Agralogics and The Internet of Food

Agralogics calls itself the “Internet of Food” company. Based in Sunnyvale, CA, Agralogics helps farmers by providing them with detailed analytics about their crops – information about everything from weather, thermal energy, soil quality, pollination, and much more. The firm is a unicorn of sorts in Silicon Valley, where tech-driven innovation in agriculture is secondary to flashier, consumer-facing products.

The DBJ sat down with co-founder and CTO Sanjay Dayal to learn more about Agralogics and the food ecosystem.

Dartmouth Business Journal (DBJ): Tell us about the data you collect.

Sanjay Dayal (SD): We look at data not as a means to an end, but almost an end in itself. First, we take data from the public domain. There’s a lot of data around satellite imagery, weather, soil, crops, pests, plant phenologies, growth stages, and much more.

Then there is partner data. There are sensor networks, which are generating an amazing volume of data, at increasing velocity.

Then there is private data, which comes from our customers. This could be collaboration noise or ‘chatter’, it could be how much water they applied, or how much pesticide they are using.

We can see that there’s an explosion of data. Whenever there’s an explosion, there is blindness. Despite so much data being produced, it’s not actually making things clearer. Because of the volume of data, you don’t know what’s relevant and what’s not.

To give you an example, there are soil sensors, which can tell you about 160 different characteristics of the soil. How do you map that to what is relevant to your crop production at different stages? It’s a hard problem. The simplest way is to do a linear regression, try to find what’s important, and then do some kind of closed loop sampling. But it just doesn’t work, because next year, the temperatures are completely different.

So I think what is needed is not necessarily giving data to our customers, but really giving them context.

I’ll give you another example. The CDC does flu modeling. It says “this is how flu is going to spread.” Google does that simply from its users, who are doing searches on ‘flu’. It can actually find out how the flu is spreading better than the CDC, just by looking at searches in different parts of the country. Data has a lot of insights, but just data itself is so much that for most people. It’s unusable. What we do is extract context from it, and then provide that context to our customers.

“Data has a lot of insights, but just data itself is so much that for most people, it’s unusable.”

DBJ: With data coming from the government and sensors, do you foresee problems with expansion in areas where that infrastructure doesn’t exist?

SD: We are very agnostic of where the data comes from. We are not dependent on the data coming in a specific format. We have built our backend to consume as much or as little of the data as is available. If we only get data on high and low temperatures, we will consume that and extrapolate other characteristics. But then there are weather stations which not only give you highs and lows, they give you surface temperature, humidity, wind speed, all kinds of stuff. And that type of data is pretty much available for most regions in the world.

There are at least eight or nine public domain satellites which give you reasonably high resolution data for the entire world. More and more, people are also moving towards putting sensors in their soil. And there’s a lot of math in the background. For places where you don’t have these things, we can do a first-order or second-order approximation.

Now, as more accurate data is available, we can improve our model, and our ability to predict. For regions you don’t have data, you do sampling and then modeling. For regions you have more data, you are more accurate.

I would say that for 90% of the globe, most of the basic data is available. And since this is just the start, we believe that in 10 or 15 years, because of the pressures of global warming, population growth, and unpredictability of weather, we will have more and more precision, which we’ll need when it comes to how we produce, distribute, and consume our food.

So I think your question is very valid in that there are regions where data isn’t fully available, but there are workarounds.

DBJ: Suppose I am a farmer, and I say “My family’s been doing this for generations, I don’t need anyone telling me how to grow my crops.” What would you say?

SD: We are not trying to tell you how to grow your crop, because we know you do that much better than us. But what we can do is make your life easier. Things that you need, the information that you need, we can provide readily to you, on any device you work with.

What crops do you grow, and where do you grow them? It’s a mathematical problem, basically. But the variables are so many, and the volume of data is so large, that individuals cannot do it. It’s not only about how much water you put in, it’s also about the other things. Is my land too wet or too dry? What phenological stage is my plant in? If you’re spraying a chemical the week before flowering, your flowers may not pollinate properly. Things like that are very contextual. Growers know about these things, but we want to make this information so easily available that they do not have to work for it.

