products) and/or customer activity (such as purchase history or browsing history). Machine learning algorithms will analyse the data they are provided and make connections that determine which recommendations are shown to the customer.

Recommendation engines can introduce customers to relevant products within an online store they otherwise might not encounter. Recommendations that are tailored to the individual customer are also a powerful tool for personalisation.

Amazon is one of the most notorious examples of retail product recommendation in action, having used machine learning to power its product recommendations since its early days. A visit to Amazon will show multiple types of recommendation engine in action, from ‘Get yourself a little something’ product recommendations based on a user’s wish lists, to ‘Customers who viewed this item also viewed- related products, to best-selling and new items in a specific category.

However, Amazon is far from the only retailer using product recommendations to good effect. The eBay homepage also features a variety of recommendation engines, from recently viewed items to daily deals to ‘trending’ items. Cookware retailer Sur la Table displays trending product recommendations when a search returns no results, to keep the customer journey going even after a dead end.

In an example of using recommendations for both product discovery and personalisation, luxury beauty brand L’Occitane integrated tailored product recommendations into a relaunch of its mobile experience in a bid to mimic the more personal in-store experience customers would normally receive in a beauty shop. After customers browse the mobile site, their personalised recommendation feed will be populated by a mixture of popular products and items based on their viewing history.

Customers browsing a product page will also be presented with suggestions to ‘Complete your routine’ offering complementary products, complete with checkboxes to help them add the products to their basket.

Visual search & visual recommendations

Machine learning is increasingly fuelling visual search and discovery tools that can intelligently suggest and surface product recommendations that are visually similar, or complementary, to what a customer is searching for.

This is a huge boon for retailers and for customers as it improves the browsing and navigation experience by allowing customers to easily move between related items, and also encourages them to purchase complementary items and put together an outfit, a matching set of cookware, or a suite of furniture.

Online fashion retailer Asos is one well-known example. It integrated visual search into its app in 2017, allowing customers to take pictures of clothing in the offline world and search for similar items on the Asos app. Customers can also use images in their photo library or taken from magazines to surface visually similar products.

Fashion brand Forever New recently partnered with Attraqt, an on-site search, merchandising and personalisation provider, to process product data and recommend visually similar products to shoppers based on their browsing behaviour. The items appear in a ‘You may also like’ section below the product page.

Forever New found this section generated up to a 135% increase in conversions from shoppers. When shoppers were shown visually similar items, their average order value increased by 21%, with full-price items more likely to be selected over sale items.

Alongside this feature, Attraqt and Forever New released a ‘Shop the Look’ tool that automatically recommends additional products that work well with a specific item they have selected. A fully-fledged visual search tool is also slated to launch in the near future.

Conversational commerce

Some retailers have also implemented chatbots, or automated assistants, on their sites to provide more informed product recommendations and facilitate navigation. These chatbots might make use of natural language processing to interpret customers’ responses and make exchanges more fluid. They can also act as a vehicle for other product discovery features.

For example, Asos launched a ‘fashionbot’ named Enki in the summer of 2018 that incorporates Asos’s visual search functionality and ‘style match’ technology to surface products that are visually similar to a photographed item. It can also surface recommendations based on the user’s previous browsing history and allows users to set a budget for the recommendations they receive.

The North Face is another brand that has tapped into the power of conversational commerce to surface more relevant products for its customers. It partnered with IBM Watson to create a dialogue-based recommendation system that would ask simple questions about where, when and how the customer would be using an item of clothing. Incorporating weather forecast data and the shopper’s gender, it would then use that information to narrow down six products that fit the user’s specifications.

During the 60-day trial of the tool, customers reportedly engaged with it for an average of two minutes, and 60% clicked through to try product recommendations.

While some implementations of retail chatbots have been a little limited, improvements in natural language processing and machine learning are allowing for more fluid, natural interactions and intelligent product suggestions that genuinely add to the user experience.

