Originally running in our 7th Edition Report, ex-Editor, reporter and contributor to WhichPLM, Ben Hanson, discusses the theme of ‘intelligence’. This piece accompanies a dedicated series from Ben around intelligence, A.I, and data-driven design and development in retail – all of which you can find in our 7th Edition.
[For any reference to timelines throughout this piece it should be noted that the publication date for our 7th Edition Report was August 2017. Similarly, you will find references to other ‘features’, which denote to the other editorial pieces in our 7th Edition Report.]
Out of the hands of research scientists, ready to start delivering new revenue streams and optimising existing ones, A.I. is currently in a tricky transitional phase between promise and proven deployment. Much like the Internet of Things (which was the subject of WhichPLM’s 6th Edition publication), A.I. is not widely understood – even by some less-than-scrupulous companies who are trying to package and sell it. As a result, with limited grasp of what this class of technologies can do and even how it all works, the market at large is unlikely to unite behind the promise of A.I. until it sees demonstrable, repeatable results. This leaves software companies and their early adopter clients to forge ahead alone, anchoring the sheer, world-changing clout of A.I. to smaller, more specific, more achievable aims. No small task.
Already, though, real-world applications of A.I. in the retail, footwear, and apparel industry are beginning to coalesce around two major themes: a cycle of heightened customer insight leading to an improved customer experience and back again; and a better-informed (or even partly automated) design and development process that increases the likelihood of the best possible products, perfectly positioned, reaching the market and selling at their target price.
Building on the A.I. primer of the previous feature, the following pages will examine different subdivisions of these top-level themes, looking for concrete examples of where retailers, brands, and technology suppliers have been able to translate theory into results.
Before analysing those specifics, however, I want to build a better picture of how A.I. – in the umbrella sense – is currently treated from a commercial and investment perspective at the highest possible level: the global economy.
While much of the potential of A.I. technologies remains unproven (but no longer speculative) leading nations have already begun to stake their claims to being either A.I. hubs, with strong startup cultures and tax incentives, or sources of the next generation of technology talent. Or both. These are not investments to be made lightly, so as you might expect, these countries are banking on the emergence of multi-million-dollar industries (at the very least) centred around A.I. in the very near future. As an example, The Vector Institute is an initiative part-funded by the Canadian government and backed by big names like Google, Accenture, Uber, Deloitte and Shopify. Its stated aim is to make its native country (and its home city of Toronto) the de facto destination for A.I. leadership. The Vector Institute is already operational, working to retain homegrown talent in-country, and to attract international assets away from the constant tech-drain of Silicon Valley, in the hope that the next big A.I. company will be Canadian.
Interestingly, though, countries may not have to actively participate in creating a welcoming home for A.I. for the technology to have a net positive impact on their futures. By dint of its ability to automate both repetitive and complex manual and mental tasks, Accenture predicts that A.I. will add up to two full percentage points to the economic growth rates of the USA, UK, China, the Nordic countries and many more by 2035. The report containing this research also estimates that labour productivity across these countries’ entire economies could rise by anywhere from 25% to 37% when compared to non-A.I. baselines.
From a sovereign perspective, then, the case for investing in A.I. and the culture that surrounds it seems cut and dry. A.I. applications are all but guaranteed to be a part of your country’s future. But is the same also true for businesses? Among the biggest and brightest, who is betting big on artificial intelligence, and why? The answer is essentially every internet company above a certain size, irrespective of industry, and for a huge range of different reasons. We have already established that Google Photos serves as a strong example of slightly secretive, narrow A.I. at work, but attached to this is the larger idea that Google is now, for all intents and purposes, an A.I. company. As well as its Photos application, the company’s entire search and advertising business relies on deep learning in very narrow niches. At the same time, its marquee mobile messaging client, Allo, is actively advertised on the fact that it invites an A.I. assistant into your chats – one who can translate messages on the fly, recommend places to eat, locate your friends on a map, and even auto-suggest replies for when you need to send a snappy rejoinder on the go.
And it certainly is not just Google betting the farm on A.I. So too are Apple and Amazon, who, together with Google, occupy the top three spots in Internet market capitalization, with a combined value of almost $2 trillion in early 2017. Tellingly, as well as their core online services and applications, these companies all sell devices that serve as Trojan horses to carry their A.I. assistants into our daily lives. On the surface, Siri, Alexa and Google Assistant want to get better acquainted with us in order to serve us better, but in a world where simple CAPTCHA forms are already being used to collect training data, it is naïve to expect these assistants’ parent companies to have totally altruistic motives. Is it a coincidence that Amazon, arguably the world’s biggest retailer, with a vested interested in knowing as much about us as possible, now has an ANI interface sat on the kitchen countertops and nightstands of more than 10 million homes in the US alone, according to a report on the Amazon Echo installed base issued in May of this year?
