Latest / Giant Robots Smashing Into Other Giant Robots / 421: Benchmark Labs with Carlos F. Gaitan Ospina
Transcript
- CHAD: This is the Giant Robots Smashing Into Other Giant Robots Podcast, where we explore the design, development, and business of great products. I'm your host, Chad Pytel. And with me today is Carlos Gaitan, the Founder and CEO of Benchmark Labs, which provides IoT-based weather forecasting solutions for the agriculture, energy, and insurance sectors worldwide using proprietary machine-learning software. Carlos, thank you very much for joining me. CARLOS: Thank you for the invitation, Chad. It's a pleasure to join you here. CHAD: You work in a variety of different industries with weather forecasting solutions using machine learning. I'm really curious, at a high level, how did you get to where you created Benchmark Labs today? CARLOS: Oh, thank you, Chad. That's a great question. I think that in many ways, it's a combination of life experiences and lots of user feedback. As a background, my mum worked for 28 years in the National Federation of Coffee Growers in my native Columbia. And we experience basically the effects of weather, La Niña, El Niño, local conditions, pests on the coffee growers. I remember growing up looking at the price in The New York Stock Exchange if the pound of coffee was going to be more than $1 or not [laughs] and so on. So, you know, we had a very severe drought in Colombia, and Colombia was heavily dependent in hydropower at that time. And I remember that we even had to study with candlelight and move to a spring savings time for the first time in the country. The country is in the equator, so you can imagine moving the clock was unheard of. So since then, I was always passionate about hydrology, the water cycle, why this happened, how weather can affect the economy at that level that people have to change their working habits. I did civil engineering hydrology, then studied these new applications of machine learning technologies, hydroinformatics, did my studies there in Columbia, my bachelor's, my master's. Then I was fortunate to go to The University of British Columbia to study my Ph.D. in Atmospheric Sciences. And then, after I finished, I moved to The United States to work at the Geophysical Fluid Dynamics Laboratory in Princeton with close collaboration with the NOAA, the USGS. And that gave that perspective also of understanding how weather climate models were done at the Department of Commerce level but also to understand the users on how they interact with weather data or climate data and what were the needs that they were expecting from the National Weather Service and the Department of Commerce and NOAA that not necessarily were fulfilled with the current information. So then I moved to the private sector, joined a hardware company, and met my co-founder of Benchmark Labs there then moved to California to work on consultancy of climate change assessments. But since the time at the Department of Commerce, it became very clear that what farmers and what users wanted was weather information that was more actionable, that was tailored to their specific location, especially for specialty crops. Think about wineries, or coffee growers, orchids, stone fruits; they depend heavily on weather, and the information from the National Weather Services was just too coarse for them. And sometimes, there are huge errors in terms of temperatures that were recorded from their farm versus what the National Weather Service was doing. And that's why we decided to create Benchmark Labs to basically solve that problem, correct those errors, and give the information that the users needed when they needed it. CHAD: Did you ever just consider becoming a TV weather person? CARLOS: [laughs] CHAD: It seems it may be easier. CARLOS: [laughs] Nah. That's a very good point. CHAD: [laughs] CARLOS: And I have great respect with my colleagues that went into forecast meteorology and TV persons. I remember some of my lab mates practicing in front of a green screen when we were doing the Ph.D. CHAD: [laughs] CARLOS: That was an interesting scenario. [laughs] However, growing up in Colombia, the weather forecasts were not very, let's say, accurate to a certain extent, and we did the opposite than the weatherman suggested. CHAD: [laughs] CARLOS: So I guess that steered me towards following that path. [laughs] CHAD: So it totally resonates with me this idea that, you know, especially for...I've been on the West Coast before where you go over a hill and the weather it's like 20 degrees hotter and sunny and on one side of the hill, it was cold and foggy. We went on a great company trip many years ago to visit some Napa vineyards, and I was surprised by that. So I can imagine how that local information just doesn't match the global information that farmers might be getting. So what is the hardware that you're actually producing, and what is it doing? What does it look like? CARLOS: [laughs] Great question. So I will go back to your story about Napa and Sonoma, and the reality is that's exactly a problem that growers face; national weather agencies give averages over a big region. They divide the world in boxes, and everybody inside of a box receives exactly the same forecast. And if you are especially in the coast or you're in specialty agriculture, you understand that weather changes with elevation. Depending on which side of the mountain you are, you could receive all the rain or no rain at all. If you are near the shores, you could also get more wind, different types of clouds, all of those situations affect the conditions at the farm. And going back to the situation of Napa and Sonoma, Burgundy or the Mediterranean Basin, they all believe in the value of what they call the terroir, that is what makes also unique their products. They're indigenous, and they understand at a very fundamental point how the local conditions from the soil, from the vegetation, makes their farm unique. So what we do is we use IoT sensors, basically hardware sensors that monitor environmental variables. We refer to them in the atmospheric science world as weather stations. I had a talk with some users when I said the term weather station. They imagined a big construction or a building with a TV station on a radar or something. But