Personalisation

Dynamic content

In online retail, dynamic content – the name given to content that is presented differently depending on the context – can be combined with machine learning and customer data to create a truly individualised shopping experience.

Dynamic content personalisation in ecommerce can range from dynamic interfaces that present the user with different navigation options depending on what they previously browsed; to featuring a user’s recently-purchased or regularly-purchased items prominently; to dynamic pop-ups that present tailored offers based on a user’s location, browsing history, or stage in the customer journey.

Craft and vintage retailer Etsy uses the shopper’s name to personalise their homepage when they are logged in, giving a personal touch to the experience of landing on the website. It also highlights items that offer free shipping to their location.

Another example of simple, yet effective, dynamic personalisation is Alloy Apparel, a clothing website catering to tall women. It uses the customer’s location to highlight products that are popular in their area – incorporating an element of personalisation even when the customer has no account or browsing history.

Dynamic personalisation doesn’t need to be on-site to be effective – it can also be used in emails to tailor communication and encourage shoppers to convert or make return purchases.

Dynamically personalised emails can make use data such as purchase history, shopping cart contents (for cart abandonments), and customer details like name, location or gender. For example, Adidas’s email campaigns for its Originals series dynamically altered the products being promoted based on the subscriber’s gender to ensure they would only receive relevant product promotions.

In another example, fashion retailer Zalando will send personalised cart abandonment emails to notify a prospective customer when an item they left in their cart is on sale – tailoring the message and providing an extra incentive to convert.

Personalised assistance

While many chatbots and automated assistants will incorporate some element of personalisation, a few retailers have developed solutions that are designed to act as a personal shopping assistant, building a one-to-one relationship with the customer and producing highly personalised recommendations.

Florist and gift retailer 1-800-Flowers partnered with IBM Watson to develop a specialised gift assistant, GWYN (Gifts When You Need), in a bid to enhance its reputation as a destination for gifts of every kind. The goal was to streamline and personalise a typically arduous customer journey, cultivate a one-to-one relationship with a customer and learn from each interaction to develop the most suitable recommendations.

Gwyn uses cognitive capabilities to interpret customers’ requests and ask follow-up questions to ensure the right product suggestions are given. Eighty percent of users reportedly said they would use Gwyn again, and 1-800-Flowers saw a 6.3% uplift in revenue from using the chatbot, as well as an increase in return customers.

Fashion retailer Zalando has used machine learning and AI to create an ‘Algorithmic Fashion Companion’ (AFC) that provides shoppers with personalised styling advice. Using more than 200,000 outfits created by stylists as training data, the algorithm has been taught to identify items of clothing and assemble them into outfits, teaching itself what makes a good outfit.

The AFC designs outfits based on “anchor items” that customers have either purchased previously or added to their wishlist. Stylists regularly tweak the algorithm to keep it up-to-date with current trends and add new outfit styles, such as a holiday-oriented or seasonal look. The algorithm enables this personalised styling advice to be dispensed at scale and round-the-clock to inspire shoppers.

Forecasting & optimisation

Demand forecasting

Machine learning isn’t all about front-end or consumer-facing applications: some of its most valuable applications are behind-the-scenes, making it easier for retailers to interpret and extrapolate from large amounts of data and keep on top of trends.

Demand forecasting is an important incarnation of this capability. Demand forecasting involves using statistical or machine learning models to analyse the impact of pricing, promotions, seasonality and holidays on sales, and then translating this into decisions about stock and pricing. It allows retailers to anticipate demand for certain types of products, make decisions about discounts, and also account for the impact of marketing campaigns on stock levels.

The advent of machine learning for tasks like demand forecasting has been particularly important for small and medium-sized retailers which lack access to the data insights of their larger competitors. It allows them to manage inventory effectively between physical stores and online, stock strategically and use data to their advantage without the need for a specialist team.

Churn prediction

The ability to predict and manage what percentage of customers will churn is key for retailers with a subscription-based product. Machine learning can aid in predicting this and in identifying which segments of a retailer’s customer base are most likely to churn and give them specialised attention and potentially retain them.