But interest in A.I. is not exclusively coming from big businesses; deep learning and other umbrella technologies are also the driving forces behind one of the technology industry’s most feverish startup races in recent memory. In WhichPLM’s home country of the UK, where government funding to the tune of almost £300 million ($395 million) was just announced to fund new research into disruptive technologies, the two-year period from 2014 to 2016 saw a new A.I. company founded every single week, according to research conducted by MMC Ventures. These startups are all working to develop real-world use cases for A.I. that run the gamut from the predictable to the outlandish, and while some – maybe many – will fail, a single successful idea, pitched at the proper market, is often all it takes for a new technology to suddenly become elevated to the next big thing.
Of these A.I.-focused startups, the same research shows that nine out of every ten were serving a B2B audience, which should come as little surprise given the money and mindshare needed to launch a B2C product. While large, pre-existing technology companies have been able to overcome these hurdles, most also established their A.I. capabilities through acquisitions. Google’s parent group, Alphabet, acquired DeepMind in 2014; Apple bought Siri in 2010; and Amazon has just this summer announced the acquisition of Graphiq Inc., whose A.I. products will be used to improve Alexa’s data analysis capabilities.
The B2B nature of the A.I. market should also seem logical when we stop thinking of the technology as an intellectual curio or wild invention, and instead begin to consider it – correctly – as a new and extremely powerful entrant into the enterprise technology market.
As with PLM, ERP, and other big business solutions that began their lives in the aerospace, defence, and automotive industries, the bulk of the B2B A.I. market will not be retail to begin with. But this is likely to change – just as it did with PLM, where fashion and consumer packaged goods are now considered key verticals for most vendors – in very short order. Recent research by UK institution Drapers and A.I.-powered eSales platform Apptus revealed that more than 90% of retailers – from a panel of 80 senior executives surveyed – are already excited about the potential of A.I. solutions to deliver real, measurable value in merchandising, customer acquisition and engagement, supply chain efficiency, and a raft of other business functions.
As readers of a PLM publication will know, though, an appetite for new technology does not always translate into adoption – particularly where that technology is industry-agnostic rather than being designed from the ground up to meet the demands of fashion and retail. Like any other enterprise solution, prospective customers of A.I. products and services will be looking at the market through their own unique lenses. First, they will want to identify the broad areas of their business that an A.I. initiative could impact, looking for a technology partner who understands the industry. Only then will they further filter the solutions and services on offer with shortlists of functionality and expectations of results calibrated by the most pressing challenges their businesses face.
Just like any other category of solutions, to succeed, A.I. products must be designed to fulfil both general market needs, and to pass through these smaller funnels of more specific demand. So, before we delve too deeply into how A.I. vendors are addressing those narrower business challenges, let us examine how a technology with virtually limitless cross-vertical potential is being tailored to address the top-level concerns common to essentially every retailer.
Recapture the essence of retail, and go real-time
“A hundred years ago, retail was a neighbourhood business, where the retailer typically owned his or her own store, knew their customers by name, and could make very personalised recommendations,” explained ShiSh Shridhar from Microsoft, when I asked where that particular technology giant saw the primary value of A.I. for our industry. “Because of the scale at which many retailers are now running, though, it’s anything but a neighbourhood business today, and they realise that they can’t possibly know their customers with that level of intimacy any more. To reclaim it, retailers need to work with data analytics and machine learning to identify the patterns that people are not capable of spotting in today’s sea of data, and use that to bring back the personal relationships that defined the old-fashioned village store – scaled to meet the needs of a much larger-scale, modern operation.”
“As a store manager today,” ShiSh continued, “I still want to be able to make personalised, relevant offers to you, but the difference is that now I’m doing it because I have a wealth of data about you behind the scenes, and a machine learning program has analysed that data. So while I don’t know you first-hand, I know from my intelligence that, based on your demographic and segmentation, buying patterns, and purchase history, I can recommend something that’s uniquely tailored to your personal tastes, and that will be relevant to you in the moment.”
