Bean Counters in Space: How Orbital Insights Sees the World

Orbital Insights is applying machine learning to a vast amount of satellite imagery to help hedge funds and governments quantify the world.

Thanks to Moore’s law, in which processing power becomes exponentially cheaper and more efficient over time, the iPhone in your pocket is “one million times more powerful than an IBM computer from 1975…which took up an entire room (1).” The meteoric rise of the smartphone has created a supply of powerful, cheap, and standardized components that in turn are allowing for rapid development, experimentation, and application of technologies that were once prohibitively time and capital-intensive.

Palo Alto-based Orbital Insights has capitalized on the dramatic resulting advances in space technology and computing. On one side, satellites are smaller, cheaper to produce, and easier to launch than they’ve ever been, generating enormous amounts of data. On the other, artificial intelligence and cloud computing allow for automated analysis of unstructured data at massive scale. By connecting the two sides of this equation, Orbital Insights is helping its clients see the world in unprecedented ways.


Planet Labs' Dove nanosat
Planet Labs’ Dove nanosat (1)

Historically, the space industry was characterized by decades long, billion dollar projects, making it close to impossible for commercial players to get involved. As recently as 2010, “the dominant imaging satellite was the size of a school bus and cost upwards of $700 million to build and launch.”(2) The development of modular, miniaturized space components has allowed companies such as Skybox (acquired by Google for $500 million in 2014 (3)) to develop much smaller (think “mini-fridge” or “wine bottle”), low-cost satellites called nanosats, without significant quality sacrifices.


Meanwhile, launches have become cheaper and more frequent. SpaceX first disrupted the industry with its relatively low-cost ($62 million) commercial rockets. Though these rockets are designed for large payloads, nanosats can “hitch a ride” for a lower fee (4). Newer entrants like Virgin Galactic’s LauncherOne project and Rocket Lab’s $4.9 million Electron rocket are specifically targeting smaller satellites for low-cost, frequent launches (5). With the legacy school bus-sized imaging satellites, images like those you see on Google Earth cost in the thousands and are updated monthly (6). A network of small satellites, however, can produce daily images of any point on the globe at “a fraction of the cost (7).”

Until recently, processing this output would require warehouses of analysts manually poring over individual images. This is where Orbital Insights comes in. Founded in 2013 by James Crawford, an artificial intelligence expert with experience at NASA and Google, the company has developed a scalable software solution for space image analysis. Orbital Insights uses “deep learning” neural networks and machine vision to algorithmically identify and quantify objects and patterns in individual images, and then big data techniques to aggregate those patterns and derive macro insights from micro inputs for its clients.

Machine vision identifies cars in a parking lot
Machine vision identifies cars in a parking lot (3)
Software measures oil volume in storage tanks
Software measures oil volume in storage tanks (2)

What does this actually mean? By counting the number of cars in the 50,000 parking lots of 90 major U.S. retailers over time, Orbital Insights can predict retail sales trends before they are reported to financial markets. By analyzing shadows inside 20,000 oil storage tanks globally (the tanks have floating lids whose height depends on the volume of oil inside), Orbital makes regional and global oil inventory trend estimates, including for regions where official data is non-existent or suspect. By tracking building shadows (which are proportional to building height), car and truck density, road building, and electricity consumption (indicated by nighttime illumination), Orbital measures economic development in China, a country known for its lack of transparency when it comes to economic data (7).

Orbital Insight’s work is of obvious interest to investors hoping to find an informational edge in the public markets, and asset managers make up most of their client base, but the company also works with governments and NGOs. Last year, Orbital partnered with the World Bank, focusing on poverty data (8).

The company’s operating model is focused only on the software side, thereby avoiding significant investment in hardware (8). Orbital doesn’t have its own satellites, but by partnering with a broad range of 3rd party imagery providers like Digital Globe, Airbus, and Planet Labs (9), it claims to achieve “a level of data and picture resolution that no single satellite provider can compete with (8).”

Potential Skybox (now Terra Bella) fleet
Potential Skybox (now Terra Bella) fleet (4)

However, I believe the company and its VCs are underappreciating the vulnerability of their business model. In early 2016, Google renamed previously acquired satellite company Skybox Imaging to Terra Bella and announced a new focus on analyzing their images through software (10). Google is no stranger to machine learning, and its announcement is a credible threat. Moreover, while Orbital Insight’s edge of working with multiple imagery providers holds weight in the legacy world of large satellites with limited coverage, the nanosat future would enable any given satellite operator to have sufficient geospatial coverage to disintermediate Orbital Insights. To preserve its edge, Orbital Insights should seek to expand its imagery inputs from satellites to aircraft, drones, and anything else it can get its hands on.


