What is this cult and how do I join? I've been screaming this from the rooftops since 2015. Everything in this post applies not only to satellite imagery but also to aerial imagery providers as well. The biggest reason why satellite imagery providers haven't grown at breakneck speeds, is still the imagery providers themselves.
I was chatting with a product manager once who worked at a prominent oil-and-gas software company. They have all the know-how, every connection high and low at pretty much every oil company in the world. One of their products is an oil field intelligence product. So I said "did you ever talk to <Company Name You Can Probably Guess>? He said "Oh yeah we talked to them. *eye brows raised* They're expensive!" And that's the problem. Satellite companies can't penetrate the market because they can't cut prices. The reason they can't cut prices is that they can't go lower their GSA contracts, less they have to lower their federal prices which would have a huge negative impact on revenue all the way around. But the federal prices are high for many commercial enterprises to kick the tires on. You've got to justify the spend, and particularly when interest rates are going up, a lot of marginal ROI projects would get cut. So they plod along, hoping to stumble on the magic algorithm that will suddenly make people see the value. They should go out, hire some industry experts that cross over to guide product development for commercial entities for last mile delivery. This frees them from GSA or state-level MSA agreements by building solutions tailored to specific uses cases.
Thank you Joe! Very interesting! This discussion can also be applied to the emerging “Commercial RF” capabilities industry. I look forward to citing this article in some coming lively debates.
Dear Joe, as an old-timer in the field, I do sympathise with your views. A few additional points. EO business follows the "anti-Steve Jobs model". Jobs said: “a lot of times, people don’t know what they want until you show it to them.” In EO market, the good clients know what they want much better than business developers". Why? Because EO data only provides indirect information. When looking at images, nothing beats the experience of having been in the field. Only those with dirty boots really understand what informations needs to be provided. There are too many developers who have never been to the trenches of real life. All machine learning algorithms have biases, but those developed without extensive support from clients are usually hopeless. I have been a leader of the longest running operational service that uses EO data (deforestation mapping in Brazil which started in 1988) I can assure you that white-collar developers in jeans and tennis do not stand a chance of succeeding by themselves. Best, Gilberto Camara (www.gilbertocamara.org).
Really fascinating take on the industry. I agree 100% that a company should pick one thing to do well and focus on that. I do think there is a market for derived data though, as long as it is being derived with a specific audience in mind. One-size-fits-all analytics will never work. I used to work at a place where we could get approval to buy derived analytics but never the people-power to derive it ourselves. We didn't really have the option to homebrew something better and I know our peers/competitors were in the same position. That being said I also saw plenty of half-baked derived data feeds because it was more cost effective for them to get their offering to 80% than 100% (or even 95%). I kept sampling those but they were never good enough for our needs.
The problem is more prominent in classification algorithms. We are mostly trying to predict "a class" which we do not have a consensus on. I was trying to classify crop type of agricultural fields. When I talk to agriculture experts, they all explain a different "corn field" which is highly dependent on the farmer. Are we talking about corn fields which the seeds are planted in 30cm or 50cm? Are they going to be harvested for animals or industry? Any definition by human itself is biased. We are trying to fit a biased reality in our mind using biased models.
Great points. Couldn't agree more. Definitely see that as data, standards, tools, and models become more ubiquitous, large enterprises (like the one I work in) will increasingly be developing and deploying these solutions internally. As you point out the true value of commercial earth observation will come from building applications; for many applications in big orgs these will be built and deployed internally where these lean on integrating multiple proprietary data streams (e.g. sensor networks, operational data etc...). It will be very interesting to see how the EO vendor space responds to these developments. Exciting and fascinating times!!!
What is this cult and how do I join? I've been screaming this from the rooftops since 2015. Everything in this post applies not only to satellite imagery but also to aerial imagery providers as well. The biggest reason why satellite imagery providers haven't grown at breakneck speeds, is still the imagery providers themselves.
You're fired.
I was chatting with a product manager once who worked at a prominent oil-and-gas software company. They have all the know-how, every connection high and low at pretty much every oil company in the world. One of their products is an oil field intelligence product. So I said "did you ever talk to <Company Name You Can Probably Guess>? He said "Oh yeah we talked to them. *eye brows raised* They're expensive!" And that's the problem. Satellite companies can't penetrate the market because they can't cut prices. The reason they can't cut prices is that they can't go lower their GSA contracts, less they have to lower their federal prices which would have a huge negative impact on revenue all the way around. But the federal prices are high for many commercial enterprises to kick the tires on. You've got to justify the spend, and particularly when interest rates are going up, a lot of marginal ROI projects would get cut. So they plod along, hoping to stumble on the magic algorithm that will suddenly make people see the value. They should go out, hire some industry experts that cross over to guide product development for commercial entities for last mile delivery. This frees them from GSA or state-level MSA agreements by building solutions tailored to specific uses cases.
Thank you Joe! Very interesting! This discussion can also be applied to the emerging “Commercial RF” capabilities industry. I look forward to citing this article in some coming lively debates.
Dear Joe, as an old-timer in the field, I do sympathise with your views. A few additional points. EO business follows the "anti-Steve Jobs model". Jobs said: “a lot of times, people don’t know what they want until you show it to them.” In EO market, the good clients know what they want much better than business developers". Why? Because EO data only provides indirect information. When looking at images, nothing beats the experience of having been in the field. Only those with dirty boots really understand what informations needs to be provided. There are too many developers who have never been to the trenches of real life. All machine learning algorithms have biases, but those developed without extensive support from clients are usually hopeless. I have been a leader of the longest running operational service that uses EO data (deforestation mapping in Brazil which started in 1988) I can assure you that white-collar developers in jeans and tennis do not stand a chance of succeeding by themselves. Best, Gilberto Camara (www.gilbertocamara.org).
Really fascinating take on the industry. I agree 100% that a company should pick one thing to do well and focus on that. I do think there is a market for derived data though, as long as it is being derived with a specific audience in mind. One-size-fits-all analytics will never work. I used to work at a place where we could get approval to buy derived analytics but never the people-power to derive it ourselves. We didn't really have the option to homebrew something better and I know our peers/competitors were in the same position. That being said I also saw plenty of half-baked derived data feeds because it was more cost effective for them to get their offering to 80% than 100% (or even 95%). I kept sampling those but they were never good enough for our needs.
The problem is more prominent in classification algorithms. We are mostly trying to predict "a class" which we do not have a consensus on. I was trying to classify crop type of agricultural fields. When I talk to agriculture experts, they all explain a different "corn field" which is highly dependent on the farmer. Are we talking about corn fields which the seeds are planted in 30cm or 50cm? Are they going to be harvested for animals or industry? Any definition by human itself is biased. We are trying to fit a biased reality in our mind using biased models.
Great points. Couldn't agree more. Definitely see that as data, standards, tools, and models become more ubiquitous, large enterprises (like the one I work in) will increasingly be developing and deploying these solutions internally. As you point out the true value of commercial earth observation will come from building applications; for many applications in big orgs these will be built and deployed internally where these lean on integrating multiple proprietary data streams (e.g. sensor networks, operational data etc...). It will be very interesting to see how the EO vendor space responds to these developments. Exciting and fascinating times!!!
I don't know if it was a marketing post for king of the hill or a destructive criticism for a business model.
Good stuff. Please follow up with "how to peruse EO substack without getting fired."