This desire to reclaim the insight enjoyed by the single-destination retailers of old is not a new one. WhichPLM wrote about it as a key theme of the NRF Retail Show several years ago, and it has likely been on the mind of every retailer ever since the combination of offshore production and online retail and marketing made the bulk of apparel production, sales, and marketing happen at arm’s length, out of direct sight. Until recently, however, it has been difficult to conceive of a practical way to recapture what was lost in the shift to international manufacturing and eCommerce. The critical data required were either scattered between different, disconnected silos, or were simply too overwhelming in volume or velocity to begin to make sense of – particularly in the case of larger retailers operating across different continents and channels.
“The opportunity for brands to collect the right consumer data, incorporate it into their decision making processes, and then to act on it has, until recently, simply not existed at the scale we needed,” said Matt Field, Founder and President of MakerSights, a machine-learning platform that provides what the company calls ‘actionable product intelligence’ to brands like True Religion and Ralph Lauren. “Before machine learning and A.I., businesses would instead have had to rely on much limited and time-consuming exercises like focus groups or in-store testing to gather that kind of insight. Using these approaches, they may have been able to engage with tens or hundreds of customers at most, over a period of several weeks. From that limited sample size, they would then have tried to form conclusions about why things were trending, and what customers thought about key products and lines. Today, brands and retailers can leverage machine learning to do the same things, but on a much larger scale, and in a window of time that actually fits with their production cycles.”
This, I believe, is where the potential of A.I. for fashion at a broad level really begins to crystallise. Time to market remains one of the major challenges for brands and retailers who need to capture consumer demand and react to it with almost no margin of error. This problem is by no means unique to fashion, but it is perhaps at its most acute in our industry; in the process of creating footwear, accessories, and clothing that consumers want to buy, when they want to buy it, every minute counts. If machine learning can deliver critical market insights more quickly, it has the potential to make a significant difference not only to initial merchandising, planning, and design processes, but every other subsequent stage of the product lifecycle – culminating in a measurable improvement to a business’s bottom line.
“Retailers now have the chance to engage with thousands of customers in a matter of hours, on channels like social media that they already use on a regular basis,” continued Field. “With the help of A.I., they are then presented with a far more detailed, far more representative of their chosen market segment on the same day, and they can then incorporate those insights into their decision-making without delay.”
With this critical common element in place – the ability to move quickly and comprehensively understand a complex market – A.I. has, for me and likely many others, passed that initial barrier of fashion suitability, and concrete results will follow. From here, brands and retailers will begin to turn to more specific applications and specialist A.I. suppliers. The remainder of this feature is given over to examining a selection of these individual use cases – beginning with initial market research and trend analysis, and proceeding through to transforming the customer experience at the point of sale and beyond.
Map the market, track trends, and seize strategic opportunities
Identifying looming trends or anticipating entire new ones is something the industry tends to think of as being instinctive. Good buyers and merchandisers are able to distil the essence of catwalk and trade shows into perfectly-pitched collections, and good designers can create styles that just seem to naturally tap into the zeitgeist. Over time, though, as the volume of information involved in identifying and tracking trends has ballooned, businesses have begun to recognise the importance of marrying the art of these processes with the science of data analysis.
In this sense, market analysis and strategic trend service EDITED is an exemplar of the new face of fashion technology, employing experts from both sides of that art / science divide to tackle broad and specific industry challenges.
“Our company is unique in that we have world class computer scientists working side by side with industry-experience retail professionals to build a powerful piece of technology, which solves fashion retail problems in intuitive ways,” said Julia Fowler, EDITED’s Co-Founder, and a regular fixture in WhichPLM interviews. “To accomplish what we’ve set out to do – help retailers see every product launch, price shift, and market event online as it happens – you need both sets of experts working together. There’s no way around it.”
As an established business, EDITED already boasts an imposing client base comprised of brands and retailers who otherwise may not have a defined A.I. strategy, meaning its solutions and services will be among the earliest experiences that many companies will have with the power of machine learning. “Using machine learning, A.I., and image recognition technologies built into our software, retailers can take advantage of customer’s shifts in interests to stock and sell the right products at the right time,” added Fowler. “For example, we saw huge success in the massively popular activewear category, where our technology is helping global retailers quantify the importance of trend and offer the exact activewear products that their customers want. By providing that degree of accuracy, we see A.I. as the winning ingredient that will allow our clients to gain a competitive advantage in a crowded retail space.”
