Growing the Future: How AI Is Transforming Agriculture

Bryan DeAngelis: Welcome to this week's episode of What's At Stake. I'm your host, Brian DeAngelis, a partner here at Penta.

Andrea Christianson: And I'm Andrea Christanson. I'm also a partner and I head up our global AI task force. Today we're talking about one of the most complex and consequential systems in the world, the global food supply, and how artificial intelligence is transforming the way we predict prices, manage risk, and plan for the future.

We're joined by Francisco Martin-Rayo, CEO of Helios AI, a company pioneering AI driven insights for agriculture and food supply chains. Francisco, welcome to the podcast.

Bryan DeAngelis: Great to be here. Francisco, I'd love to just jump right in with your story. We don't have a lot of CEOs of, uh, AI agriculture companies on the pod, so we'd love to hear kind of how you got into not only this technology, but then now you got into this business as well.

Francisco Martin-Rayo: [00:01:00] Yeah, absolutely. So it's, it's really fun to be here and it's a little bit of a, a weird background, I would say. I, I spent a lot of time at BCG as a principal there doing digital transformations. I ran a couple AI businesses, but then a, a long time ago, I used to import avocados from Mexico. And then my co-founder and CTO Eden, John Alar, you know, senior AI ML engineer at Google, she'd won the Abbey Award.

She also used to own a bunch of restaurants. And so we had these fairly unique backgrounds of kind of food and tech. Um, and what one of the things that we had noticed is, and we mean this apolitical. Is that, you know, climate was changing and the weather patterns were changing pretty dramatically and we felt that that was only going to accelerate.

So we knew that we had this looming crisis in the food supply chain coming on board, and we were gonna see crazy price volatility, a lot of disruptions. That was kind of a negative hypothesis, but the positive hypothesis for us, just given our [00:02:00] technology backgrounds was. We really felt that technology could make an enormous difference and we could build this, you know, engine and brain or platform, whatever you wanna call it, that could predict a lot of these disruptions before they happened.

Andrea Christianson: Yeah, when we were talking last week, we were talking about the global food supply chain, and while supply chain has been in the news a ton, a lot of people think about cars or electronics, and you said that the food supply chain is probably the most complex supply chain. So I'd love for you to talk a little bit about that and, what your product does to help, normalize some of the supply chain issues.

Francisco Martin-Rayo: Absolutely. If you, if you, if we take a step back and we think about it. The last time you went into the supermarket, let's say you may or may not have bought, but you probably saw mangoes somewhere. A lot of the mangoes that you have on the shelves in the United States of America come from a very rural province in Peru called Tambo Grande.

Uh, you know, mangoes are perennial trees, which means they're always there. And so [00:03:00] somebody right at Kroger or Walmart or Safeway or Publix, whatever your, you know, retailer of choice is, is contracting out 18 months ahead of time with that grower in rural Peru so that they're able to, you know, contract it out.

You wait 18 months until you know, you have the flowering, you have the harvest, and then you have to transport that mango from rural Peru to a port, from that port to another port in the United States, and then onto the shell. If anything goes wrong in those 18 months, right from a climate perspective or labor strikes or price volatility, you're not gonna get that mango, and that means you're not gonna have the mango on the shelf, or if this is one of the key ingredients from some of the products that you're creating, you're not gonna be able to create that product.

And so what's so fascinating about us is there are 10,000 things that can go wrong. And if any of them go wrong, you don't get the end product, which is so important. And yet, despite how sophisticated and how complicated this is, [00:04:00] because you're worried about geopolitical uncertainty, you're worried about climate, you're worried about the supplier risk, you're worried about, you know, uh, tariffs is a good example, right?

And what that impact is and where they're gonna be . a lot of the tools that these procurement managers are using are, haven't really changed since the eighties and nineties, um, which, which to us is extraordinary, right? You have these, uh, you know, really senior, highly experienced procurement managers that know their job better than anyone and are dealing with this extraordinary level of uncertainty and, and technology hasn't really caught up with them to help them do their jobs better.

Bryan DeAngelis: Well, let's jump there. I want to hear more about how you're using particularly artificial intelligence to, to do that and catch up. I think you told us before, you cover now about 75 different commodities across 90 countries. Yeah. Break down the technology itself for us. What kind of data and and methods are you looking at to forecast that?

How do you [00:05:00] ensure. Like you said, anything can go wrong in that time, that these predictions stay reliable over that same period too.

Francisco Martin-Rayo: Yeah. Yeah. So we co It's exactly right. We cover 75 commodities across 90 countries. That's more than anyone ever has in the history. Um, we really started initially, you know, three-ish years ago, really focusing on weather and what those weather disruptions could be.