Here’s an example. Food production doesn’t happen on Gregorian calendars. You cannot say, “I have planted my tomatoes in April, so I will get a harvest on the first of August.” Tomatoes don’t grow that way. Tomatoes grow based on how much thermal energy is given to that plant. If it’s colder, it will take longer. If it’s hotter, it will take a shorter period of time. You might have heard of the Pennsylvania groundhog, Punxsutawney Phil, which people use to predict whether the winter will be short or long.

And guess what? Most farmers have this clock in their heads today. They say “March is warm this year, so I think I should plant early, because this year kind of resembles how it was four years ago.” They are all working in their minds. What we can do is give them a calendar based on a thermal clock. This tells them that in two months time, this much thermal energy will be given to your plant. And this amount of thermal energy looks like what you got three years ago.

Just that information is very important for the farmer to say “I’ll do what I did three years ago, because I had a great crop that year.”

A lot of planning in the farm happens around when it is hot or not, when it has rained or not rained, how much water has been given to my soil, both by nature and by me? How is the soil losing water, because that determines when I need to water my plants again. All of that is done by heuristics, and because conditions change so fast, your past wisdom may not be applicable to your current situation.

What we can do is help by doing a large-scale analysis of, for example California, and understand how water availability is changing based on reservoir levels, aquifers, etc. We can predict that, and tell a farmer, “these are the new areas that you can expand,” if their current areas are becoming less productive. All of that is very data-driven, and that’s where we think we can help the entire food ecosystem.

The biggest problem for today is that people work for data – data doesn’t work for them, especially in the food ecosystem. We want to turn that on its head.

DBJ: Do you think Agralogics will get to the point where its technology can predict a drought, or a massive crop blight?

SD: Absolutely. I think some of that work is already done by some companies. You might have heard of Climate Corp. Their claim to fame was that they could provide rates for crop insurance based on the weather.

What we have is not only weather-based data, but we have soil data, private data, management practices. The combined data is a much better predictor of success than just the weather. In a few years, I think we’ll have enough statistically significant data worldwide that our system could start to approach that problem.

Right now we’re focused on California, which provides a great sample. We have customers from 24 different crops, across all counties, using our platform. We can actually, in an anonymized way, what’s happening in their fields, both from remote sensing and their private data. From this, we can make larger scale predictions. We’re not there yet, because we need more data. But I think we’ll be there soon.

DBJ: In this day and age, data privacy is very important. But without using private data, is it possible to make such predictions regarding droughts and other phenomenon?

SD: All private data is never shared with anyone else. This is like Google doing an anonymized analysis of your Gmail, but not sharing those contents with anyone. It’s in a very similar spirit, where we only use anonymized data from our customers for our analysis.

Coming back to your first question, ‘can this be done with only public data?’ It can’t. That’s why Google has the public data, but they can’t do it. There is no feedback loop telling them that their analysis is right or wrong. It’s like a scientist only having theories that they can’t test.

So it’s very important to have that private data.

DBJ: You mentioned ‘chatter’. Could you elaborate on that?

SD: When we talk about data, we aren’t just talking about ‘transactional data’, such as how much water or pesticide I used. It’s also about how you came to the decision to apply this much water, what ‘chatter’ happened prior to that decision. Think about a decision to apply a certain amount of water being a communication between the ground staff and the field supervisor. We want to capture that chatter, and see if there was a better way to collaborate. Because unless you optimize that process, the outcome will always be suboptimal.

It’s not just about what happened, but also how it happened.

Around that, we are creating collaboration. So think about a Facebook wall, where field staff collaborate with everyone else around pesticide management, land management, pollination, and the food supply chain. Collaboration between the grower and the processor. Collaboration then between the processor and the distributor. It’s a highly connected ecosystem.

If you look at the food graph, it is superconnected. It is more connected than Facebook. Facebook is a relatively uniform graph, where each person has around the same number of friends. The food graph is much more connected. The way the farmers are connected to consumers: it’s not six degrees of separation. It’s sometimes two degrees of separation,  sometimes 10 degrees of separation. So the graph complexity is pretty high. And these are not reciprocal relationships. So once you look at the graph and say, my god, this graph has to be the basis of any information flow and understanding, you come to that “aha” moment. It’s the data and the ‘chatter’, in the system which needs to be captured first, in order to optimize what’s going on.

