The use cases, challenges and benefits behind retail AI

A tale of customers and data

At the launch of Ocado Technology’s educational online game AI:MMO, Paul Clarke, chief technology officer (CTO) at Ocado, said AI technology can help retailers make “smarter use of scarce resources” and that firms don’t have a choice about whether or not to adopt it.

As previously reported by Computer Weekly as part of its CW@50 anniversary coverage, knowing your customer 50 years ago meant a local merchant manually making sure the right stock was in the right place at the right time – a friendly face-to-face service for a small market.

Decades on, retailers now collect so much customer data from so many different people that it is impossible to offer this personalised service without the help of technology.

As pointed out by Brian Kalms, partner and retail lead at consultancy Elixirr, some retailers have so much data there is no longer the option of analysing it by hand, especially when adding new online ventures into the mix. 

“Historically, when you went into a store, you didn’t identify yourself,” he says. “Online brands know who you are, so retailers are going to have to learn to be data-savvy, and that’s one of the first applications of AI – it’s been in the form of bots and communications, and it’s moving into data analysis.”

Utilising data to understand customers

Where retailers used to categorise their customers in a “simplistic way”, now data can be used to better understand customers individually.

For example, using old customer demographics based on socio-economic background, earnings and gender, a consumer who buys high-end food but “value” tissues should not exist, be we know this isn’t the case.

“People don’t shop their demographic, they shop their behaviour,” says Kalms. “Some of this information has taken a while for organisations to find.”

With digitally native businesses such as Asos, Ocado and Amazon acting as market disruptors which deeply understand their customers, he says it is becoming more difficult to justify spend on physical retail locations, forcing retailers to explore other ways to engage customers who increasingly expect more from the retail experience.

“Customers will be expecting that of everybody, and if you’re offering a 48-hour stay-in-all-day type service, you’re finished,” he says.

Digital retailers quicker to adopt new technologies

But the difference between digital-first retailers and traditional bricks-and-mortar retailers is that digital retailers have always had these technologies in mind, and find it easy to adopt new ones as they come along.

“They live only in the digital world, so they view AI not as a thing to go and find out about – it’s just built in to everything you do. It is the business,” says Kalms. “That’s probably the biggest split in retail at the moment – the split between the legacy businesses and the digital-first businesses.”

For larger firms, experimenting with and adopting technologies such as smart mirrors, data analytics and AI can lead to increased purchases and fewer returns. But Kalms says there is a difficulty integrating these experiments into the core way of doing business, something that is much easier for digital-first retailers.

As Kalms put it, larger retailers have a hard time trying to “redirect the mother ship”, having a past of being “product and service innovative” as opposed to agile and digitally innovative.

Some retailers have made this jump successfully – video games retailer Game, for example, took advantage of the data it had and used it to develop a personalisation project to adapt both its online and in-store offerings to a modern digital audience.

But Boohoo.com’s head of IT operations, Steve Roberts, says so many projects of this nature fail because people “haven’t actually worked it out” – they race to adopt the technology but they don’t really know what to use it for or how it fits into their wider business.

“It sounds great, doesn’t it? It’s obvious there are lots of buzzwords in the industry and AI is probably one of them at the moment,” he says. “I think a lot of people don’t understand how to use those results, and they end up with a bit of tech that’s quite smart technically but not so smart commercially.”

The use cases and the challenges

Roberts lays out some of the common use cases for AI in the retail industry, including using machine learning for fraud prevention or using customer data for personalisation.

Another retailer claims it can come in useful for predicting the volume of calls coming into customer help centres, store footfall or website traffic volume, and adapting accordingly.

Boohoo.com is trialling the use of AI for chatbots that will answer some of its customers’ more frequently asked questions, such as where orders are or how to return a product.

For the brand, it reduces costs as a person and does not need to be involved in answering those questions – and customers are more satisfied because questions are answered quickly.

Boohoo worked with a third party visual search firm called Syte to implement this. Roberts says working with third parties can help retailers solve problems they may not be able to fix themselves.

Past experience

In the past, much like government, retailers saw technology projects as a huge cost, and in many cases they would fail, leaving retailers out of pocket and with no solution for the problems they faced.

But even when working with a third party, can a retailer really rely on the data it has collected to fuel any kind of AI technology?

“Can it join it all together? Can it understand the data we’ve got and learn from it? It might be that we’ve been provided data that looks fine to a human, but not quite right to a machine,” says Roberts.

