Orbital Insight Opens Its Satellite Network to the Masses

(Bloomberg Businessweek) -- Google’s long-running quest has been “to organize the world’s information and make it universally accessible and useful.” This corporate mantra focuses, for the most part, on arranging and analyzing data produced by humans, be it websites, books, calendar appointments, or the location of businesses around a city. But what if instead of gathering the world’s information from the ground up, you could begin organizing all of that data from above by looking down at Planet Earth itself? This has been the mission of Orbital Insight.

Founded in 2013, Orbital pulls in images snapped by satellites and uses them to watch and analyze human activity. It can monitor the number of cars in Walmart parking lots across the U.S. to see how busy the back-to-school shopping season is, the number of new homes going up in Houston, the amount of oil in China’s storage tanks, or the production activity at Tesla’s auto factory. Traditional economic data also measure these types of things, but Orbital says its images are more accurate indicators of what’s happening on Earth. “What we are selling is truths about the world,” says James Crawford, its founder and chief executive officer.

To pull useful information out of thousands upon thousands of images, Orbital built a complex software system infused with artificial intelligence. It’s spent years holding the hands of hedge funds, government agencies, and other customers to teach them how the software works and how to customize analysis, acting almost like a consultant. On May 15 the company released Orbital Go, a product it’s billing as more of a self-service application that lets customers hunt for fresh insights on their own. It’s part of a mission to make the technology widely available to businesses, governments, and other organizations, allowing anyone to interrogate the planet.

Crawford, 56, is tall and thin and carries himself with a self-confidence that makes it seem he was predestined to bring this technology to market. His career began in the 1990s at Bell Labs, where he worked on artificial intelligence systems—back when it was anything but cool to do so. From there he moved between business software companies, did a tour at NASA adding AI to the Mars rovers, then ran the Google Books project.

As he bounced around Silicon Valley, Crawford began to catch wind of a revolution in the aerospace world. A couple of startups, SkyBox and Planet Labs Inc., were making tiny satellites designed to surround the Earth and photograph every square inch of it, every day. Crawford quickly recognized there would be an opportunity for a startup to buy access to these images, combine them with free government image sets, and do something useful with all of it.

To test whether its technology could work, Orbital set out to predict the corn yield of U.S. farms. The company relied on public data to determine the location of the farms and got pictures matching those locations, which it used to gauge the health of crops—some satellite images are detailed enough to show their condition. When the actual yields came in as forecast, Orbital knew its technology was working and moved to Wall Street. The hedge funds wanted counts of cars in retailer parking lots and a measure of how much oil sat in tanks around the world.

In the case of the cars, Orbital had to look at four years of data for about 250,000 parking lots to find useful correlations between the number of shoppers and sales. To do this, it trained a computer-vision algorithm to spot cars and tabulate their activity. “We kept iterating and then started to see patterns,” Crawford says. “You could see Black Friday, Christmas, back to school.”

As is the case with many modern AI projects, humans must train their computer-based overlords. Orbital, which has about 130 employees, has spent years recruiting and teaching some 50 contract workers to look at images, tag useful things in the pictures, and feed those observations into a computer. These people will, for example, look at pictures of Houston and label images of freshly poured home foundations, halfway-finished houses, and newly built ones. Eventually the computers learn to do this on their own. Where it used to take months to refine a data set, Orbital can now create one in a few weeks, which has enabled it to count more and more things.

In the past, customers looking to create a data set would work individually with Orbital on custom coding to get things to work. With Orbital Go, they can tap into the startup’s data feeds with a few clicks of a mouse. Someone can, for example, select what to analyze—vehicles, ships, residential housing, planes, land use, infrastructure changes, foot traffic—and then specify the period they want to examine. Next they can pick from an existing database of places—countries, states, cities, ports, even specific stores—or create their own area of interest by drawing on Google Maps. For example, a customer could analyze the auto and foot traffic at all of Brooklyn’s gas stations over a year to find out where to put a new station.

The foot traffic information is derived from anonymized location data Orbital pulls from services that track smartphone use and then pairs with satellite images to estimate how many people work in a factory and how much stuff each worker produces. Orbital’s quest is to build a mathematical model of how the global economy operates. “It often sounds like the sage people in the world actually understand it,” Crawford says. “But there is an awful lot of mystery.”

Jeff Meyers, CEO of Hanley Wood, has spent the last year working with Orbital to refine data on the U.S. housing market. His company sells information detailing the state of millions of construction projects and, historically, much of that data has been gathered slowly by people on the ground. The satellite images, by contrast, provide a weekly—and sometimes daily—look at the market. “It gives us clarity that has never been available before,” Meyers says. “If this system had been in place in 2008, you would have known there was an abundance of inventory very quickly.” He expects Go will coax more customers to give the technology a try.

To date, Orbital has raised more than $80 million from Sequoia Capital, Lux Capital, and other investors. (Bloomberg Beta, the venture capital arm of Bloomberg LP, is an investor in Orbital. Bloomberg also offers Orbital data via the Bloomberg Terminal and the Bloomberg Enterprise Access Point.) Crawford declines to provide a revenue figure but says sales are doubling every year. The company’s major issues have been making businesses aware that this type of data even exists and finding customers interested in more than a one-off job. “You may find valuable insights that people will pay for, but it’s challenging to repeat that success,” says Tim Farrar, a satellite and aerospace consultant at TMF Associates Inc. “I don’t think we know how big this market is yet.”

To contact the editor responsible for this story: Dimitra Kessenides at dkessenides1@bloomberg.net

©2019 Bloomberg L.P.