AI in retail: Survival depends on getting smart

The retail sector is the poster child for the use of artificial intelligence. Self-driving delivery robots, automated warehouses, intelligent chatbots, personalized recommendations, and deep supply chain analytics have been making significant impact on the bottom line — if you’re Amazon.com.

Other retailers, however, are struggling to adapt. In fact, only 19 percent of large retailers in the U.S., UK, Canada and Europe have deployed AI and are using it in production, according to Gartner.

[ Cut through the hype with our practical guide to machine learning in business and find out the 10 signs you’re ready for AI — but might not succeed. | Get the latest insights with our CIO Daily newsletter. ]

That number will change quickly over the next few years, with 31 percent of retailers piloting AI technologies, and another 27 percent planning to deploy by the end of 2020. “There’s an awful lot of experimentation going on,” says Gartner analyst Bob Hetu.

Retailers that are embracing AI do so to lower costs, and increase revenues. And while the upfront investment is an obstacle for many companies, not investing in AI could be deadly.

“Over the next five years or so, AI is going to become one of the key elements of differentiation for retailers,” Hutu says. “It’s really urgent that they get proficient with it.”

“We consider AI to be an absolute building block, not an optional choice anymore,” agrees Suketu Gandhi, partner in the digital transformation practice at A.T. Kearney. “If you’re not using it, you should be counting your days.”

Efficiency improvements

According to IDC analyst Jon Duke, retail and banking are the two industries most aggressive about AI investment. “Retail in particular is widely understood to be one of the most promising industries for applying AI,” he says.

Today’s use cases largely comprise automation of administrative processes, repetitive process, and low-level decision making tasks, Duke says, such as automated customer service agents.

Home improvement retailer Build.com is one such organization deploying AI to boost efficiency. Patrick Berry, the company’s senior director of technology, credits machine learning with freeing employees from mundane tasks, enabling them to focus on higher-level problems.

For example, the technology is reducing hold times for customer service calls. “When we have a better idea of what a customer is looking for or wants to do based on the actions they take on our website, we can give them what they came for more quickly,” he says.

Machine learning-based systems are also having an impact on application deployment at Build.com, where the technology is used to establish baselines and detect anomalies. “We’re saving a lot of time not having engineers watching logs or monitoring systems,” Berry says.

The company is also looking to use intelligence in code review. “Automated reviews will free up valuable engineering time by taking many of the mundane reviews out of the queue,” he says. “This will allow our teams to increase delivery velocity without sacrificing quality.”

According to Gartner, 88 percent of retailers cite cost savings as the biggest impact of AI, and efficiency-related AI technologies are among the most frequently cited for planned use, including robots in the warehouses (59 percent), AI for fraud or anomaly detection (56 percent), and delivery robots (45 percent).

But only a small minority have these technologies in production. Just 19 percent of large retailers currently rely on AI-powered fraud detection. Robots in warehouses remains primarily a pilot play, with 41 percent in the testing phase.

“There’s a lot going on,” says Gartner’s Hetu. For example, Fedex and Amazon and even Domino’s Pizza are all experimenting with delivery robots, and UPS just got FAA clearance for delivery drones. “This could move more quickly than we might imagine.”

A smarter supply chain

Retailers are also turning to AI to improve supply chain management, with 77 percent of large retailers making use of the technology, according to Coresight Research.

Key opportunities for improvement include inventory management, demand predictions, and pricing calculations, as markdowns cost U.S. nongrocery retailers $300 billion last year, or about 12 percent of sales, mostly due to bad inventory decisions leading to too much product, or the wrong type of product, according to Coresight.

Gartner found that 64 percent of large retailers are using or plan to use AI for product development and selection, and 60 percent are using or planning to use AI to improve pricing.

And in a survey of 200 retailers, 86 percent had identified specific use cases for advanced analytics in decisions such as which products to buy and how much inventory to keep on hand, according to Coresight.

Enhancing the customer experience

Customer experience is a key focus today, and retailers are looking to leverage AI and ML for product recommendations and to optimize promotions. Personalization and virtual assistants are also viewed as technologies that can fuel revenue growth.

“Consumer expectations have changed,” says Arpit Jain, vice president of cross-functional delivery and capabilities at Nerdery, a digital services consultancy. “We have higher expectations, and we want lower friction.”

Chatbots are one example of this, he says. Another must-have is being able to present the right content to consumers where, when and how they want it. “Four or five years ago, this was a nice to have,” he says. “Now, it’s a necessary thing.”

At Bluestem Brands, for example, AI is used to improve search in an effort to present the most relevant products and services to customers. The Minneapolis-based ecommerce company, with $2 billion in annual sales, has seven branded websites, including Fingerhut, Haband, and Appleseed’s.