DBJ: What has been Agralogics’s biggest challenge?

SD: Every startup has some basic challenges that all startups face. Having a startup is like having a baby. I have two, so I always compare them. Besides the work, it is a lot of faith. You need to believe that what you’re proposing has value. Most of the time, that value is not seen immediately. There is a lot of that effort and proselytizing you have to do, for people who are the change makers, to see that value.

When you try to drive something as big as what we’re trying to do, you cannot do it alone. We need the opinion makers, the people who can make things happen, to be on our side. And we believe that it has to be a much bigger effort than who we are – a tiny little company.

Especially for Agralogics, I see that as a big challenge. We are not trying to come up with a better algorithm to match consumers with products. We are trying to disrupt the food ecosystem. There are very powerful stakeholders who we have to work with, and convince that what we are proposing is good for all of us. And that’s definitely a challenge. We will continue to do that because we believe in this.

DBJ: Where do you see Agralogics in 10 years? or maybe 5 years?

SD: Where we want to be is a ubiquitous platform for anything related to food collaboration. Collaboration can be as simple as the ability of a food processor to inform its customers about what is in their food. For example if there is a pesticide scare, I as a consumer should be able to scan my label, and get a response saying that this particular product is completely free of pesticides. Or the label should be able to tell me exactly how many miles away this was produced. Or it should be able to tell me its actual nutritional content. Today, an apple is an apple is an apple. It doesn’t matter whether it’s an organic apple, whether it grew in rich soil, or whether it was a hydroponic apple. You get the same nutrition information for all of them today. So that is one type of collaboration, where the consumers want to find out information about food.

The other type of collaboration is that which enables growers to grow their crops more optimally. Suppose a bank is trying to give a loan to a farmer, and they want to understand their risks. Right now, the farmer would have to submit documents, monitoring of his fields. The bank guys have to physically visit the fields to make sure the crops are growing.

And then there’s a predictive side of it. Based on all this data, we can start to help countries, financial markets, future markets, to plan better on how to grow food and feed humanity.

But unless we have the data, we’ll never get there. So our first phase is to make sure that the platform is ubiquitous, or at least used by enough people, and enough stakeholders in the ecosystem, that we can go to our second purpose, which is to plan at a much larger scale, at a county, state, national, or even continental scale.

DBJ: That’s great, because my last question was going to be you see any humanitarian potential for Agralogics?

SD: That’s what drives us. It’s a big dream, but all big dreams start small.

DBJ: Is there anything else you wanted to mention Agralogics?

SD: We are about a year old company. One thing I would say is that there are some visionary investors. We are just a seed stage company right now, but we already have some of the most visionary investors in Silicon Valley, who are willing to part with some of their money in pursuit of this dream, which is very satisfying, because sometimes me and [my co-founder] Soumesh say to ourselves ‘We’re not smoking dope, are we?’, and when there are so many very smart people willing to help you, then I think it’s a confirmation that it’s some good stuff. Support of people who are influential is what I believe is the key for doing something at a scale of the problem that we’re trying to address. So far, we have had very good encouragement from people who are the fathers of the Silicon Valley.

Now we have customers, who have validated our solution. Most of the customers we have presented this to are wowed. They tell us that “we were waiting for something like this.” We are very happy to report that there is a strong adoption that is happening as we speak.

DBJ: You’ve worked at startups before. How is it different as a co-founder?

SD: The startup journey is very challenging. This is my first startup as a founder, but I have done two more startups before this. One of them went public, and one of them busted. Then I was at Sybase in its very early stages pre-IPO. I have seen phenomenal successes, and phenomenal failures, but mostly from the eyes of an employee. This time, I am seeing things from the eyes of a founder, and it’s very different. But it’s very exciting. Because that ability to pursue a dream, and seeing the milestones, is more satisfying than anything you can imagine as an employee. The only other feeling better than this was the birth of my sons. It’s at that level.