Follow the leaders: Machine learning and AI in the retail sector

Many of the real-world applications of AI are to be found in retail, but it's the big players that are making the running

Retail is the place where rapidly changing consumer demands meet just-in-time distribution and production. The sector is in a state of constant flux:

"If you think about it, the retail store that you visit today is fundamentally different from what it used to be even a couple of years ago," said Manish Choudhary, SVP global SMB products and strategy at Pitney Bowes.

"Above all else, customers are looking for ease and convenience, and this is driving retailers to think about new merchandising strategies, sourcing, supply-chain optimisation and delivery options."

Unsurprisingly then, the retail sector has typically been a leader when it comes to adopting new technology. Innovations such as big data analytics were taken up with enthusiasm by retailers, and the sector made early running in the IoT with RFID tagging. Other examples include pioneering use of the web, cloud and mobile, and marketing across multiple digital channels.

The same story is playing out with machine learning. Retailers, following Amazon's lead, have been using recommendation systems for nearly a decade now.

The trouble with AI, though, is that apart from simple chatbots and one or two other standalone solutions, there are very few off-the-shelf products that you can simply purchase and install, with most being highly use-case specific, requiring large amounts of data to create any useful benefits with side effects across the organisation that are difficult to predict.

Nevertheless, in a competitive sector like retail, there is growing interest in applying machine learning in areas such as sales and CRM, manufacturing, distribution and logistics, fraud detection and payment services.

So where should retailers start? The initiative needs to come from the top, Choudhary said.

"C-suite executives need to be front and centre when driving AI projects for their organisations. They need to have a vision and plan for enterprise-wide AI strategies before they begin to be implemented," he said.

Evanna Kearins, VP global field marketing at DataStax, agreed.

"It tends to be the CEO or COO that makes this happen. They are the ones that have to meet their targets, and they have to be able to demonstrate how they will compete with the global e-commerce players out there."

While AI may garner the most attention at the consumer end, the efforts most likely to get signed off are those with the greatest RoI, which will generally be about automating back-end systems and business processes. These will also be among the most complex to deliver.

If a coherent business strategy with backing from the board is essential, so is a commitment to data management, quality and governance. To operate with any sort of ‘intelligence', algorithms require data that's consistent, accurate, timely and clean - and lots of it. In this, the big players have an advantage as they simply have more data to start with. They are also better able to experiment with integrating machine learning into existing processes than smaller firms, which may find it conflicts with the day-to-day running of the business. Recruiting rare data science talent is also likely to be problematic.

"For smaller retailers and e-tailers, integrating AI requires transformation of core business systems to support these interventions, and that can hold them back from investing in new technologies," Choudhary explained.

The real-time requirements of AI applications can also be a challenge for those seeking to integrate machine learning into existing systems. Back-end systems generating offers and recommendations must be able to communicate seamlessly with the apps consumers use on their mobiles.

"If you can't implement a recommendation to a customer when they are actually buying something, in the moment, that result will be less helpful," said Kearins. "You have to work in real-time, and at the moment of the transaction. That real-time factor is a big hurdle to overcome."

The massive data requirements of machine learning mean that for most retailers cloud will is an essential and unavoidable component. One of the biggest cloud vendors - Amazon -also happens to be a retailer, meaning that many in the sector will not use it. Beyond that though, with large volumes of data to host and process and with cloud vendors offering advanced AI platforms of their own, lock-in is a real issue to contend with no matter who the provider is. Retailers need cloud, but they have to make it work for them, Kearins said.

"Getting to this stage needs an understanding of multi-cloud and hybrid cloud, and that involves getting your technical experts and your line of business teams together. That kind of collaboration is essential."

So, plenty of challenges remain, and the big ecommerce players are currently making the play. But there are already any number of examples of AI being used in retail, according our correspondents.

Supply chain optimisation can to reductions in surplus inventory stocks and cost savings. Data science and predictive analytics can provide retailers with a deep understanding of their customers across multiple touchpoints. On the manufacturing side, robots that can adapt to changing conditions can be significantly more efficient, and retailers like Ocado have already demonstrated the effectiveness of robot-staffed distribution warehouses. Meanwhile large ecommerce firms are looking at ‘hyper-personalisation' through customer engagement chatbots and omnichannel support.