“AI and ML are powerful tools for discerning correlations and patterns in our customers’ search and purchase behavior,” says Jacob Wagner, the company’s IT director. AI-powered tools help Bluestem better understand what customers find desirable, and then create models to predict which new products or services to suggest, he says.

For example, if a customer searches for something, say, “jeggings,” that the search engine hasn’t indexed, it assumes the customer meant “leggings” and gives those results instead. When the customer scrolls down the results and clicks on denim-colored leggings, the system takes note.

“Now we have a signal that this specific product is a match for the term ‘jeggings’ even if the product itself doesn’t use that term anywhere,” Wagner says. “This is classic signal stuff.”

The latest AI technology — specifically, convolutional neural networks — takes this up a notch.

“We can look at the images on the product the user clicked on,” he says. “We can then use similarity scoring against all the other products’ images.”

For example, the system might find all the items shaped like pants that are denim-colored, and add those to the recommendations. “The next time a customer types ‘jeggings,’ the products that user clicked on using that term can pop up first, the products with similar images second, and the rest of the leggings third,” he says. “We’ve now trained the engine how to associate many products with a term it’s never heard of using nothing more than a single users’ behavior signal.”

Bluestem is also looking to use AI to identify shipping locations where customers aren’t getting their orders because “porch pirates” are stealing them with the idea of offering better protection services for those packages on checkout.

Wayfair is another ecommerce company using AI technology, including computer vision, to improve customer service. The home goods retailer, which had $6.7 billion in revenue last year, serves more than 15 million active customers.

“In categories like furniture and decor, it can be difficult to precisely describe an item in mind, especially in a way search engines can understand,” says Dan Wulin, the company’s head of data science and machine learning. “So we built an AI-powered visual search tool.”

For example, Wayfair recently announced a “search with photo” function that uses smartphone photos to help users find products. The app, released in early November, includes an augmented reality tool that enables customers to preview of how furniture will look in their homes.

“Because our catalog is so big and there’s such a big variety of products, and our customers shop in such a visual way, we have to use AI in a different way to be successful,” he says.

AI and machine learning are so important, Wulin says, that Wayfair has 2,300 engineers and data scientists working on applying AI techniques to business problems. One area they’re working on is using computer vision and natural language processing together to create new models that better mirror how the brain learns, he says.

But at what cost?

Few retailers can afford to hire dozens of data scientists, let alone thousands. So for most retailers, the large initial investment in AI is prohibitive, despite its long-term upside.

“Retailers are thrifty,” says Kim Knickle, retail industry specialist for digital innovation at consultancy Insight. “Unless they know that it’s going to directly impact the top line or bottom line, they’re cautious about what deciding which IT investment they’re going to make.”

Moreover, many still need to take basic digitization steps before AI can even be applied.

Take Seattle’s Ste. Michelle Wine Estates, for example. It’s the third-largest premium wine company in the U.S. with more than 1,000 employees and 14 brick-and-mortar locations, plus additional distribution through pubs and its ecommerce site.

The winery had three separate sets of data about its customers from the three different channels, says CIO Joe Gregg. The first step on the path to AI has been to solve that data problem, he says. “We’re really just beginning that journey.”

The company was well aware of the fact that it wasn’t going to be competing against Facebook and Amazon in the AI department. “We can hire the best winemakers in the world, and we do,” he said. “But we’re never going to be able to hire the best data scientists.”

So Gregg opted for Microsoft’s Dynamics 365 Commerce product, which is built on top of Dynamics 365 Retail. The rollout is expected to be completed by the end of the year. Once the data issue is solved, the winery will begin to use the platform for predictive analytics — for example, to recommend new types of wines to customers based on what they’ve purchased before.

This exemplifies another trend: As AI becomes commoditized, it becomes much more accessible for smaller retailers.

“As the technology is more available, specialty retailers are starting to get an advantage,” says Tracie Kambies, principal and lead for the US retail analytics and information management team at Deloitte Consulting. “It’s just been a matter of time” to make the technology more accessible and cost-effective, she says.

One development that hasn’t been much of a hindrance to the adoption of AI is the increase in privacy regulations. In fact, they may even be helping.

“The privacy regulations actually provide some guardrails on how you do this,” says Ray Wang, principal analyst and founder at Constellation Research.

Europe’s GDPR and California’s CCPA have made it easier for retailers to gather and analyze customer data, he says. “Because now you have some ground rules.”

More on AI and machine learning:

• A practical guide to machine learning in business
• Reskilling IT for the AI era
• 9 IT projects primed for machine learning
• The future of ERP is AI
• 5 artificial intelligence trends that will dominate 2018
• 6 secrets of successful chatbot strategies
• How AI is revolutionizing business productivity
• Machine learning success stories: An inside look
• 9 machine learning myths
• 10 signs you’re ready for AI — but might not succeed