IN THE CORNER of a packed conference room in Box headquarters in Redwood City, Aaron Levie is coiled on top of mini fridge. A little taller than the conference table in front of him, at which Box’s key marketing leaders are assembled, the perch allows Levie room to fidget – and pepper the team with real-time feedback.
It’s a sunny September day in the Bay Area, just a few days before Box’s annual conference, BoxWorks, is set to begin in nearby San Francisco. Levie and his enterprise software company will be unveiling something big — Box’s push into artificial intelligence – and it’s important to get the details right.
“We need examples that will be practical and realistic,” Levie tells the team as Ben Kus, the product management leader running the AI group, runs through examples of Box’s new tech at work on a projected screen. “And we need something with documents on Day 1.”
One quarter away from three years as a public company, Box and Levie have changed, but plenty remains the same. Levie still works late into the night, napping at the office when everyone else goes home, his signature sneakers suggesting a youthfulness only partly kept in check by his unkempt salt-and-pepper hair. And Box, which has added a host of products and updates in the months since from security to collaboration, is still famous for its documents—the files that customers upload and share within Box.
At a market capitalization of $2.5 billion and with a stock that’s rebounded in recent months, Box is neither frothy Silicon Valley unicorn nor outlier public success – a fact not lost on Levie himself. By betting on artificial intelligence, Box is joining a crowded field of tech companies to hitch their names to AI. But Box has something they don’t, Levie says, in the 30 billion files customers store on Box’s software today.
With annual revenue of about $400 million today, Box management is focused on a path to $1 billion. To get there – and join the ranks of software’s largest companies – Levie’s turned to AI.
IN THE SPRING OF 2016, an engineer named Reshma Khilnani got an urgent email from the boss. Google had just opened up its machine learning tools for image recognition, called Cloud Vision, for any developer to use. At Box, Levie had dutifully watched announcements from Google, Apple with its Siri technology, Microsoft with Cortana and many more. He’d partnered with IBM in 2015 to integrate some APIs from its own AI platform, Watson. But with the Vision API, the dam burst.
“That was probably a bit of an ‘Aha Moment’ collectively,” Levie says. “It was pretty clear that AI was going to be this remarkable trend in technology.” But now, to hear Levie tell it, the tech was clearly mature.
Box wouldn’t be starting from scratch – the company had its core base of 76,000 paying customers who combined for 30 billion files stored with Box, some with tens of millions at a time. (At the upper end, GE Digital has more than one billion.) The hardest problem for artificial intelligence, Levie believed, was training the data. Few companies had more data to work with than Box.
As an email thread heated up and a “tiger team” of volunteers formed to work on the project over the summer of 2016, Levie had to decide just how much Box would build on its own. Box had acquired a startup called dLoop in 2013 to beef up the analytics administrators at big companies could have about how their employees were using Box, and cofounder Divya Jain had worked on ways for Box to better classify and filter up content to its users (Jain left in 2016). Under Kus, that work, along with new efforts, developed into Box Graph, a project to embed AI tools within the typical workflow of using Box.
Graph would add small improvements such as predicting what documents you might want to see based on which coworkers you were talking to, or pulling up relevant files based on the people on your calendar for meetings that day. But for more ambitious uses in image recognition and text analysis, Box had to decide how much it would outsource. Box bet that competition among the leading players like Amazon, Apple, Google and Microsoft would make such tools a commodity – and one Box couldn’t afford to try to build better. “It instantaneously became unattractive,” says Levie. “So we wanted a general fabric to integrate any of their services.”
The result, unveiled on Wednesday publicly for the first time, is Box Skills. Skills functions as a clearinghouse for other tech company’s AI tools, allowing any Box customer to take advantage of Google for images and Microsoft for text without committing long-term to either.