EDITED also provides what might be termed a soft introduction to A.I. There is nothing, technologically-speaking, that prevents another algorithm from taking that quantified trend actually creating the beginnings of an activewear category by itself, but advertising this kind of functionality at an early stage of A.I. adoption may be ill advised. While Fowler is keen to talk about the technology that underpins its recommendations, EDITED is not overtly an A.I. company, and its solution leaves critical decision-making firmly in the hands of the retailer or brand. This is understandable because, for many customers, trusting an A.I. to actually make automated decisions during what remains a creatively-led process is currently a step too far – and may remain so for some time.
This is not, however, an opinion that every business shares, and Cosabella’s Courtney Connell believes that delegating at least some limited decision-making to A.I. will be essential to truly carving out a competitive advantage in the longer term. “Working with A.I. is going to require a shift in mindset from many people,” she explained. “Today, when you mention A.I., someone will inevitably say “that’s going to give me so much more information that I can use to make better decisions,” and while that’s true, they’re overlooking the fact that A.I. can, and in some cases should, be making those decisions on their behalf. To put it bluntly, if you keep thinking of A.I. as just a really awesome analytics platform, you haven’t understood its full potential yet.”
Collaborate with your customers and lock in loyalty
A big part of having the right product in the right place at the right time is starting from a strong impression of what that product should be. Obviously, broad trend prediction and market analysis are critical, but A.I. technologies also present the opportunity for retailers and brands to go beyond the broad strokes and engage with their customers, collecting invaluable information that will directly inform design and development.
On the surface, though, this seems to run counter to what we’re often told are the defining characteristics of today’s shoppers: fickle, disloyal, difficult to please, and willing to drop their passion for your brand in a heartbeat when a better price exists elsewhere. You could be forgiven for thinking that these were not sensible targets for any kind of collaboration, but Matt Field of MakerSights disagreed, calling this a misrepresentation of what loyalty actually means in 2017. Where previously a retailer could expect to create loyal customers by having great products and good service, he said, today they must seek active engagement with the customers or risk losing them to competitors who make them feel more valued.
“Analysts who talk about disloyal consumers aren’t necessarily incorrect, but they are talking about a world where brands are either incapable of differentiating themselves on engagement and experience, or are choosing not to do so,” Field said. “In that situation, customers will indeed go where they find the most obvious points of differentiation, and apart from in the luxury industry, the first and most prominent point is generally price. Where I’d strongly disagree with those analysts, though, is that I believe there is a significant opportunity, with A.I., for brands to reengage customers in new ways that re-engender loyalty. It’s not that customers are inherently non loyal; they just have not been engaged in a way that influences their behaviour away from price sensitivity. Our research shows that more than 50% of millennials – and even more from older generations – do actively want to co-create with brands if the opportunity is presented to them.”
Like the desire to achieve a more intimate knowledge of a customer base, this kind of co-creation also has some strong historical analogues. “To see this ethos in action, look back to the early 2000s, when Nike ran their Nike ID program, asking consumers to create their own versions of shoes,” adds Field. “It was wildly successful, and people lined up around the block to participate. That’s an example from more than fifteen years ago of how to thoughtfully engage customers in a loyalty building exercise, and the tools we have today in machine learning and A.I. are so much more capable than the ones Nike had at their disposal then. And what’s maybe even more important is that Nike didn’t just attract new customers and strengthen their relationship with existing ones; they also collected millions of data points on customers’ preferences at very specific item levels like material, colour, print design and so on. That has no doubt proven incredibly useful for powering broader design activities in their standard categories in the years since.”
Better data for buyers, designers, and merchandisers
Where customers are not invited into the design and development process, the major influence on what products brands and retailers actually bring to market remains in-house merchandising processes. As Ganesh Subramanian from Stylumia explained, though, these are often informed more by intuition than by true intelligence, with merchandising teams working from small pools of historical information, making them a prime candidate for A.I.-driven improvement.
“Most design and buying decisions in fashion are largely based on a very small amount of data about what the brand or retailer has done in the past, and what parts of that worked or did not work,” Subramanian told me. “That’s the intelligence base that most businesses have to work from, which they supplement by either recruiting or contracting third party experts whose job it is to help them make better bets on the future, based on a combination of that small data pool and their own experience and intuition. That always seemed to me like a very speculative way to run a business, and when we looked at what proportion of those brands’ and retailers’ products sell at full price, we discovered that figure has held at around 50% for decades. There are obviously exceptions, but across the overall industry, that means one out of every two products is sold at a discount. So, designing and merchandising the traditional way, what’s the chance of any individual fashion product selling without a markdown? It’s very close to a coin toss.”