And we cover, you know, we have 500 billion climate risk data points. We take daily weather readings from 14 million locations around the world. That's temperature, precipitation, humidity, wind speed. It's super cool. And then we've built individual machine learning models for each of those crops because you know, cocoa's very different than soy is very different than tomatoes, is very different than corn.

They have different growing seasons, they have different temperature ranges, they care about different precipitation ranges. So we've put all of that into individual machine learning models. So every day across these 500 billion climate or data points, [00:06:00] we're constantly looking at. How is this global crop doing?

We also cover 90% of all the places in the world that are growing these crops for export, which, you know, prior to Helio being around, no one did that. And so if you had asked us, you know, before we came along, Hey, how's the global tomato crop doing? You probably would've looked at a couple USDA reports that were six months behind.

We love the USDA. It's not a criticism, it's just, it's a tough business, right? That's typically when you're putting out your reports or you'd be looking at something on Reuters or something, right? That happens every three to six months and is also behind. And people didn't really have that global coverage of all the places in the world that were growing these commodities for export.

And so we had to build that, that knowledge base had to build these individual machine learning models. And then what we launched about six months ago is. In addition to the climate risk, we also forecast all of the prices of now these 2000 plus price series globally [00:07:00] that'll get to 5,000 by the end of the year.

And so we forecast that out 12 months in advance. What's really cool about that is because we're covering all of the places in the world that you're growing these commodities for export, because we have access to, you know, 250,000 global news sources. We have historical pricing data, we have climate data, we have seasonality data.

We've put all of that into our pricing engine, and so our price forecast are five to 20 times better than than the industry average because when you look at almost your traditional competitor. And this is insane, guys. You have a single model that you're using to forecast the price of plastic and potatoes, uh, which to us is still insane.

Yeah. So we've just built those individual models for each of them. So that's the core of it. And then we, we just launched something really cool, a multi-agent platform, which we can talk about. It's a muddy point of helpful.

Bryan DeAngelis: Well, I'm curious too, where do you, um, if you do, and I assume you do combine the human expertise [00:08:00] with, with that AI work.

Francisco Martin-Rayo: Yeah. I think one of the things that's really special about kind of the industry that we get to work in, how familiar are you guys with like horizontal versus vertical ai and, and, and your listeners, how familiar do you think they're. Getting there. Awesome. Right, so, so what's cool is like horizontal ai chat, we love it.

We use it all the time. And you know, you can ask it anything from, I have a four month old so you know, how much, uh, milk should my four month old drink all the way to, can you help me draft this email? That's super cool. But when you ask it something like, Hey, where do we think the price of bananas is gonna be in six or 12 months?

And what are the biggest risks that I need to worry about? The way a generic LLM works is it'll look at, you know, probabilistically, what's a lot of the contextual information that I have. It's gonna do a prediction based on that, and it's gonna give you an answer that's probably going to be incorrect.

When you look at [00:09:00] kind of our age agentic platform, it's called Helios Horizon. It has access to, you know, all the proprietary climate data we have, all the proprietary pricing data, we have all of the workflows, plus a lot of the new sources, all the USDA reports, all the European commission reports, et cetera, et cetera.

And so then it's able to take all of that information, site it, and give you the right answer about where banana prices are gonna be. So this, this is a, a perfect example of what we call vertical ai, right? Where you need to deepen that age agent AI instead of the generalist. But where it's most helpful is when you have humans side by side.

And so the way it works for us is you've got your highly experienced procurement managers that know their space better than anyone, and they're able to ask specific questions. They're able to put in specific workflows that require that level of expertise to really be successful at. Otherwise, if you don't have that level of expertise, you won't know what the right [00:10:00] questions are and you won't know what the right processes to ask, uh, for it to do are.

Andrea Christianson: Well, you know, it, it seems really transformational for a lot of procurement folks. Is there a kind of favorite story or case study you can share of, you know, how, how this works in practice?

Francisco Martin-Rayo: Absolutely. So our, one of our favorite customers is Libby's. You know, Libby's is the second largest, I think, second largest buyer of pineapples and mandarins in the United States.

Um, we, we've been working them for a long time. Um, we saved them 15%, uh, recently on one of their costs of good sold, they were buying Mandarins, we had flagged to them, Hey, there's gonna be a big disruption in Mandarins. The price is gonna pretty dramatically. Our recommendation is for you to buy a lot more and a lot earlier than you would otherwise.

The beauty of it is, right then they take this information, they put it to their procurement team, and the procurement team is able, one, to test the assumptions that our platform is [00:11:00] making and then go have conversations with their suppliers. They went ahead and bought a lot more and a lot earlier than they would've, and it saved them 15%.

