MioTalk

It’s a bird... It’s a plane... It’s satellites!

Tom Diamond 2019-04-03

In this report, we speak to Tom Diamond, President and Co-Founder of RS Metrics, on the rise of satellite imagery data, the promise of alpha from geospatial analytics, and the financial possibilities for business and investors with a bird’s eye view.

Please tell me about how RS Metrics began.

 

Before founding RS Metrics in 2010, I was working for a management consulting firm called Stax Inc. where we worked with the likes of Prudential, H&R Block, or American Express conducting quantitative surveys and analysis of customers and merchants (anywhere from Delta Airlines, Best Buy to convenience stores and gas stations) to support corporate strategy and growth. I also worked with private equity firms and hedge funds in their due diligence on markets or companies they were interested or had invested in.
 

It all started when a private equity firm who was looking to do an investment, wanted data on the company’s overseas operations. They had no way of conducting due diligence on how the location had been operating over the last couple of years. My brother, Alex, who at the time was working with a satellite company, DigitalGlobe, was able to acquire some images of the plant which proved it had been operating in the way it should. When the client saw the time-series of images, and saw the trucks and the employee cars, he immediately pulled in 20 people from the floor. Everyone was giddy, laughing and clapping because they could see, without a shadow of a doubt, what was happening with the offshore location. So I realized at that moment that there's something there.
 

Today, RS Metrics is a satellite imagery data and geospatial analytics company for businesses and investors. Leveraging advanced computer vision and machine learning, we extract insightful, ready-to-use data from a variety of location-based sources, providing predictive analytics, signals, alerts, and end-user applications for decision making in financial services, real estate, retail, industrials, metals, government, and academic research.

 

How was the satellite imagery landscape back when you first started?

 

When we started out, everything was almost 100% manual. Using satellite imagery at scale for geospatial analysis was still in its early stages back in 2010 which meant we had to figure out with the satellite companies how to deliver batch orders to us. Airbus, DigitalGlobe, both had never done this before. Google Earth and RS Metrics were among the biggest consumers of satellite imagery from DigitalGlobe. It's not like today where a business can just go to DigitalGlobe and sign up for their GBDX platform and just start pulling batches of imagery. We had to build it.
 

Back when I was with Stax, I had already been working with large retailers like H&R Block. So I had a good handle on retail and knew that hedge funds were interested in retail and there was sufficient satellite imagery available on retail. So we decided to start with retail, to cover and test namely Walmart, Home Depot, Lowe's and McDonald's. We pulled thousands of images of stores and we used very old-fashioned counting methods to measure and do our initial testing. What we saw was that there was a correlation; that the number of cars in the parking lots actually does relate to revenues for the entire chain.

 

At the time there were fewer satellites in operation, so our samples sizes for larger retail chains were relatively low but still good enough for quantitative analysis.  Now there are many more satellites operating which is making sample sizes higher and results more granular and accurate.
 

What's the most challenging moment in your entrepreneurial journey that has forced you to rethink the business?

 

We faced a shift in the business when the number of satellites that we had anticipated to be launched over the years in ‘15/’16, never materialized. While we had large quant firms and big clients like Lowe’s, Target, Kroger, we faced fierce competition when cell phone and GPS data started to come out and attracted attention away from satellite data. There are certainly more satellites now and much more data available for measuring retail traffic trends, and privacy laws are restricting the access of personal data from cell phones/GPS so things are more realistic and satellite data is again in the spotlight for retail.  But back then, it was the beginning of people realizing how granular the data could be from cell phones. Even though it only captures a small percentage of total customers at a location, it’s collected every minute. While our data was collected only every couple of days over each location.

So we really needed more satellite images to increase the sample sizes and make the retail product stronger and more powerful. Ideally, we needed to measure every store, every day. We knew we had to make a pivot in order to survive. We needed to have other products that didn't require that level of frequency.



Source: RS Metrics
 

So we decided to focus more on our products for commodities.  Specifically, MetalSignals which is a global survey of outdoor metals storage at the smelters where metal is made, where they meltdown concentrates, And storage sites, like ports and terminals where they store and ship base metals, steel, coal and iron ore. And of course, the mines themselves.

We are still doing our retail data. But we're going to relaunch it in a bigger way at the end of this year. With the availability of more satellites, we now are able to provide global coverage of retail and commercial real estate, CRE in the frequency and accuracy that clients need.
 

Please tell me more about your MetalSignals product.

