AI Realities & Futures In Retail

by Hugh Williams

The retail sector has often been regarded as slow when it comes to the adoption of artificial intelligence. However, an industry that once seemed unsure how best to leverage this technology, is slowly getting to grips with its proper application. In this piece, Peter Ellen, co-founder and CEO, Big Data for Humans takes a look at the purpose, and deployment, of AI in retail.

It’s confession time. Two years ago I kicked a robot hard; but I believe there were mitigating circumstances. It was a proto-type delivery robot; the love child of Henry the vacuum cleaner and a self-driving picnic cool box. I mistakenly back-heeled it as it approached me from behind. I was speaking at the Wired Retail Conference and its inventor was a little upset. I apologised to both of them, then pondered how it might perform making grocery deliveries on busy downtown streets, and what loss-prevention strategies would reduce the chances of a [robot] kidnap.

This didn’t seem to be the immediate future of AI.

The retail sector has struggled with the deployment of AI, hindered by confusion over how it can be harnessed to change the game. A recent report from McKinsey considers that retailers are the laggards in AI-adoption, with financial services, high tech, and communications businesses taking a significant lead. The heritage of these vertical markets goes some way to explaining why subscription-based businesses find it easier to use data-driven intelligence, as analytics models and data science have been essential tools for decades. Meanwhile, retailers have been traditionally product-driven, transaction-focused businesses where the majority of data resides in product, supply chain, and EPOS systems. However, Amazon is blurring the lines between subscription and transaction businesses with its Prime service, and makes massive investments in AI.

A revolution

Retail disruption is moving the status of AI from vanity project to an essential part of the customer-centric toolkit. With increasingly complex channel habits, and game-changing new entrants re-orientating their business functions around specific customer groups [see], retailers are deploying AI to power central decisioning, merchandising, and marketing personalisation at scale. This is less about AI-powered robots and more about automating powerful systems for actionable insight.

Brandon Purcell, senior analyst at Forrester Research, notes the difference between traditional DIY analytics workbenches and DIFM [Do-it-for-me] technologies that automate actionable insights for specific sectors like retail. The need for automation is driven by the scarcity of skills on the one hand, and increasingly complex multi-channel customer behaviour on the other. In this case, AI is not a straight human replacement because traditional methods simply can’t perform the job fast enough. With AI, marketers can keep up and personalise at significant scale.

At a recent fashion retail event we hosted, executives from a luxury department store group noted that the launch of a hotly anticipated new beauty brand suddenly delivered a box-fresh generation of customers to their stores that they hadn’t seen before. This required rapid decisions on how to retain them beyond a mono-brand relationship. Young customers often start as brand or product patriots, but offer great long-term value, provided the retailer meets their changing tastes and growing incomes. Understanding these new customer requirements requires systems of insight that can keep up. This is where AI steps in to support smarter merchandising and marketing personalisation across digital and physical channels. Automation removes many of the constraints of traditional DIY insights so that businesses can focus on automation and personalisation at scale. Combined with cloud computing, much more can be achieved from self-learning systems that move fast and hunt new sales opportunities. In the future, they’ll become as common as cool boxes at a picnic.

AI Deployment – A Greek and Roman Style Problem

My home country of Scotland is the claimant to many inventions of the modern world, and if you visit the bars of Glasgow, it’s not too hard to find a patron who will proudly reel them off into the small hours. Some would have you believe there’s some kind of genetic magic that produced this super human invention conveyor belt. Others more credibly cite the nexus of a competitive education system that encourages people to think out-of-the-box. What’s unquestionable is that most Scottish inventions went on to be exploited elsewhere.

It’s worth bearing this in mind when thinking about the AI revolution that is upon us. Inventing a proprietary system of insight could fill you with misty-eyed pride, but working out how to use it at scale is more likely to bring in the dollars. People of a scientific bent love tough problems, but leaders have to ask themselves if deployment strategies make sense when it comes to business outcomes. There’s a significant risk that AI becomes another wave of under-used technology.

The Harvard business review notes that around 70% of corporate data scientists’ time is spent working with IT teams trying to operationalise their findings in the business. This is the root cause of employee-retention challenges, as scientists are the ancient Greeks that really enjoy hard experiments and lose interest rapidly when placed in Roman-style development teams.

At Big Data for Humans, we’ve benefited from an altruistic approach to AI and we are world class at operationalising large-scale network maths. Take our client AirAsia. Our team calculated that a laptop would take 120 years to build the graph of their passenger behaviour that our platform performs in just a few seconds. But we temper our pride in that capability, because it’s not really the purpose of our efforts. It’s more important that a platform can support a sequence of models, AI, and machine learning to produce actionable insights for marketing folks. We are as happy to share the maths with inquisitive data science people as we are to support others to operationalise their projects. You won’t find our team leaning on the Horseshoe Bar claiming to have invented n-dimensional cosine matrices. What they’re really interested in is providing timely central decisioning that joins up teams in marketing and operations to drive revenues.

For leaders who see potential in AI, the key challenge is clearly defining, and often constraining, business problems then looking at outcomes required so that projects remain under time and cost control. This tends to be easier if you start by defining the daily business outcomes required and testing that they resonate with end-users. In our case, we deliver a simple app that marketers use to run omni-channel campaigns and measure revenue uplift across all customer channels.

For thought leaders in data, the opportunity is big since the emphasis has progressed from analysis to prediction, prescription, and now the stewardship of automation. As a result, we see a future where our marketers become more strategic, capable, and profit-focused.

I, for one, will drink to that.