The beauty of the 15% is all of these businesses are very low margin. I mean, your top businesses are making seven to 8% margins. So when you're able to save someone 15% on that cost of goods sold, it's incredibly impactful.

Andrea Christianson: That's awesome. And we've talked a lot about the procurement side of this, but talk to me a little bit about the farmer and producer side of this.

So, you know, a big thing in American farming families, and I think. All over the world is how do I make sure that my children and grandchildren can, you know, continue this if they so choose. And so talk to me about how maybe your technology could help, uh, families as they think about, you know, building for the future.

Francisco Martin-Rayo: Yeah. So in, in our experience, you know, farmers are some of the most innovative group of people we've ever worked with. And what's [00:12:00] really interesting about. Kind of the, the innovation and that adoption scale is, you've got about 40 seasons, right? And so you're really thoughtful about any new technology that you're gonna put in.

Because if you, if you do it incorrectly, if you do it too fast, if it's too expensive, you lose, you know, one of your 40 seasons, right? You're not gonna be able to get that back. But what's interesting is once you're able to prove. To farmers, that farmers it is cost effective, it saves you money and it works.

That adoption curve is, is pretty extraordinary. Like nothing we've ever seen in any other industry. This is probably one of the hardest times historically for farmers in the United States. Um. Not just soybeans in general, right? Which we, I think China has now started buying again, but had stopped buying for a while.

Um, but you are seeing this change in weather patterns. I mean, a good example of that is Florida oranges. Florida used to do 200 million boxes of [00:13:00] oranges a year. That's down 98%. Uh, vast majority of orange juice that we drink in the United States is now Brazilian orange juice. Right. Um. So one of the things that we're able to do for growers, and, and I'll say, you know, for any growers that are listening, we make the platform available to most of you guys for free.

Uh, so please reach out. We are so happy to work with you. We're happy to give you access to the platform for free. Um, one of the best ways we work with them is we give them access both to what the price predictions are gonna be and also to what all of their competitors domestically and internationally are doing.

We're never gonna know your field better than you do. We're definitely gonna know your competitors field and international fields better than you do, and we're gonna have a pretty good indication of where we think prices are going and when. And so a lot of the growers we work with are using that information to better inform, you know, how much do I sell?

When do I sell? And even in some cases, based on climbers. You know, what, what is the, the ratio and the different [00:14:00] combination of crops that I'm planting? There's a very large farmer in the UK we work with who uses our system to better inform, you know, should I grow more potatoes and more onions in the field that I have, they have, you know, 10 or 20,000 hectares.

Bryan DeAngelis: You know, Francisco, there's a third actor in this that I keep coming back to, particularly because of where we sit, which is the, the government, and we hear so much about what. If the government, you know, does for farmers or in some cases maybe makes it harder on farmers, and it's also such a, you know, obviously critical, you know, need for us with our, our food supply and the disruptions that happens.

So talk to me about where government is playing a role or maybe could be playing a bigger role as they deal with some of the same disrupting weather patterns and other issues that we see out there.

Francisco Martin-Rayo: Yeah, so we, we don't work with government yet, but I, I will say, um, we've seen [00:15:00] pretty good indication so far from Secretary Rollins that artificial intelligence is something that they really care about and, and they're really thinking of how can the USDA be a better bridge?

To bring a kind of cutting edge technology to farmers, which we think is awesome. I think that is a wonderful role for USDA to play, and I think it's, it's really exciting that they're thinking of being so innovative about this. Um, and so I think really the role of the government is, you know, how do you de-risk a lot of these newer technologies for farmers at scale?

How do you make it free or much easier for them to engage with them? Uh, how do you taste test them for scalability? Because, um, you know, we're, we're, uh, very friendly with the International Fresh Produce Association, the wonderful group. We were part of their accelerator program. We were the only software company in their accelerator program.

Um, and so for us, scalability is built in, right? It doesn't matter if I have 10,000 customers or a thousand customers, it, [00:16:00] it doesn't really change the software that much. Um, but everybody else was more of a true ag tech hardware type. Where they really struggle, understandably, is, you know, you come up with a good prototype that takes you one or two years, 'cause you have to test it out.

Then you go to a bunch of farmers and you ask them to test it out. Some of them are gonna say, all right, we'll test it out next season, right? Leave me alone, I gotta finish this. And then you're gonna test it out in 5% of your crop, understandably, right? You're not gonna put a hundred percent of your crop at risk.

And then maybe if that goes well, maybe the year after, you're gonna do it in 10 or 15% of your crop. And, and one of the things that's hard for a lot of these Ag Tech startups is they're gonna run out of funding and they're gonna die on the vine before you're able to get to scale and really prove it out and get that revenue model.