At subscription service Graze, for example, customers can sign up to receive regular snack box deliveries to their home or workplace. If customers move to cancel their subscription, the brand will often offer them discounts as an incentive to stay subscribed.

By using machine learning to analyse customer data, a company like Graze can identify customers who might churn and pre-emptively offer them rewards as a ‘thank you’, or calculate which customers are valuable enough to offer multiple discounts versus those aren’t worth expending the effort to retain.

Pricing optimisation

Retailers need to pay attention to a number of different factors when setting prices. Market positioning, production and distribution costs, product availability, competition and time of year are just some of the factors that need to be taken into account.

Machine learning can be of immense help in incorporating this wide variety of variables, understanding the patterns involved, and learning from them. Much like demand and churn, pricing also benefits from the creation of predictive models that allow retailers to determine the best price for a product or service in advance. It can also help retailers to evaluate the potential impact of sales promotions or gauge the right price for a specific product if they want to sell it within a set time period.

Machine learning can also be used for dynamic pricing: a strategy in which prices are adjusted in real-time based on factors like competitor pricing, customer demand and profit margins. Pricing tools evaluate a large number of internal and external variables – from inventory and KPIs to competitor pricing and demand – to generate prices, a process that can be partially or completely automated.

Amazon is a pioneer of this technique, reportedly changing its prices 2.5 million times per day, or roughly every 10 minutes. In a bid to remain competitive, major retail brands like Walmart, Target and Kohl’s have followed suit.

Dynamic pricing runs the risk of alienating consumers who notice the strategy, particularly if they feel as though they are being cheated out of a potential discount or being discriminated against. Target was the subject of widespread consumer backlash in February after an investigation revealed that product pricing on Target’s mobile app jumped when a shopper entered a store, though the app didn’t transparently indicate that prices had changed.

Retailers also risk getting drawn into a price war if they constantly attempt to compete on pricing with a much larger competitor such as Amazon. However, when used carefully, dynamic pricing can act as an incentive for customers to shop using a particular channel – for example, in-store, if prices are lowered when a shopper enters a store – and help retailers navigate changing market forces while maintaining a profit.

Sentiment analysis & customer feedback

Sentiment analysis is another important tool in a retailer’s arsenal. It refers to the process of using a combination of natural language processing and machine learning to analyse bodies of text – such as social posts, customer feedback, or chatbot conversations – to determine customers’ thoughts and feelings about particular products. It can also inform retailers how deeply customers feel about a product, as well as which features are responsible for those feelings.

Thanks to sentiment analysis and customer feedback data, retailers can quickly understand how their products are being received, and if necessary, act on that information to make improvements and satisfy consumer demands. It can provide context for the shifts that retailers might notice in their sales, churn, or demand – they may know that these things are happening, but not why they’re happening. It can keep them abreast of problems before they arise – or help them learn about trends they can exploit and cater to.

Multiple ecommerce brands who were interviewed for Econsultancy’s Ecommerce Trends report spoke about the importance of analysing customer feedback for delivering on the customer experience and improving the customer journey. Cyril Lambard, global head of ecommerce at Nespresso, detailed how reviewing customer complaints on a weekly basis and working out how to alleviate them is a “company priority”.

“We look at the customer voice across all touchpoints, so monitoring customer satisfaction, effort score and NPS,” he said. “We also get feedback from a range of sources, and anything we do we assess with qualitative focus group to ensure we are constantly optimising the customer journey.”

How is machine learning changing bricks-and-mortar retail?

There’s little question that the applications of machine learning outlined above have effected a transformation in how retailers do business online. But what about offline?

Bricks and mortar retailers have been slower overall to adopt similar innovations, but some developments are beginning to emerge in in-store technology that promise to transform offline retail into an optimised, tailored, ecommerce-style experience.