Like many of the other solution vendors we interviewed for this Report, Stylumia was founded on the principle that both a wider data collection net and a more intelligent approach to analysis were the keys to improving on processes that had historically been something of a bottleneck in the product lifecycle – not replacing them. “We recognise that forecasting remains an important part of the apparel business, and we are certainly not saying that intuition is a bad thing,” Subramanian clarified. “A decade ago, small data pools like these were all retailers and brands had; it was difficult to get to know your customer and understand their preferences. Today, though, customers are easier than ever to track online. They leave their likes and dislikes lying around the Internet in structured and unstructured form, and A.I. technologies allow us to consume and use all of that information in something approaching real-time. Deep learning lets us listen to what customers are saying and learn from it; computer vision lets us see what they are seeing; and A.I.-created knowledge graphs can help us drive insights from those signals. Combining traditional intuition with A.I. really presents a totally new way to understand what your customer wants.”
Pricing, on point
As Subramanian points out, one of the primary criteria used to judge product success in retail is full-priced sell-through. Achieving this demands the right products in the right assortments at the right time, but it also hinges on a keen, up-to-the-minute understanding of customers’ price sensitivities, as well as the pricing and promotional structures being used by the competition.
Talking to Cheryl Sullivan, Chief Marketing and Strategy Officer for Revionics, she revealed that pricing is not something businesses can leave to assumptions or guesswork. Sullivan’s company suggests that 93% of shoppers use a digital device to browse and research while shopping, 90% leave stores to go and buy elsewhere, and approximately 70% of promotions do not achieve their objectives. In this context, it is little wonder that machine learning has the potential to redefine the way we think about price, and Revionics’ solutions promise to leverage machine intelligence to optimise customers’ initial pricing, markdown schedules, and promotions.
Interestingly, Revionics offers different levels of A.I. engagement, with data science and automation employed to varying degrees depending on how far the business at hand is willing to part with their traditional financial planning methods – with the deepest level doing essentially everything but changing the price tags itself.
“The simplest form of what we do would be classed as analytical insight,” Sullivan explained. “We capture real-time inventory, pricing, and assortment information on a retailer’s competitors, and provide our clients with the ability to search and query it through intelligent dashboards. From there, we move to the level that I call predictive, which means taking that data, testing it based on different hypothetical scenarios, and being able to predict the outcome in terms of margin, revenue, customer loyalty and other metrics. Finally, we have the prescriptive level, and this is where machine learning really enters the picture and the science goes into overdrive. Rather than just providing a retailer with insights and predictions and allowing them to take whatever action they think best, the solution will instead tell them what to do in detail, instructing them on the exact promotional offers to make, and even suggesting dynamic pricing, where a price point changes in real-time according to demand.”
Stand out from the crowd
In an extremely competitive market, some companies may be able to differentiate themselves from the competition purely through price, but for many brands and retailers, razor-thin margins afford little wiggle room. Instead, the search is already on for other ways in which A.I. can be used to help businesses stand out from the crowd.
“Today’s retail environment creates the need to be a rapid innovator, in products and experiences,” explained Steve Laughlin, General Manager for IBM’s Global Consumer Industries, when we spoke. “Companies need to be constantly monitoring what the marketplace and their consumers expect, and quickly coming up with ideas – not just for new products, but for changes to their services and business models. And they need to be able to test those ideas rapidly, and gauge their impact through a range of different channels.”
Needless to say, given his position, Laughlin understands retail, and he is acutely aware of the challenges that come with any attempt to do truly new things in a digital age. “Creating standout experiences also creates an explosion of what we call unstructured data. While structured data fits into spreadsheet columns and rows, unstructured data is more like a mass of images, video, text, social media entries and so on. Now that data is filled with valuable insight and information, but you have to be able to unlock it. That’s why investing in A.I. – our version of it being cognitive computing and Watson – is so important. A.I. can digest that unstructured data, make sense of it, and create use cases that can really transform the way brands and retailers interact with their consumers, helping them to stand out from the competition.”
“If I put myself in a brand’s shoes,” Laughlin continued, “I can now have customers query my product catalogue using natural language – asking the A.I. to suggest a red dress for a summer cocktail party. Or I can even allow them to upload a photo they snapped from a magazine with a smartphone, and the A.I. can then suggest the closest match from my complete catalogue.”