[Word Count: 790]


Works Cited

(1) Roger Cheng, “Moore’s Law is the Reason Your iPhone is So Thin and Cheap,” CNET, April 16, 2015,, accessed November 15, 2016.

(2) Murray Newlands, “To Infinity and Beyond with Investments: Rocketing into Space Investing,” Forbes, August 27, 2015,, accessed November 15, 2016.

(3) Christopher Mims, “Amid Stratospheric Valuations, Google Unearths a Deal with Skybox,” Wall Street Journal, June 15, 2014,, accessed November 15, 2016.

(4) Samantha Masunaga, “Small Satellites are Back, with Down-To-Earth Expectations,” Los Angeles Times, May 27, 2016,, accessed November 15, 2016.

(5) Bernie Lo, Nishka Chandran, “Rocket Lab Nears Completion of World’s First Private Orbital Launch Site in New Zealand,” CNBC, August 28, 2016,, accessed November 15, 2016.

(6) Klint Finley, “How AI Can Calculate Our Oil Surplus…From Space,” Wired, March 16, 2015,, accessed November 15, 2016.

(7) Orbital Insights, “Solutions,”, accessed November 15, 2016.

(8) Connie Loizos, “Orbital Insight Lands $20 Million from Investors, Led by GV,” Techcrunch, June 27, 2016,, accessed November 15, 2016.

(9) Joshua New, “5 Q’s for James Crawford, Chief Executive Officer at Orbital Insight,” Center for Data Innovation, April 4, 2016,, accessed November 15, 2016.

(1) Frederic Lardinois, “Google Renames its Satellite Startup, Skybox Imaging, to Terra Bella and Adds Focus on Image Analysis,” Techcrunch, March 8, 2016,, accessed November 15, 2016.


Featured Image:






PAYTM and India’s Cash Driven Economy: Opportunity for Online Payments


CHOPCHOP – Smart Cooking Backed by Artificial Intelligence

Student comments on Bean Counters in Space: How Orbital Insights Sees the World

  1. Dan, this is really incredible stuff that I did not know was happening – the use of satellite imagery as a means to market predictions is a pretty innovative way of thinking (and executing). I’m curious if there is any indication that Orbital Insight is working to mold their future business model to account for the potential threats from Google and others, or if they are truly, as you surmise, overconfident in their unassailability. Regardless, I can see this space (pun intended) becoming increasingly competitive and highly-contested as the capabilities become more accurate and reliable, and the stakes get even higher. Has there been any pushback from the investment community or increased efforts to partner with Orbital Insights, specifically from those who do not have access to this data and may feel at a disadvantage?

  2. That’s really impressive work. It’s amazing to see how far technology has come. However, I’m in agreement with you on their precarious market condition. The barriers for entry into this market are dropping fast. And “anyone” can do machine learning. Without specific IP or physical assets, what is their edge?

  3. While Orbital Insights is generating in-demand data, what are the long-term barriers to entry for this business? Over time, can’t anyone send up a satellite, or leverage someone else’s satellite and analyze the pictures? I would be concerned about investing in this business, as I don’t see sustainable grooves of differentiation — what’s their edge (as Prof. Greenwood would say). Separately, it seems they are gathering valuable information before others have it. I imagine this data is only valuable to the purchaser if it is sold exclusively to them (if everyone has the data, what’s the point?) As a result, I wonder if other stakeholders of this proprietary info, such as the government, will intervene and demand that the info become more widely available. If the insights have deep implications for society, the government is unlikely to be pleased with the info being concentrated in the hands of a few profit-seeking hedge fund managers.

  4. Nice post! I think the increasing ease and decreasing cost with which we can put satellites into space raises loads of interesting concerns, particularly when it comes to intelligence gathering, privacy and espionage (including personal, corporate, government). As I see it, space is currently a free-for-all, no one owns it, no one regulates it – who possibly could right? But once upon a time the same thing was probably said about oceans – I wonder if, as the value of satellite ownership becomes more apparent to many, “ownership” of space will become an issue for the international community to grapple with?

Leave a comment