FIGHTING BACK
Jetblack, a personal-shopping service, used to develop AI to compete with tech giant
Walmart’s personal-shopping service, Jetblack, is its latest weapon in war against e-commerce giant Amazon,
Epiphany Davis arrived at work in lower Manhattan on a recent morning, consulted her cellphone and set off by foot in search of products ordered via text message by wealthy New Yorkers.
From her company’s loft-like headquarters, Ms. Davis walked to a health food store to get SmartyPants Kids vitamins, but the variety was out of stock. Checking her cellphone often for instructions, she walked to a grocery store for a single bag of Guittard milk chocolate chips. She rode the subway to a Nespresso store for three boxes of coffee pods, then walked to Bloomingdale’s to pick up a $245 navy blue MZ Wallace backpack.
Ms. Davis works for Jetblack, a personal-shopping company targeted at mothers launched last summer by a surprising newcomer to the field—Walmart Inc. A few hundred shoppers in New York City pay $600 a year to order anything by text message except for fresh food. Members were invited by Walmart, or referred by current members, and need to have a doorman to join.
Their orders go to Jetblack headquarters where dozens of agents sit at computers and field requests, from reordering diapers to making suggestions on high-end cribs, organic snacks and yoga attire. Couriers fetch the items and bring them back to a Manhattan delivery hub, where they are wrapped in black packaging and hand delivered, usually the same day.
It’s a labor-intensive operation that loses money. But making money isn’t the goal, at least not right away.
Walmart executives are betting the upstart becomes a powerful weapon in an escalating technological ground war with Amazon.com Inc., as the two companies battle over shoppers who are increasingly making all sorts of purchases online. Amazon reset the landscape with Prime, in which more than 100 million people globally get twoday delivery and other perks for $119 a year. Even though Walmart is bigger in sales overall, it is an underdog online, and it is fighting for a larger presence.
Walmart is using Jetblack’s army of human agents to train an artificial intelligence system that could someday power an automated personal-shopping service, preparing Walmart for a time when the search bar disappears and more shopping is done through voice-activated devices, said Jetblack CEO Jenny Fleiss.
“It’s the tech of the future, right? It’s not what everyone is doing today,” said Ms. Fleiss, who previously co-founded apparel rental company Rent The Runway. The CEO said it could be five to seven years before the system is mostly automated and less reliant on humans. “This is a long journey,” she said. “And I think we were aware of that going in.”
Walmart is competing with Amazon, which has $233 billion in annual sales, including web services. In addition to Prime, the online giant has same-day grocery delivery from Whole Foods stores in some cities, plans to open dozens of small physical grocery stores and has sold millions of Echo speakers that let shoppers skip stores and websites altogether, and shop for products or request music with their voice.
Walmart is the world’s biggest retailer by revenue, with $514 billion in annual sales, but ecommerce makes up only a small percentage. That’s out of sync with where retail is growing fastest. Across the U.S., online shopping accounted for 9.7% of total retail sales last year and grew 14.2% from the previous year, according to the Commerce Department.
Walmart bought India’s big- gest e-commerce site. It has been buying up small online retailers including men’s apparel company Bonobos and is testing autonomous cargo vans for home grocery delivery in places such as Surprise, Ariz.
The rivalry is clear: At an annual meeting for Walmart store managers last year, attendees watched a parody video that included a clip of Darth Vader’s ship superimposed with an Amazon-like logo chasing Princess Leia’s smaller vessel, according to people familiar with the event.
Jetblack is a small piece of Walmart’s online investments, but it is one of the biggest gambles Walmart is making to attract wealthy shoppers and burnish its tech credentials.
Walmart primarily views the company as a research hub on AI and voice shopping. Some pieces of the business “could very readily be applied to the broader ecosystem in time,” she said. Jetblack’s software is learning to make agents more efficient, already suggesting language to use for many text interactions, said Ms. Fleiss.
Jetblack’s goal is that over time, through these interactions, the computer algorithm will learn to respond to requests with humanlike nuance but machine efficiency.
For now, the majority of interactions with Jetblack members require a human agent to press send on a text message or research a product recommendation, according to current and former customer-service agents.
On-demand delivery has been a treacherous business. In the late 1990s, Kozmo.com Inc., a startup promising delivery of nearly anything in under an hour, raised more than $250 million in financing, including $60 million from Amazon, but folded in 2001 after losing too much money. WunWun Inc. let New Yorkers order anything for fast delivery through an app, sometimes free, but found buying inventory by sending couriers to local stores without realtime knowledge of what was in stock too inefficient to become profitable, said founder Lee Hnetinka. WunWun closed, selling assets to a competitor, in 2015. Food delivery is also proving to be an expensive gambit for restaurants and grocers.
Ahost of companies have tried or are trying to automate textbased concierge and shopping services by using human agents. Facebook Inc. last year shut down a personal assistant dubbed “M” within its messag- ing app that fulfilled tasks, such as making restaurant recommendations or purchasing a birthday gift for a friend, using AI software being trained with human interactions. The service shut down but was useful to power other AI projects at Facebook, a Facebook spokeswoman said.