Meanwhile, advances in inventory and merchandising that make use of machine learning are equipping bricks-and-mortar retailers to keep abreast of trends, monitor competitors, and prepare for external factors that could affect purchasing behaviour.

In-store technology

Facial recognition

Machine learning (and sometimes deep learning) is an integral part of facial recognition technology as machines are trained on visual datasets over time to identify faces accurately. In bricks-and-mortar retail, it can be used for security purposes to prevent shoplifting, but also for some more futuristic applications.

The use of facial recognition in retail is more commonplace in Asia, particularly China, than it is in the west. Thanks to two major payments providers – Alipay and WeChat Pay – rolling out facial recognition-based payment systems, shoppers at supermarkets, grocery stores and other retail outlets across China can use facial recognition to pay for goods quickly and easily.

In January, a historic shopping street in Wenzhou, a city known for its private entrepreneurship, became the first to extensively apply facial recognition in making payments. Then in April, Japanese retailer Aeon revealed plans to debut a chain of self-serve smart retail stores in China that would identify shoppers via facial recognition and display recommended items and coupons on their smartphone based on their purchasing habits and digital payment.

While markets like the US, Canada and Europe aren’t as far along as China, similar developments are on their way. At the National Retail Federation’s 2018 annual expo, boutique sweet shop Lolli and Pops demonstrated an opt-in facial recognition program that allowed in-store cameras to identify shoppers and associate them with data from the brand’s loyalty programme, allowing shop assistants to better serve shoppers according to their preferences and past purchases.

French retailer Carrefour has also initiated a trial of an ‘Amazon Go’-style automated store that uses facial recognition and register-less shopping. The store, called Flash, is currently only available to Carrefour staff at its head office in Massy, near Paris.

Retailers will need to tread carefully when it comes to implementing facial recognition, as there are obvious potential ramifications for consumer privacy, and shoppers are likely to be wary at first. Making facial recognition strictly opt-in, as Lolli and Pops did, and being as transparent as possible with how it works and what the data is used for will help to allay any concerns, particularly while the technology is still new.

Augmented reality/smart mirrors

Augmented reality relies on various forms of machine learning, including computer vision and pattern recognition, to work successfully. It has seen a few innovative applications in retail that hold considerable promise for the future.

One of these is augmented reality mirrors, also known as smart mirrors, that allow shoppers to virtually try on clothing, accessories and cosmetics and experiment with different styles, fits and colours without lifting a finger.

Beauty retailer Sephora was one of the first to experiment with this technology, launching a 3D augmented reality mirror in its Milan store in 2014 that allowed customers to virtually try on different makeup products. It has since introduced a similar concept to its mobile app, called ‘Virtual Artist’, which provides tutorials on how to achieve specific makeup looks and allows users to virtually overlay them onto a photograph.

In a slightly different application of the technology, fashion retailer Neiman Marcus debuted a ‘memory mirror’ that could capture images and videos of clothing that customers tried on and then visualise the outfits in a different colour or compare different outfit variations side-by-side. It launched another version of this concept in 2016 that could be used for trying on pairs of sunglasses.

Earlier this year, the retailer opened its first department store in Manhattan. It is packed full of digital experiences including two variations of the Memory Mirror: the Sunglass Memory Mirror and a Memory Makeover mirror that records beauty demonstrations and make-up tutorials to be emailed to customers.

Inventory & merchandising

Trends analysis

Just as machine learning allows ecommerce retailers to forecast demand, optimise pricing and adjust their inventory in real-time, it is giving bricks-and-mortar retailers access to trends data they can use to inform choices about stock, product launches and promotions.

Edited, for example, is a searchable database of product and pricing data that allows fashion retailers to make informed choices about the most commercially viable products and trends. They can compare data on competitor prices and monitor discount rates in order to price things strategically.

They can also look back over retail trends from the past five years to determine the best time to launch a promotion, seasonal update or marketing campaign. Plus, the tool incorporates a live stream of consumer sentiment from blogs and social media that can help retailers decide which trends to invest in (and which to avoid).