So one of the things that the DOD, the Department of Defense does really well is it has these kind of SBIR and STTR programs. Okay. And you know, from a DOD perspective, I think it's, I dunno, quarter million dollars, one and a half million [00:17:00] dollars. Nothing for the dod, it's, it's very meaningful for a lot of these ag tech startups.

And so one of the, one of the places I think that USDA and other kind of ag folks agencies should start looking at is how do you increase that type of program to fund some of the more innovative technologies through that chasm. Because a lot of them will get there and there's really cool stuff that's happening.

They just, they just need that bridge to go from, right? Yeah. We've got a few, you know, farmers and growers that are trying this out to, Hey, we've reached scale. It just took us that, you know, two or three seasons to be able to do

Bryan DeAngelis: it. Is that a issue in many jurisdictions, same kind of deal in the EU or the uk or Asia, or do

Francisco Martin-Rayo: you No, we're at a huge disadvantage against the eu.

Um, just to be really transparent, I think the amount of support that they give their farmers is, is pretty extraordinary. Not just from a subsidy perspective, but from a tech perspective. [00:18:00] But, you know, our three top competitors are all UK based. Um, and the European governments, uh, basically fund those competitors so that they can go to their farmers and give them this technology for free, which is a huge edge.

Against their American counterparts. Um, you know, because they can better understand our Americas, uh, you know, how's the harvest growing? Where do you think the prices are gonna be and they can better position against us. Um, we, we should, we should do a little bit better there. Yeah.

Bryan DeAngelis: I have one last question for you, and maybe Andrea has a couple follow ups too, but, um, we always like to talk about what's next here at Penta.

So, you know, fast forwarding maybe a few years or even a few months, who knows, you know, as. As you all mature, more as a company, have more adoption, what is the, what does the future of AG look like with these tools or in the future of the company maybe?

Francisco Martin-Rayo: [00:19:00] Yeah, so I'll, I'll talk about ag broadly and then I can talk about Helios Ag broadly is one of the places where I.

I am just so excited to see that the impact that, you know, artificial intelligence will have. And what's really cool about where we are today in AI's journey is you're seeing a bifurcation, right? I think there's a lot of jobs that are disappearing, which we're, we're very worried about. Um, but in areas like agriculture, education and healthcare, there's so much potential.

Um, and so from an ag perspective, if you think of. You know, the way the majority of farming goes, um, it's relatively sophisticated in the United States, but probably you're still putting out, you know, fertilizer in general or weed killers in general. You're irrigating in general. Um, the beauty of artificial intelligence is you're going to be able to do that in a hyper custom way, almost on a per plant basis at scale.[00:20:00]

You know, using drone technology, using robotic technology, using just sensor technology. Um, and that's gonna make an enormous difference because then you're spending less water, you're spending less fertilizer, your margins are gonna go up, your yields are gonna go up. And so I, I do really think we're entering this golden age where you're gonna see increasing yields and increasing profit margins.

Which I think is gonna be really great for producers. And the question is, how do we, how do we just get through that, the chasm we talked about? Um, but I think it's gonna be a great five to 10 years for Helios. One of the most exciting things that we launched is this, you know, multi-agent platform, Helios Horizon.

Um, and it, it really is like a chat gt, but specific to kind of ag procurement, right? And ag in general. What's really cool about it is, you know, we're a SaaS company and, and the pain with SaaS for the last 20 years is you, you put up a dashboard and you wanna put as much information on the dashboard as possible, right?

Yeah. But then the [00:21:00] dashboard kind of sucks, right? Because you're, you're, it's a really difficult time to navigate it, and then you try and customize it. That makes it really hard. But if you don't put a bunch of stuff on it, then it's not helpful either. The beauty of agentic AI and, you know, generative AI is.

You don't have to do that anymore. You can just query and create these analysis at scale using natural language. I mean, people have asked questions to our platform and Chinese, not Chinese, uh, Japanese, Ukrainian, um, French, Spanish, uh, and that's, that's awesome. Right? They don't have to go into this dashboard that we've spent so much time building and, and we think is so pretty, but it's still imperfect because everybody has different questions.

Right. And so that I, I think, is just the continuation of the future. Everybody will be able to have this hyper customized software and hyper customized insights at scale.

Andrea Christianson: Well, you've captured kind of the optimism of the future Very well. And, um, you know, we're at Penta too. We're, we're pretty big AI optimists, [00:22:00] uh, ourselves and feel you on the platform Magenta piece of it.

And, and, and. Going a little bit in your direction, but with more of a media intelligence thing. But, um, Francisco, thank you so much for being with us today and helping us better understand your company, the technology, the future of agriculture. Um, and we're really grateful you were here. Thank you.

Francisco Martin-Rayo: Great to be here.

Thanks again.