Box has publicly touted partnerships with many of the leading companies in the AI race since going public: Amazon, Apple and IBM, Google, Microsoft and more. With Box Skills, the company plans not to play favorites, even as it recommends particular use cases to customers. “There’s a little bit of tightrope walking,” Levie admits. “But most of this industry is understanding that we’re in a world where you compete on product value, not exclusivity.” In a fast-moving market like AI, in which Alexa and Cortana can partner up even as they compete, Box is betting that there’s room for a software Switzerland.
THE DIRTY SECRET of file sharing and storage services like Box’s original business and competitors like Dropbox is that much of the data is uploaded and never seen again. Levie estimates that only 5% to 10% of Box data is “hot,” or ready to be pulled up quickly because a customer is likely to use it, at any given time. Box Skills hopes to change that by tapping the “massive long tail” of images and text gathering digital dust. Levie imagines a marketing agency that’s worked with a client for years. When a new staffer joins the account, the staffer could create and access a folder with all past work, images and video alike, for that client in just a couple clicks. A lawyer working on a merger could use different AI tools such as optical character recognition (OCR) to group together all past sensitive files referencing the corporate target.
Kus, the product leader behind Skills, believes his team is building the first centralized system for unstructured data – a term technologists use to differentiate between files like documents and images compared to easily sortable numbers and code. “People flip through images all day long on the job,” he says. “If you have 100,000 images, that gets too hard.” Box has its own APIs, he notes: a worker could start their day in Office or Slack, and pull up Box information. “This does give you a reason to go back into Box,” he says.
Back in the go-to market meeting in Redwood City, Kus shows the marketers and Levie the demo use cases his team has put together to show off Skills. There’s a demo for extracting data from drivers licenses for ID verification; another for extracting info from a large set of forms. The team also shows off a non-public set of images from one customer to show how recognition software could be easily customized within Skills to tag photos that don’t show up in Google search results. Customization is where advanced Box users will see the most value, the team agrees – but first Box needs to have individuals on the ground to walk companies through its most basic AI features. Kus requests that every sales rep pitching Skills make one herself first. Above all, the executives remind each other of one thing: don’t assume the typical Box customer is an AI enthusiast – yet.
A typical Box customer resembles Sunbelt Rentals, the second largest provider of rented industrial construction equipment in the U.S. While Sunbelt’s business depends on heavy-duty machinery like backhoes, its sales people win because of their ability to close a deal from a mobile device, says IT chief Dean Moore – almost 10,000 iOS devices in all. For its images of construction sites and the equipment it leases, Sunbelt uses Box to the tune of about 3,200 users today. The company’s currently testing using Google Vision’s API to tag photos taken of equipment in the field at job sites to track its whereabouts and usage history.
“There are a ton of potential uses we are looking at,” says Bill Moertel, director of Sunbelt’s ecommerce sales. The executives imagine scenarios in which sales people could photograph a new construction site to predict what type of equipment its contractors would need, even homeowners conducting renovations. Sunbelt could also track the makes and models of its equipment in the field without manually inputting that data, and track lost equipment from natural disasters or theft. Sunbelt was unlikely to use machine-learning tools from companies like Google and Microsoft directly, says Moore. “We’ve made Box the center of where we’d house this information. That made Box a fairly easy choice.”
At The Coca-Cola Company, about 20,000 employees use Box and have for the past three-plus years, with Box connected to its Salesforce CRM platform, video conferencing and Office suite. Like Sunbelt, Coca-Cola has many terabytes of data stored with Box, including sensitive documents it’s trusted with Box’s security tools. For Coca-Cola, Box Graphs can help employees on large projects work better across teams, says Barry Simpson, senior vice president and chief information officer. Box is also critical for sharing assets with internal and external marketing teams and with franchise partners, Simpson says.
As a large corporation with its own machine learning team, Coca-Cola has already worked on its own AI tools, similar to another large customer that’s tested the tools, GE Digital. Groups at both companies can now focus on more differentiated offerings while the company uses Box and its partners to tag product images and sort documents. “We spend more time gluing all this stuff together versus actually leveraging the technology investments the other players are putting in,” says Simpson. “Having Box is a major accelerant.”
