Domino’s uses artificial intelligence to determine its pizza delivery times and, in Australia and New Zealand, to monitor quality. Sephora’s ColorIQ scans shoppers’ skin to provide custom recommendations for foundation and concealer. Walmart’s Intelligent Retail Lab is experimenting with a number of artificial intelligence tools. Then there are the ecommerce retailers who rely on AI, sometimes in partnership with humans, to conduct visual search, provide product recommendations and chat.
If it seems that AI is somewhat at an inflection point, that might be a smart guess. But it might not be the be-all, end-all solution that it appears.
Sucharita Kodali, a principal analyst with Forrester in ebusiness and channel strategy, says the many facets of AI can leave retailers’ heads spinning. “Retailers don’t know what is possible or what is most valuable,” she says. “Many are chasing an AI strategy, when they should be looking to solve problems first by whatever means work best.”
AI’s best deployments, she says, include customer-facing solutions, as well as those for knowledge workers and store associates. “The second — optimization of data, data visualization, finding exceptions — is by far the most valuable,” she says.
Brian Kilcourse, managing partner for retail market intelligence firm RSR Research, believes retailers “are starting to understand some of the business use cases that are going to drive its adoption. They’ve probably been using some AI in the past without realizing that they’re doing it, personalizing the offers to consumers in the digital space.” Some software-as-a-service offerings like Salesforce, Demandware and SAP Hybris employ AI in their predictive analytics, Kilcourse says.
Doug Stephens, founder of retail consultant firm Retail Prophet, also believes AI is in the “very, very early days. We’re really only scratching the surface.”
With all this scratching, what lies beneath? It helps to look at the various sectors of the supply chain — from product idea to customer delivery — to explore the opportunities.
Demand and disruption
Currently, the best uses of artificial intelligence are behind the scenes, Kodali says. “Robotic process automation and low code are things my colleagues have been talking about that are very effective. It’s not really sexy though.”
Basic, perhaps, but potentially the most potent — especially given that modern supply chain technology is decades old. “Inventory tracking continues to be a mess,” Kodali says. “If every move I make can be tracked, why can’t we track all the movements of any item in a store, even upstream from its production? I think marketplaces aren’t incented to invest in this. Otherwise, it would have been done by now.”
Put Kilcourse in the camp of those who envision the possibilities for more AI in the supply chain, through demand forecasting and product visibility.
“Retailers are starting to see that they need to be able to model demand throughout their supply chain and to be able to deal with disruption more than they ever have before,” Kilcourse says. “Having said that, retailers — except for the top ones — are struggling to find talent for an internal team to bring that kind of data science on board. There is a great push to find solution providers with AI embedded. It basically democratizes the capabilities that AI brings to the table.”
The coronavirus revealed just how much the supply chain relies on China, even if manufacturing doesn’t. Kilcourse pointed to threads manufactured in China before garments are made elsewhere in Asia. Consumer electronics’ reliance on Chinese components also is significant. That comes as suppliers have been lessening the number of days in the supply chain to respond to shifting demands.
“The downside is when you have a disruption, you have to look elsewhere and really fast,” Kilcourse says. “Coronavirus is just one perfect storm of an example. It may be a political action, a pandemic or weather. Retailers are realizing they need to not just respond in a quick way but also be able to model scenarios so that they can be ready for these disruptions when they occur.”
The Amazon effect
No conversation about high-tech retail can omit Amazon. “Amazon is first and foremost a data and technology company,” Stephens says. “Retail simply turned out to be a convenient means of obtaining and leveraging consumer data at scale.”
Still, it is proving a useful tool to help “calculate how many delivery drivers are needed in a particular geographic area at any given time, to use an inordinate volume of data of sales in specific geography to deploy the right workforce, at the right time so that every customer gets the package they were promised,” Stephens says.
“On the front end, Amazon has been using AI to formulate recommendations for customers for years. When the rest of the retail industry was trying to figure out what the term ‘AI’ meant, Amazon was using it to generate customer recommendations.”
It also was, rather successfully, convincing customers to use AI in their homes through Alexa-enabled devices. Stephens believes that goal isn’t about shopping, but about data. “Only about 6 percent of households that have an Alexa device are using it for shopping. That objectively qualifies as a failure on Amazon’s part,” he says.