What I found especially interesting from my conversation with Laughlin and other senior technology figures was that A.I. is credited with being simultaneously responsible for improving existing processes and experiences, as well as creating entirely new, hitherto impossible, ones. This, I believe, will make technologies like machine learning especially attractive to fashion retailers. The market has become so homogenised, and so cutthroat, that when it comes to responding to competition, retailers often have to choose one or the other – and doing existing things more efficiently usually wins out.
This is a view shared by Raj De Datta, CEO of BloomReach, which has created what it calls “the first open and intelligent Digital Experience Platform” targeted at improving customer experiences and personalisation, and which counts names like Neiman Marcus, Nordstrom, Williams-Sonoma, and Forever 21 among its clients. “I’ve sat with a lot of large retailers, and I believe that most of them are still struggling to identify the true source of their competitive advantage,” De Datta told me. “Because how can they not ask that question when Amazon sells many of the same products they do, delivers them faster, with a great shopping experience, and isn’t saddled with the same costly real estate footprint that they are?”
“Realistically,” De Datta went on, “I think there are only a couple of possible answers to that question. Either these retailers create a product selection that nobody else in the world has, which is easier for speciality retailers. Or they try to create customer experiences that are fundamentally unique. The difficulty is that neither of those differentiators is easy to deliver at scale; retailers simply cannot hire enough human beings to achieve them, so much of the customer journey is fragmented across mobile apps, websites, stores, and social media, while retailers typically also have disconnected technologies, different data silos, and different teams working in isolation.”
“This is where machine learning and data science come in,” De Datta concluded. “A digital experience platform can power every interaction between a brand or retailer and their customer – from informing merchandising , to building marketing campaigns, to supporting the experience a shopper has in-store. For us and our clients, A.I. is going to be absolutely essential to delivering competitive advantage in a crowded market.”
With all of the preceding elements taken care of – on-trend product, perfectly mixed and properly priced, with a shopper experience that catches people’s attentions – we can begin to look at far more measurable metrics.
“A lot of retailers’ attitudes to eCommerce have focused on driving more people to their websites, rather than actually improving conversion rates,” Andy Narayana of Sentient told me, referring to the proportion of website visitors who actually go on to engage with the company or make a purchase. “The biggest barrier to converting a visitor into a buyer is that people are often not able to find the products they actually want; their experience is being constrained by the database centric way we built web applications in the late 1990s and early 2000s. We believe one of the best ways that A.I. can deliver additional revenue for retailers and brands is to improve the experience of online storefronts with a paradigm shift; instead of the retailers dropping products into their channels and hoping that someone will click on and buy them, we want shoppers to interact with the A.I., and for the A.I. to pull the products it knows they’re likely to want, with context taken into account. We know from experience that conversion rates among shoppers who engaged with the A.I. in Sentient Aware and Sentient Ascend are around 30% higher, and that average dollar values of shoppers who interact with the A.I. are up to 15% higher.”
Another noteworthy change, powered by improvements to online shopping experiences and the deployment of Internet of Things technologies like beacons and trackers into physical stores, is the ability for retailers to see a blended view of past customer behaviour, and for an A.I. to recommend – and even take – actions based on that intelligence.
“There is a huge amount of data that businesses have only recently gained access to because of beacons and other IoT technologies,” explained ShiSh Shridhar from Microsoft. “We can now collect behavioural patterns and traffic information to see where people dwell in physical stores as well as online storefronts. And on the opposite side, we can also achieve some level of insight into abandoned baskets – what was picked but never made it to the point of sale – in bricks and mortar locations.”
“Machine learning is what will allow us to find patterns that cross the barriers between the two, and allow us to actually make use of that information to try and influence behaviours,” Shridhar added. “With the right insights, how can we, the retailer, convince you, the shopper, to revisit the display you dwelled at? How can we get you to the next stage of the customer journey? In both cases, A.I. can help suggest or take action. For instance, our purchase history data tells us that you bought a certain product a few weeks ago, and our in-store sensors are now telling us you’re standing in front of a product that matches it. Machine learning can infer from that that you’re likely looking for a matching product – consciously or subconsciously – and our recommendation engine knows that the product you’re now near ranks highly on the list of possible matches. Based on those patterns, the A.I. tells us that you have a 90% affinity to buy that product, so we can push you an offer for 20% off and see if you’ll convert.”
Cross-sell and up-sell
Of course, convincing customers to buy something they are already considering is only the beginning. In a market suffused by collections, complementary pieces, and even more capsule collections, the RFA industry thrives on crossselling and up-selling. Unfortunately, both of these promotional methods are most effective when a retailer has a large, diverse, product range to draw related recommendations from – which places them firmly in the category of things Amazon is able to do better by sheer brute force.