AT BOX’S WEEKLY BRAINSTORM session among top executives, Levie – in his corner spot as usual – grills his technical and product leaders on a deeper dive into the security capabilities made possible by machine learning. These improvements aren’t announced at BoxWorks – we’re talking six, even twelve months in the future. Box is bullish that the same tech in its Box Graphs can lead to improvements in security for its customers. Graphs can easily map companies by who’s working with whom. Today that helps with recommending content to groups or catching someone you might’ve missed. In the future, it could help flag irregular activity like a “Snowden event” when an individual employee goes rogue.
Box makes much of its security tools today, such as its encrypted keys, and security advantages are even more important in growing Box’s business internationally, a priority for new COO Stephanie Carullo. “We have an opportunity to go a little wider,” she says. “International expansion is a really important piece for us.”
But Box will have to successfully navigate its partners while making sure the team doesn’t move too fast – offer AI capabilities that don’t work, and you quickly lose credibility. “AI is a nascent space, and many of the technologies have yet to be developed,” says Jia Li, head of Google’s Cloud Machine Learning group. Google’s team benefits from partnering with Box by getting exposure to industries newer to such technology such as agriculture, education and healthcare, says Li. Like Box, Google isn’t playing favorites or working with the company on an exclusive basis.
Box product chief Jeetu Patel argues that the company’s AI push will prove a money maker for its partners, minimizing concern that it’s building a short term middle layer that will see its margins compressed and ultimately bypassed. “Watson will make money from this,” he says, as the AI tool providers make money from the scale brought to them by Box’s customer base. “A piece of content in Box should be infinitely more valuable to you than if it wasn’t in Box.”
The greatest challenge facing Box in its AI push could prove a variation of the “garbage in, garbage out” problem. In that challenge, algorithms trained on low quality data sets learn bad insights – they spot patterns or make inferences that are incorrect, limited by the quality of the information available. Box will face a similar challenge by outsourcing the machine learning and AI tools in its platform, says Gartner analyst Darin Stewart. Quality and capability could vary widely working with third parties, he says, adding that Box will need to carefully monitor what it chooses to expose to customers. Box is also not alone in pushing into AI among content management players, says Forester Research analyst Cheryl McKinnon. “AI and machine learning will be the big area of disruption in 2018 – so the ability to move quickly and deliver real value to customers will be key.”
WHEN HE NEEDS ADVICE, Levie turns to John Chambers, the former longtime Cisco CEO and outgoing executive chairman legendary for his track record and influence with several generations of Silicon Valley’s best and brightest. Chambers has informally coached Levie for the past five years and calls Levie one of the five most impressive young leaders he’s met in the past decade. To Chambers, Box’s move into AI is one as significant—and as brave—as Levie’s commitment to focus on enterprise customers over individual consumers nearly ten years ago. “The move to enterprise, and then high-end enterprise, were bigger bets, and he made those well,” says Chambers. “Now he’s having the courage to change again.”
Box won’t follow every wave of popularity – executives joke not to expect a Box flying car – but Chambers’ argument that Box has weathered the worst from doubters in the past is held up by the company’ stock. In February 2016, just weeks before Levie’s AI team got to work, Box traded at just $9.90 per share, down 57% from its first-day high. Today Box has clawed its way back to over $19. The company has guided to reach its $1 billion revenue target by fiscal 2021.
Levie anticipates that Box’s move into AI might induce eye rolls among those worn out by the term’s overuse. “We’ve all probably been collectively overhyping, at the highest level, what AI is going to do,” he says. But just as every company needed a mobile strategy with the rise of the smartphone, Box’s longtime CEO believes the same will prove true with AI.
“There’s going to be a very wide spectrum of companies who use this, and for some it will be transformative,” Levie says. “Between nominal and revolutionary, we hope to be more on the revolutionary side.”
More Info: www.forbes.com