“I know a full cemetery of companies that have tried to do that and failed,” said Alex Lebrun, former head of engineering for Facebook Artificial Intelligence Research who recently started his own Parisbased firm to build an AI system that integrates human agents but doesn’t totally replace them.
Today’s AI technology allows companies to train text or voice software to respond to narrow requests, for example, to ask what color flowers a shopper wants to order, but so far can’t handle the broad range of questions humans tend to ask of a concierge-like services, said Mr. Lebrun.
The businesses that have attempted speedy delivery or chatbot concierges in the past didn’t have a giant retailer like Walmart behind them, said Ms. Fleiss. Walmart has existing retail and supply chain infrastructure Jetblack can use, she said.
Marc Lore, head of Walmart’s U.S. e-commerce business, had in mind a Jetblack-like service before Walmart bought his ecommerce startup Jet.com in 2016. That website sells products that appeal to urban shoppers, including higher-end brands that won’t sell on Walmart.com.
Mr. Lore was fascinated by the idea of a premium service that allowed shoppers to order products for speedy delivery by speaking into the air, said people familiar with his thinking. “This is a Marc Lore passion project,” said one former Walmart executive.
Since at least 2017, Mr. Lore encouraged executives to build a Jetblack branded, voice-enabled device similar to an Amazon Echo or Google Home that members could use to order products, former employees said. One prototype looked similar to a cylinder-shaped Ama- zon Echo, but with more colored lights. Mr. Lore declined to comment.
Jetblack considered a device, among other voice ordering options, but decided text communication is more useful for now, said Ms. Fleiss. The technology isn’t ready, and “customers actually weren’t ready and didn’t find voice currently to be appealing,” she said.
“When I’m laying in bed at night and I’m thinking about something, rather than going to Amazon and searching, I just text,” said member Julia LeClair, co-founder of a high-end fashion e-commerce site and mother of a 1-year-old. She has asked for recommendations on which sippy cup to buy and for help planning her daughter’s recent birthday party. Jetblack recommended a theme, decorations and party favors, and then ordered the items for delivery.
“You can definitely tell that some of the responses are from an automated bot,” said the 33year-old, “but maybe it’s not.” Her texts after 11 p.m. have received the response, “Bots need beauty rest too. I will be quick to respond in the morning!”
Walmart said it uses automated texts at night when agents aren’t working.
One former Walmart executive said Jetblack is “the first thing that we’ve tried that will unwind you” consistently from Amazon Prime. “The early indication is that it has legs,” even if the point isn’t earning profits, the former executive said.
Jetblack members are spending an average of $300 a week for products because the ease of the service encourages more frequent purchases, Mr. Lore said on the sidelines of a company party in September. “We are ramping it up gradually,” he said.
The average shopper is buying more than 10 items a week, said Ms. Fleiss. Average spending a week is higher than last September, said a company spokesman, but he declined to say how much Jetblack members spend a week or how many of those products come from Walmart.
Customer-service agents, often recent college graduates drawn to the startup culture of Jetblack, need to become experts on wealthy New York City moms.
Agents have two weeks of training, in part to learn what products babies need as they move through different developmental phases so they can make better product recommendations.
Moms—the vast majority of members—sometimes text fast requests like “reorder cereal.” When a Jetblack member joins, an employee usually goes to the customer’s home to inventory the products the person uses, giving agents a database of frequent purchases. Software automatically suggests a product if it is a frequent purchase or was scanned in the customer’s home. The software gives agents “responses that match nearly every possible situation,” said a former employee.
It takes agents slightly longer to place an order for “item requests,” when the shopper knows exactly what they want but hasn’t ordered it before. Even more time-consuming are “recommendations,” openended requests such as “I need a new yoga mat” or “I need a birthday present for a 9-yearold.”
Agents consult a file that combines past purchases, products agents have researched and recommendations made by Jetblack merchandising workers, then suggest around three items for the customer to choose from, current and former employees said. Sometimes agents head to Google to do research, said one of these people.
Through the dialogue, the system is learning which follow up questions to ask, said Ms. Fleiss. For example, if a shopper asks for a new stroller, the system might learn to next ask “For how many children?” and “Do you need your child to nap in the stroller?” Members buy one of the recommended products 80% of the time, she said.
Jetblack also learned it still needs traditional technology. It created a companion app because some shoppers don’t like to update credit card information or review past orders over text.
Workers stock up on frequently purchased items by taking a daily van to a Jet.com warehouse and Walmart stores in New Jersey, since New York City has no physical Walmart stores, ordering products online for pickup.
Later this year Jetblack plans to use a new Walmart fulfillment center in the Bronx to collect some products faster, said a spokesman.
Current and former employees said they first try to buy products from Walmart or one of its units but will buy products from any stores Jetblack customers want. That includes Amazon, where employees use Jetblack’s Prime account.
“It’s the tech of the future, right? It’s not what everyone is doing today.” JENNY FLEISS JETBLACK CEO