 

MetalSignals is a platform, app and data feed that provides satellite-imagery based data, trends, and analysis for global aluminum, copper, zinc, and steel production, based on changes in daily, weekly, and monthly growth in metal concentrate and finished product stored outside at hundreds of individual smelters and storage facilities across major countries such as China, Chile, Russia, USA, Australia, and more.
 


Source: RS Metrics
 

MetalSignals not only forecasts futures price and inventory for the metals themselves, but they also are very predictive of price for 100s of equities, currencies, and interest rates. Copper, also known as the “PHD of Metals” or “Dr. Copper”, for example, is known to be very related to the economy. So our copper signals, generated from our data from measuring copper at smelters and storage globally, forecast price for the Euro, the Great British pound, the Australian dollar and the Japanese Yen, interest rates like the 5 and 10-year U.S treasuries and hundreds of equities and indices like Glencore, Alcoa, COPX, very well.
 

Source: RS Metrics


A lot of MetalSignals customers are hedge finds and macro/FX firms, and companies and traders in the metals industry.  But also some companies you wouldn't even think would be related to metals like Consumer Goods companies are working with us for hedging aluminum for their aluminum cans and other product packaging. They use this because they can see how much aluminum is being produced in the world and forecast the price and figure out where it all is. At RS Metrics, while we still serve a lot of financial institutions, we have a broad range of corporate customers as well.  We also have excellent distribution partnerships with Bloomberg Enterprise Data, CME Datamine, and Quandl/Nasdaq that provide global access to our products.
 

How accurate are your metal signals?

 

We cover about 500 global locations, and growing and we measure the amount of finished metal products, concentrates, and vehicles that's sitting at these locations… like the Al Taweelah smelter in Abu Dhabi or Escondida in Chile. To ensure that we get 99% accuracy, we have a human quality control workflow that corrects errors from our AI/machine learning. We take the pre-measured images, and check and correct them, which then provides training and feedback for the system.
 


Source: RS Metrics

 

That data we extract from imagery, when aggregated into an index is highly predictive of inventory and futures price on the exchanges like the LME, CME and SHFE. It's about 70% to 80% directionally accurate forecasting changes in price and inventory, one, two and three months out.


It’s kind of like an iceberg. The top of the iceberg above the water and where you can see, is just a small part of it and that's what's on the exchanges like the LME, CMEs, Shanghai exchange. Those are the contracts that are called on warrant. Everything else below the water, no one knows about. And what that is, is the aluminum sitting at a smelter in Russia that no one has ever reported about. But we have it and we see it. We can see what’s sitting at China’s copper smelters and storage facilities three months before it sits on the exchange. It still is totally shocking to people that it works that well.
 

How has AI enabled your business? Are there limitations?

 

AI/machine learning is powerful but still immature, and really is a means to an end for us. It surely helps us do the bulk of the work for us, but it still often needs human interaction to get to 99% accuracy.
 

Over the years RS Metrics has combined computer vision and machine learning with rigorous sampling techniques to train the underlying machine-learning models that power MetalSignals and generate high quality data at scale. The platform is proprietary and patented, and includes a scaled QC workflow to generate high quality, predictive and consumable information.

 

"If someone tells you they can do 100% accuracy with AI for satellite imagery, they're going to show you a picture of a Walmart on a sunny day, with glistening sunlight coming off of every car."

 

But for the most part, especially at smelters, it's foggy, with trees, puddles and snow on the ground. When we're looking at China, where there's smog in some cities and smelters are dirty places to begin with, you really do have to have that human component involved in the workflow.

As more satellite imagery data is available, it's going to be like a fire hose, spraying at us and we need to have the computers to help handle that scale, and at the same time, it has to still be accurate.

So it's a challenge for everyone in the industry. You won't hear many people talk about it because almost everybody, like I said that's in this space, comes out of the pixel world and their entire reputation is based on that side of things. But the truth is there's a long way to go. But the future is very bright.

 

Are financial institutions you engage quick to accept this form of alternative data?

 

A picture tells a thousand words and that's a true statement. Geospatial analytics is fact based, which people like about it. It takes the risk and “opinion” out of the situation. There's a lot of cases where simply looking at images answers the question, but in most cases all customers need to do is look at the data and the signals that come out of the images. Its clean and easy to understand and use; we're not saying, “Here's raw data. Now, good luck!”   The data is available in easy-to-use end-user apps, alerts, signals, and tools, and as daily data feeds for customers with quantitative focus and capabilities.
 