“Yes, in the long run, Amazon would love to think that everyone is using Alexa to shop but that’s not the real endgame. All the other uses — the queries, the questions, checking the score in the game, the weather — all of that data can begin to inform Amazon’s strategy and choices. It’s not simply about order data but being privy to some of the most intimate queries that customers are making.”
Amazon is hardly the only retailer using artificial intelligence. Stitch Fix provides another use case, pairing the tool with human beings. “AI can do a lot of the heavy lifting and take a tremendous amount of data that no human being could ever be capable of finding,” Stephens says. “But Stitch Fix uses a degree of human creativity applied by a designer. AI can take a lot of menial, repetitive data work off our hands and allow humans to focus on what makes us unique — creativity.”
Ulta and Sephora are frequently mentioned as leaders in using tools like AI, and for good reason. “The sheer volume of products in the beauty sector is daunting,” Stephens says. “If you’re a beauty company, to make recommendations, there are so many variables that apply in that category, everything from skin tone to the climate that someone might live in and their own personal style. You match that against the volume of choice and selection, it makes it really ripe for AI interventions.
“If you’re Sephora and can quickly take a customer from 3,000 shades of lipsticks down to the three to five shades that will really work, you’ve achieved an enormous task and an effective use of that technology.”
Still, there may be room to learn; Kodali puts much of AI-enabled beauty as “gimmicky.”
Chatbots are “still a very basic and narrow form of AI,” Stephens says. Kodali agrees that this technology is “still nascent. It’s OK for highly structured interactions, but anytime there is an exception, which is often, a human still needs to be involved.”
Perhaps not for long. Stephens recounts a demonstration he witnessed in which a technology developed by IBM Watson and Soul Machines created an avatar “that looks incredibly lifelike and human,” Stephens says. The avatar and a human had a three-minute, natural language conversation about choosing a credit card. “Ultimately, the AI made a precise recommendation based on the questions asked,” Stephens says. “It was far and away much more advanced than anything I’d seen.”
Back to the supply chain
These days, even the name might need rethinking. “Supply chains aren’t really chains, but ecosystems or networks,” Kilcourse says. “Retailers have to stop thinking about a chain and think about a network that behaves like an ecosystem. They have to be able to constantly monitor and adjust their supply chain strategies.”
AI can play a role, particularly in helping retailers meet demand, which may not necessarily be in the store. “This is changing a fairly straightforward supply chain model to a complex one,” Kilcourse says. “You have to be able to model various scenarios and switch scenarios whenever the situation on the ground demands. This is giving rise to two important concepts: the AI-enabled supply chain and a digital twin, which is a precursor to AI in the supply chain. You have to be able to see something that you can model. This is getting a real hard look by Tier One retailers.”
It also means helping retailers anticipate that local assortment — something that even the biggest companies are exploring. Kilcourse points to Walmart, “the last company in the world I would say has a localized format. They are using next-generation demand forecasting to understand demand and deploy the greatest assortment for the customer and profitability. They’re trying to lower their overall investment in inventory and making the investment much more efficient.”
Solving this problem may be the ultimate key to survival, especially as customer needs change, Kilcourse says. “If they’ve learned anything in the last 10 years, NRF’s Big Show has shown that consumers are doing a lot of the legwork to determine what their lifestyle problem is and the potential solutions before they ever set foot in a store. They don’t want to be convinced to be shown things that don’t matter. Personalization in that context is really important.”
While it may be obvious to see the role that personalization plays online, Kilcourse notes that it’s time for this to come into bricks-and-mortar stores, too. “Retailers are using AI engines to be able to sense your behavior, or put your shopping experience into context, into a lifestyle problem you’re solving and put appropriate messaging in front of you that will enhance your experience.”
Geolocation and inner-aisle marketing play a role in this. “If they happen to know that you’re in aisle 14B and by looking at your basket can tell what kind of a meal you’re preparing, they can send information on how to make that meal spectacular and may offer it as a special price,” Kilcourse says.
“All of that is driven by modeling technologies that can associate you and your shopping behavior to a model that they want to promote.”