Unable to compete by having product catalogues anywhere near as large, though, apparel retailers have a chance to do things differently and more personally, using machine learning and computer vision to examine behavioural patterns and product characteristics, and then making recommendations that are informed by current context.
“There have been many different approaches to create the perfect eCommerce recommendations systems, but historically they have all started with hard-coded algorithms that simply identified similar products based on their colour, fit or material – or they have simply looked up where previous customers have bought one product and then another,” explained Daryn Nakhuda of Mighty AI. “With machine learning, though, you’re able to pull in so much more signal, and take advantage of a much fuller range of data and context. For example, a simpler recommendation system might serve up a few tops that match a particular black skirt, but it will not take account of the context in which the customer is shopping. Has she just finished browsing blazers and other workwear? If so, then the top you recommend should not be a summer tank; it should be something more formal, recommended within the context of the customer’s current browsing experience.”
“There’s a classic cautionary example from the consumer electronics industry that works equally well in fashion,” continued Nakhuda, expanding on the reasons that traditional recommendations systems have fallen short. “A customer has previously bought some batteries, so the next time they visit, a simple system recommends more batteries for them. An A.I. system, on the other hand, would understand that those previous batteries were bought to support a toy or some other device, and would be able to identify an appropriate direction to steer recommendations towards as a result. And context also extends to appropriate price: if a customer has only previously bought some $5 socks from your web shop, you’re unlikely to get an up-sell to a $2,000 suit that just happens to go with the socks.”
This is an approach that Eric Brassard of Propulse endorses, but he places greater emphasis on the distinction between archival data and real-time information – the latter of which he believes is the key to making automated product recommendations that align with personal taste, not just purchase history.
“Product recommendation engines are not new,” Brassard explained. “But until now they have all worked on the same simple premise: that by analysing big data we can see patterns of purchase and, from them, make predictions about what someone is likely to buy in the future. The fundamental assumption here is that past behaviour is all you need to predict the future. We set out to build a different kind of recommendation engine, because we believe that the past is not enough of a base from which to make appropriate recommendations in the present – certainly not when it comes to visual products like fashion, furniture, art, and accessories.”
At this point, Brassard’s model for better product recommendations begins to sound very familiar, relying as it does on replicating the role of a real human being with on-the-ground knowledge of their customers and products. “Instead of purely historical data, we use A.I. to try and mimic the performance of outstanding salespeople – the kind who intimately understand what you want even as you’re picking clothes off the rack,” he added. “Our approach is to use visual recognition to build a model of personal taste informed by history, but not totally reliant on it. Rather than looking for exact matches between products, our AI looks for common elements that suggest an appeal to certain tastes. That’s what you might call a human-like aspect to A.I., or a soft skill, which is only possible to deliver at this scale with the help of A.I.”
Even with the best possible products, in the optimum mix, priced according to bulletproof machine intelligence, and recommended to the most relevant market segments, a final hurdle remains before a collection can really capture consumers’ attentions: marketing. Easily the equal of its impact on actual shopping behaviours, the rise of eCommerce has had a tremendous effect on the way that brands and retailers advertise. According to the Wall Street Journal, luxury brands increased their online marketing budgets by more than 60% in the three years from 2013 to 2016, and a similar shift in promotional spend was also likely to apply to other market segments.
Like other key stages of the product lifecycle, however, much of modern marketing – even cutting-edge digital content creation – is still directed by human intuition. To improve on the performance of their online marketing, lingerie brand Cosabella turned to Albert. Billed as “the first Artificial intelligence marketing platform for the enterprise,” Albert (formerly known as Adgorithms) was given access to Cosabella’s existing campaigns, and was, according to Courtney Connell, able to deliver results in social media that far exceeded those the company had been able to achieve through traditional methods.
“Now we have Albert, I’m not sure I’ll ever have another human manage my Google AdWords or Facebook campaigns again,” Connell told me. “Give it access to your campaigns and the A.I. ingests everything you’ve done to date, spends some time optimising that, and once he’s finished fixing your current content, he begins to create his own campaigns – adding keywords, taking keywords away, creating new micro-segments and so on. And then, with even more time, Albert starts making real recommendations like “your competitors are running promotions and you aren’t,” or “social ads are fatiguing at a rate of two weeks, so we need to produce new content every fortnight.” Every piece of this is new, actionable insight, and it’s had a dramatic impact on our KPIs.”