Source: RS Metrics


Take MetalSignals for example. Everything our client needs to run our demonstration forecast models is in one zip file. We have white papers and make the entire code available; where we take the inputs from the markets and our satellite imagery inputs. We then run a KNN, a nearest neighbour’s algorithm or forecast model which then outputs directional forecasts for one month ahead to three months ahead for all the different metal types. And RS Metrics does this with a 70-90% accuracy by combining the best in AI with human supervision. We’re the only company that’s been able to this even today, at scale.
 

We've gone that far because even companies that have data-scientists, they're busy. They don't have a lot of time and resources to test and interpret the data. That's why we don't just make the data, we deliver the Alpha and insights right to their door.

 

RS metrics has always been run by people who’ve come out of business and investing as our pedigree. Anyone else in this space that you talk to for satellite imagery or geospatial data, almost all of them, their pedigree is on image processing, is on pixels. So for example, how to detect an airplane using AI. For us, in our experience that is maybe 10% of the problem. No one can do anything with that information. You need to put it into context or an index that can be compared against itself, that can generate insights or predictive signals.

 

That's how people use it in the industry. There has to be a baseline and change off of that baseline that they can react to. So there's a lot of steps in between producing raw data from satellite imagery, making it accurate and then taking it all the way to models that are proven to generate Alpha.

 

Any advice you would give incumbents who are looking to adopt alternative data?

 

I would recommend reading a book by Geoffrey A. Moore called “Crossing the Chasm” which expounds on the gap between the early adopters (visionaries) of new technology and the early majority (the pragmatists) and suggests methods to successfully cross the "chasm".

 

From what I observe, the biggest investors and corporates in the U.S are still slow on adoption. But there have been significant movements into the mainstream recently with Bloomberg, CME, NASDAQ, Quandl, Factset, IHS. All of these firms are building alternative data hubs now. As satellite imagery advances or technology in general advances and barriers to entry are lowered, the inevitability of tech adoption goes higher. So when we talk to big institutions about what we do, everyone understands that this is the future. It's going to come and if they don't use it, somebody else will.

 

"But one important thing I’ve learned working with customers who are some of the smartest people in the world over the past almost 10 years, is if you're going to use and test alternative data, whether it's satellite imagery or credit card data or point of sale data, or social sentiment data, you have to get out of your existing box."

 

One example I would give is with our retail data. Sometimes someone would call us and say, “I have a model that tries to forecast the exact revenue number down to the dollar for 50 retail chains. Can I use your data to make my model more accurate?”. And the answer is no. Maybe there's a way. We haven't tried it. But that's not the way to use it.

 

The way to use the retail data is to look at signals that are generated based on change in traffic at these locations. And when a signal is generated, it’s predictive that there's likely to be a revenue beat or a miss. And you can do a lot more than just that with those signals, but essentially that's how you use it. It's not necessarily going to make your current model 1% more accurate. Mostly, it's a separate stream of information. It's much like a mosaic of research. It’s an input to compare with everything else that you're seeing.
 

So that's one piece of advice for anyone that's looking to use alternative data, look at it as a separate stream that feeds into your overall research mosaic. It's not necessarily going to make your existing number more accurate, but it's going to make your overall process a lot less risky and give you more ideas.

 

With barriers to entry decreasing, how does RS Metrics remain competitive in the market?

 

No industry is safe from competition or evolution or any advancement. You’ve got to keep going. But for us, we have had the advantage of a head start. It’s already hard enough as it is to monitor hundreds of thousands of global locations on a daily basis going forward. If say this store moved across the street, or if this store shut down, or this smelter started up or this company bought this storage yard; all this is hard enough to keep up with moving forward on a day-to-day basis. Trying to reverse-engineer our products, to go back in time to figure out what has happened in the world (which you have to in order to provide a baseline), it's almost impossible to do.

Anybody starting today, is going to have years of work before they can get anywhere close to what we're doing.

On top of that, our differentiation is really that we provide the whole gamut of geospatial analysis for clients, everything from pulling and managing imagery, counting and measuring data, to models and forecasts in generating Alpha. We even go as far as developing not just research but also financial products like ETFs. With our retail and metals data, we're already beginning to work with index providers. We want to provide the indices, the “Intel Inside” for ETFs that track brick-and-mortar retail or metals and other spin-offs.

So in the future, we're going to be scaling further up the value chain, providing more value and ease of use to our customers. Competition wise, alternative data and satellite imagery in general, has many pockets for opportunity. There's retail, agriculture, energy, commodities and I believe it’s going be a long time before it gets too competitive and too commoditized.
 

 

Share this to
Share this to