“It’s still relatively immature as a technology, but it’s an area that will rapidly gain traction, so I expect that maturity to increase a lot.”

While machine learning and AI is most commonly focused on using data to improve customer experience and add a personal touch, Andy Britcliffe, group technology director at Holiday Extras, points out that as simple as that sounds, it can be just the “tip of the iceberg”.

For example, when using AI in a call centre, Britcliffe says: “With machine learning, there’s a potential to get better at that modelling so we have the right people on our phones when people call us.”

Knowing how to collect data

But knowing how to collect data and what data to collect can be make or break for these systems. 

Britcliffe says “classic machine learning problems” are dependent on “good quality data”, as well as having good software engineers and data specialists who can make sure the data is collected and sorted correctly while being secure and anonymous.

In an attempt to bridge this gap, firms of all types are on the search for data scientists – and in the summer of 2018, retailer Marks and Spencer partnered with Decoded to teach its employees how to better use data.

Stepping outside of a data analysis use case, Britcliffe says other AI technologies that could be considered are voice or vision, or using open source frameworks for deep learning – all of which have the potential to be of benefit to a retailer when implemented correctly.

Overcoming more challenges to find the benefits

Even once a retailer has decided to use AI technology, knows where it fits into the business and what it will be using it for, and has decided how to implement the technology – whether through a third party provider, in-house or experimenting with relevant startups – there are still more challenges ahead.

In many cases, legacy systems stand in the way of using AI and other technologies in the way a business intended, according to Diana Parker, senior director for retail, consumer goods and transport at Microsoft UK.

“There’s a lot of legacy estate out there that organisations need to put their arms around and work out the best and most effective way to get hold of that data,” she says.

For example, when implementing websites or e-commerce, there are often two systems that are then not integrated, so there is no single view of stock. Fixing things like this isn’t just a case of shoving all of a firm’s data in the same place.

Increasingly, to tackle these issues, retailers are choosing to partner with startups, or to take a “lab” approach to new technologies to determine how they can be adopted.

In many cases, this depends on the attitude of the business. Parker says the organisation’s culture can be a challenge when trying to adopt new technologies of any kind.

“It’s not exclusive to AI, but if you’re going to make a big, tech-based change, you need to take people with you,” she says. “Be really clear about why changing the way you work is going to be of value to you as an employee. Create a culture where people are able to participate in change.”

Considering initiatives as business projects

Rather than just being an experiment or an IT project, these initiatives need to be considered as “business projects” that contribute to the business as a whole.

There is also the issue of AI systems bias. This can apply not only to societal biases that affect the choices the AI makes, but also systems making poor decisions based on poor data. “AI is only as good at the dataset you asked it to reason over – if your dataset contains bias, it goes into the decisions you make,” says Parker.

With all of these roadblocks in the way, retailers might think they are better off without these systems.

But Toys ‘R’ US, House of Fraser and BHS are just some of the retailers which have faced trouble because they failed to change.

Parker says AI is not just “a new shiny fad that will go away”. And retailers know it – according to Fujitsu research, 95% of retailers understand that AI and other emerging tech will affect the sector.

The benefits of machine learning

For those willing to put the effort in, AI can make a significant difference to a business; Asos has used machine learning and AI to make recommendations to 15 million customers when browsing on the site.

Morrisons has used AI for demand forecasting and customised stock based on its stores’ local demographics, maximising sales and making savings in the supply chain.  

Overwhelmingly, retailers, technology providers and experts in the field alike all advise that the best thing to do is to “give it a try”.

Karina van den Oever, who began her career on the retail shop floor and is now a partner at Elixirr, says any retailer which thinks it can avoid AI adoption is in denial. “Retailers are in denial, and I think it’s because they’re so distracted by the operational activity,” she says.

“What retailers need to be doing more of is just experimenting. Will drones deliver? Will there be store robots? We don’t know, but if you don’t experiment now, you’re going to be behind the curve forever.”

Tapping into the tech ecosystem

Where retailers used to want to build systems like this themselves because they may be at the core of the retailer’s unique selling point, Van den Oever says now is the time to build and tap into the tech ecosystem and not go it alone unless you have the capability to.

“For every problem a retailer or company is trying to solve, there are five startups out there trying to solve the same problem,” she says. “Tap into that disruptive tech before it disrupts you. Find that solution that could potentially be the future.”

Whether taking a small step or a big leap, it’s clear that benefits such as increased sales, better understanding of customers and cost savings are on the horizon for those retailers that are ready to put the required effort into AI adoption.