Find fit and reduce returns
As important a metric as full price sell -through is, understanding retail performance and consumer satisfaction does not end at the point of sale – at least not the first time. Every sector of the RFA industry is plagued by returns, with online customers routinely ordering multiple sizes of a single product and returning the ones that do not fit. Similarly, consumers also have clear expectations of consistency in fit, and one of the holy grails of consumer satisfaction is a shopper with total confidence that whatever he or she buys, it will fit the same way as the last product they bought from the same company.
Footwear fit may seem, on the surface, to be less subjective than apparel fit. Shoe sizes are clearly graded for different markets, and none of the vagaries of small, medium, and large apply. Nevertheless, as Timo Steitz, CEO and Co- Founder of ShoeSize.Me explained, machine learning has the potential to permanently solve what has become a major problem – with return rates higher than fashion in many cases – in his native Germany and beyond.
“Worldwide, we see footwear return rates from 10% up to 70%,” Steitz told me. “In German speaking countries, returns for both fashion and footwear average out at 50%, and our research tells us that the latter is a lot higher. A recent survey revealed that as many as 80% of all Germans are wearing the wrong shoe size if foot length is used as the sole measurement. We also know that between different styles and different brands, actual sizes can vary by as much as three full sizes for the same kind of product. Footwear sizing also differs dramatically from country to country, brand to brand, and style to style, and the combination of all those variances can be extreme. We call the result ‘sizing and fit chaos’, and we believe that data holds the key to solving it forever.”
As for how ShoeSize.Me uses machine learning? “We collect data from three fields: product data, user data, and production and sales data,” said Steitz. “Product data encompasses material, form, category and so on. User data refers to information about the customer, such as their age, gender, body metrics, and whether they consider their foot to be small, medium, or wide. Then we look at how people buy: what articles they purchase, what they keep, what they return, and why. By uniting those three worlds into a single dataset and allowing our algorithms to run on it, we can really begin to make sense of how shoes fit, give customers the confidence to shop, and provide brands with the insight they need to create better products.”
Build better relationships
Today, as we have seen from previous examples, shoppers can be convinced to look beyond pure value if they leave a retail experience feeling as though they have been personally valued. The majority of retailers currently try to create this feeling of reciprocity by offering incentives for repeat spending (the loyalty card model), but these can often feel impersonal. Only rarely do retailers have the level of insight and segmentation required to present consumers with offers or rewards based on their own personal spending or behaviour patterns. But with the help of A.I. technologies, this level of personalisation could be readily achievable, helping retailers to increase their chances of retaining customers without resorting to bulk discounting.
“With A.I. and cognitive computing, the opportunity exists to personalise relationships with consumers – not through a points card, but through relevance,” explained Steve Laughlin of IBM. “Because you, as a brand or retailer, better understand your customers, you understand their past interactions with you, you understand their personalities, and you can therefore shape your communications in a way that’s going to resonate the most with small groups of people – or even individuals. That sort of relationship is sticky, and it fosters a different kind of loyalty that should endure.”
Retailers and brands who do obtain that level of insight into individual customers will also have the opportunity to use it in other areas, to deliver a consistent level of personalisation (or at least familiarity) across channels, as Daryn Nakhuda from Mighty AI explained. “Fashion and retail are the perfect stages for conversational bots and A.I. assistants, equipped with real insights into consumer behaviours, to flourish. In real life, a personal shopper is only as good as how well they know you; the same will go for an A.I. assistant, but with the key difference that they can get to know you a lot faster and a lot better if you give it access to the right personal information.”
While one of the major applications for chatbots is in customer support, Nakhuda also pictures a near future where – just as IBM and The North Face have done with Watson – conversational interfaces come to the foreground of eCommerce. “I think this will completely change the entire online shopping experience,” he said. “Rather than going to a retailers web store and browsing for evening gowns or a wedding outfit, we’ll simply tell the chatbot our names, and it will know enough about our personal preferences to ask a few situational questions before recommending a product based entirely on what it thinks we – as individuals, not as a market segment – will want. In the future, our calendars may even be linked, so the underlying A.I. will already know that the wedding is in Dallas, in August, so it’s going to be hot – or that the couple’s wedding website has some suggested colours. Either way, the A.I. can then apply that information against a retailer’s product catalogue and make recommendations that will be so much more compelling than just showing a sidebar ad for something that’s similar to things we’ve looked at before.”