How will artificial intelligence affect the retail sector?

“People who liked that, like this”

Online shopping started to diverge from the offline experience when Amazon introduced its collaborative filtering technique, matching a user’s selection of a product with other products selected by other users who had also bought that first one. The system was one of the first attempts to mimic online the function of a physical sales assistant: inconspicuously to help the shopper spend more in store, but had the added advantage that whereas modern shops have a low ratio of assistants to shoppers, every single visitor to a site benefits from those timely prompts. Even without using the full power of artificial intelligence (“AI”), but purely passive data analysis and response, its recommendations were sometimes uncanny, though sometimes hilariously off. Nevertheless, it drove Amazon’s turnover up 35%.

“People who liked that, like this” 

Online shopping started to diverge from the offline experience when Amazon introduced its collaborative filtering technique, matching a user’s selection of a product with other products selected by other users who had also bought that first one. The system was one of the first attempts to mimic online the function of a physical sales assistant: inconspicuously to help the shopper spend more in store, but had the added advantage that whereas modern shops have a low ratio of assistants to shoppers, every single visitor to a site benefits from those timely prompts. Even without using the full power of artificial intelligence (“AI”), but purely passive data analysis and response, its recommendations were sometimes uncanny, though sometimes hilariously off. Nevertheless, it drove Amazon’s turnover up 35%. 

 Beacon systems attempt to replicate this level of one-to-one engagement in real shops, and do involve AI functions. Recognising the customer entering a shop (through RFID on a mobile device or loyalty card, for instance, or in future by pure facial recognition), they track his or her direction of travel; predict what category of goods is most likely to be of interest; and make targeted offers accordingly. The need for customers to actively opt-in by downloading an app and enabling Bluetooth in-store is a significant limitation, however. In 2014, only 44% of consumers entering a shop with a smartphone had Bluetooth enabled. Retailers are also concerned by the risk of appearing to invade customers’ privacy by bombarding their smartphones with coupons as they browse. John Lewis and Mothercare have eschewed the advertising function, instead using beacons largely to provide information and alert staff to help customers navigate. 

Both these systems will get ever more sophisticated as the data available to them increases, enabling retailers to capture more of a given shopper’s budget. 

Know your customer 

Genuine artificial intelligence (“AI”) is dependent upon vast collections of data. Rather than merely applying a filter, however, AI can identify patterns and learns from them much like a human researcher studying social trends or stock exchange movements. The offers presented to users online and on mobile can thus be increasingly personalised – including knowledge or estimates about not only the consumer’s history, needs and interests but also their budget, time or storage space constraints, for instance. Data privacy issues are clearly fundamental. Bearing in mind the amount of behavioural information (such as voting patterns) which can already be inferred simply from a person’s home address, adding individual data from online buying patterns enables AI apps to learn fairly precisely what basic goods someone is likely to want or need and when. (Target in the USA can already predict from her purchases whether a woman is pregnant and when the baby is due.) The question then is how to influence their choices: of brand, and of source (physical or online). 

Kraft’s iPhone Assistant attempts to drive the choice of brand before the shopper has even left the house. It learns from the user’s behaviour to infer, for instance, how many people need to be cooked for. The app can indicate where the nearest grocery with the right (Kraft) ingredients in stock for a downloaded recipe is, and suggest what can be cooked from the leftovers.

 In the face of competition from the producers themselves, retailers need increasingly sophisticated strategies to bring customers in-store, and an AI approach to in-store service may be more effective than a rapidly changing band of sales assistants making fallible, human judgments about the consumer in front of them. On- to offline retailers such as cosmetics subscription service Birchbox, which has now set up a physical shop in New York, bring the AI-learning based approach with them into the real world. A large touchscreen and iPads make up an important part of the shop’s customer interface. In such a scenario, the traditional shop assistant role may still be needed – a shop where each customer collects an iPad on entry and is then left completely alone might feel rather unnatural – but evolve to include something more of an audio-visual technician’s input, mixing and presenting content according to the AI analysis of the customer’s digital interactions both on- and offline. 

It’s not just the act of buying

Birchbox’s move also reflects the fact that shopping is only partly about finding and buying the right product – clearly, that can already, for Birchbox subscribers, be done online. Another component is the experience of browsing, looking at things the shopper is not presently looking for. Sharing that experience is also an important form of social engagement with friends and family. Shopping at this level is a form of entertainment, like visiting a museum where everything is part of the gift shop. AI may ensure that the customer leaves having bought all the products they either need or want, but it cannot substitute the shop’s function as a place to hang out, enjoy temptation and simply let curiosity loose.

 But given the costs involved in maintaining shops, does providing showrooms partly for consumers’ amusement make sense long-term if the majority of actual buying will ultimately take place online, a substantial proportion driven by AI-optimised product cross-selling? 

 The answer may still be yes. Granted that AI may be effective at maximising the customer’s spend, the first step – getting the consumer to choose which shop, on- or offline, to visit – requires brand recognition and loyalty, a different challenge. Physical engagement with a place and an experience is one of the most effective forms of learning; the role of the sales assistant, always in part a performance art, may shift still further towards entertaining the customer. The question then is what terms are appropriate for the customer’s access to the space, if actual buying is not expected of them. Will entry be conditional on membership of an online loyalty scheme, or subject to signing up to one? Shops may even have to move towards the differentiations now presented by airline lounges for different classes of scheme members, or work out how to deliver graded experiences within the same space. Indeed, some retailers may move towards an integration of the showroom function into a broader leisure and entertainment offering. The terms needed to manage that will need to address a range of issues very different from today’s store card schemes. 

Behind the scenes, AI can also help in developing a brand. By analysing data from across social media it can identify what a brand currently means to its target audience, and what issues drive brand-related actions such as sales. For instance, Virgin discovered that news about its space-related activities kept consumers tuned to it as a technology brand even though its sales were in far more conventional sectors. That opens opportunities for marketing better aligned to consumers’ image of the brand, increasing the potential for brand loyalty and advocacy. Maintaining appropriate protection for brands in such a constantly evolving relationship with their markets will require a sophisticated and forward-looking intellectual property strategy. 

Just in time 

That ‘last mile’ of the process, where the consumer selects and pays for products, has had the most attention but AI will affect the rest of the retail ecosystem as well. 

 Whilst Amazon managed to raise eyebrows with its proposed “predictive shipping” patent granted in 2014, in reality this only added another set of datapoints – consumers’ web behaviour – into a process which is already routine, namely ensuring sufficient supply is available locally of goods for which there is a forecastable local demand. AI set to work on a sufficiently complete database will be able to improve on current levels of demand forecast accuracy, and AI systems are already being developed to optimise logistics planning. By eliminating redundant product journeys and warehousing this should reduce waste, as well as storage and transport capacity demand. But collating the database to enable really accurate forecasting will be the challenge: a database of “anonymised” data about shoe sales, for instance, including all units sold by all retailers, would be far more effective than any single retailer’s data alone. However, agreeing to pool data between competing suppliers could involve anti-competitive conduct, so must be approached cautiously.

In a world where demand forecasting is really precise, and 3D printing is able to replace the current production process for low volume products, a whole new approach to manufacturing will be needed. Fluctuating consumer demand will be satisfied by AI-powered “agile retail”. Companies may face the associated marketing, commercial and legal changes needed sooner than they think!