Another tool aimed at providing insights to fashion retailers is Stylumia. It uses artificial intelligence and computer vision to process social media impressions and brand/retail signals and turn them into actionable insights for retailers. That includes insights such as which fashions will be a commercial success, which visuals will draw consumer attention, and which garments and styles are trending in different parts of the world.

It has partnered with a range of fashion brands including Pepe Jeans, Puma, Jack & Jones, Aeropostale and Global Desi in India, and expanded into Europe with partnerships with senior fashion veterans in the UK and Italy.

Inventory management

The last thing a bricks and mortar retailer needs is to be caught out by a lack of stock. Unfortunately, it has historically been difficult for retailers to predict the peaks and troughs in demand that can impact inventory availability, particularly when external factors like weather are involved.

Machine learning is helping to alleviate this challenge by enabling retailers to stay on top of trends that could affect purchasing behaviour. In 2016, IBM acquired The Weather Company to make use of its vast database and collection systems; it has since developed a tool called Weather Signals that provides insights to retailers into how local weather is likely to affect sales and footfall, allowing them to anticipate demand and optimise store performance.

Another tool, NextOrbit, was developed to help retailers combat out-of-stocks (OOS), which are said to be more costly than losing a sale and can also impact customer trust and loyalty. NextOrbit uses predictive analytics to forecast customer demand on a granular level and predict when and why a given item will go out of stock, reportedly reducing OOS by 25% and leading to a 2% to 4% increase in sales.

Some retailers have even taken inventory management to the next level by programming autonomous retail robots to monitor inventory using computer vision and machine learning and look for patterns in product or price discrepancies.

Lowe’s is one such retailer. Over the summer of 2016, it implemented the LoweBot in 11 stores across the San Francisco Bay Area as part of a year-long pilot programme. The autonomous robot could help customers find the products they were looking for via a searchable computer display, advanced voice recognition, and laser-based sensors that helped it navigate around the shop floor.

After the LoweBot’s trial period ended, Lowes used some of the technology from the experiment, such as the computer vision, to detect empty space on store shelves and thus track gaps in inventory.

What’s next for machine learning in retail?

While there are no shortage of real-world example to show how retailers are making use of machine learning, it’s also far from the norm. Many retailers have yet to implement any level of personalisation on their ecommerce websites or consider how machine learning could improve product recommendations or add to the customer experience.

And although machine learning has given rise to many promising innovations for bricks-and-mortar retailers, only a select few brands have succeeded in implementing them.

However, the availability of machine learning is constantly improving as tech partners develop new offerings, technology is licensed, and the tools for implementing it become cheaper and more commonplace. On top of this, retailers are waking up to the need to transform their offerings and set themselves apart by offering something that their competitors can’t replicate.

Equipped with the same tools as Amazon, ecommerce companies are increasingly  recreating the personalisation techniques that made the retail giant such a force to be reckoned with and even finding ways to set the bar higher. However, the majority still have some way to go before they meet the customer experience expectations that are increasingly becoming the norm – and Amazon isn’t backing down any time soon.

Meanwhile, those bricks-and-mortar retailers that can successfully implement machine learning will have the edge in an increasingly tough and competitive sector. China is currently experiencing a bricks and mortar retail boom as retailers and tech giants alike rush to take part in the New Retail revolution, to see what it could look like if bricks and mortar retailers can truly make a success of machine learning.

In the 2010s, ecommerce began to truly come into its own as a shopping channel as internet connectivity and accessibility improved, mobile devices proliferated, and machine learning enabled ecommerce retailers to harness a customer experience that (in many ways) could outdo bricks and mortar retail. As we enter the 2020s, the same transformation may just be over the horizon in bricks and mortar retail, with machine learning (and other forms of artificial intelligence) once again at the centre.

The post How machine learning is transforming retail both online and offline appeared first on Marketing Week.

Back to live news feeds    Share news feed article