in this case, there are IoT devices that are totally portable, the size of a Wi-Fi modem in some cases. And we use those sensors as ground truth that will basically tell us the local conditions. We use the information from the National Weather Services and the information from those IoT sensors and correct the forecast as they come. CHAD: And is that where the machine learning comes in because it's actually correcting the forecast being received? CARLOS: Exactly, our machine learning aspect of it is fully operational, non-linear correction of weather data as it comes in from the National Weather Services to correct it to the conditions that are experienced at the farm level, at the sensor level. And a farm could be also an agricultural farm, or it could be a solar farm, a wind farm. Or, as we talk with some users in ski resorts that actually they consider as snow farmers, it's also affected by microclimates. So at the end, it is about providing value to all these areas affected by microclimates that are not being resolved correctly by the current generation of forecast from the National Weather Services. CHAD: Are most customers able to get the coverage that they need with one weather station, or are they deploying multiple ones? CARLOS: So that's a great question, and the answer probably is it depends. Our customers, original customers, have thousands of stations over multiple fields under management. For specialty crops, it's common to have multiple IoT sensors in one acre. For other scenarios, they might have only one station or one sensor every 10 acres or so on, so it depends on the condition. It depends on how technologically inclined are the users if they already invested in these IoT sensors or if they are looking into buying IoT sensors and then scaling up the number of sensors in their farms. CHAD: How do all the sensors report their data back? CARLOS: That is a very interesting question because they are, let's say, tens of hardware manufacturers globally. We also created kind of a Rosetta Stone that puts all the sensors to communicate to our back-end systems. We integrate different languages of each hardware manufacturer. It has its own ways of naming the variables. So we do the translation in our end. We receive the data via an API. These IoT devices are Internet of Things in many ways because they transmit data via Wi-Fi, satellite internet, you know, cellular. CHAD: Cell, yeah. So different manufacturers might have different ways of actual communication, not just the protocol, but one box might be using Wi-Fi, and another one might be using a satellite. CARLOS: Exactly. And sometimes, many manufacturers give you the options of connecting even using Wi-Fi or Bluetooth for IoT sensors that are near, let's say, a farm that has internet connectivity. If they are on the field farther away, they might need to get access to a data plan from a cellular carrier, 3G usually or 5G. In some areas, there is limited coverage so far. And if it's a very remote area, there are options to get satellite coverage. CHAD: Now, I'm asking somewhat naive questions based on my understanding. And so if I start butting up against proprietary information, just tell me, "No." That's totally fine. CARLOS: [laughs] CHAD: So when we're thinking about the amount of data coming in from all of these different weather stations that your customers have, is it a lot of data? Is it a lot of data points? CARLOS: [laughs] It's a great question. So in many ways, yeah, each weather station communicates at different frequency. Sometimes what we are offering now is hourly transmission rates, but we also have access to government stations that sometimes only refresh once per day. So yes, it's a lot of data coming in, most of the data from the weather stations. Fortunately, it can be transmitted as a txt file, or it's only for one location. So the files are not big, but they are many per day. And so, we have probably done millions of operations already to assimilate data and provide the forecast. While on the other hand, The National Weather Service provides one forecast for the globe, let's say every...some models are every hour, other models are every six hours, and so on. So that is more, let's say, a bigger data set because it's a global data set that then you have to query to extract the information locally that is relevant for your servers, for your users. CHAD: Yeah. And I think it's neat how this is all happening centrally from all the data coming in, right? CARLOS: Yeah, exactly. We get data coming in for each specific location. We do the corrections, and we provide the forecasts. So there are lots of operations involved in the data handling activities, pre-processing, post-processing, but it's very rewarding at the end to provide the forecasts that are tailored to specific locations. And we had seen users that they basically told us, "Okay, we are using provider B or C; can you beat them? Show us that you can beat them, and the contract will be yours." So we showed them, and then they are like, "Yeah, that's fantastic. This is exactly what we have been looking for, information that is more accurate for our farms," so yeah. CHAD: Now, does your system correct itself based on what actually happened in an area after the modified forecast goes out? CARLOS: That's not a very relevant question because some of the models are static. I used my experience when I did an internship in Environment Canada, and I found that they were adjusting their models, let's say four times per, at least the operational models they had, four times per year. They kind of tweaked them to the local, let's say, spring, summer, fall, winter conditions. In our case, we make our models to correct themselves as more data comes in so they can adjust to weather events and have short-term memory, let's say, of what they will wait heavily on and forget the distant past. CHAD: I mean, it seems obvious, not necessarily easy but obvious, that you've made a prediction about what the weather is going to be, and you have all the data coming in from the stations to confirm whether your prediction was correct or not. So I'm sure it's not easy to adjust the model based on that. CARLOS: [laughs] CHAD: That seems obvious to me. CARLOS: Yeah, it's just a different approach in many ways. As you said, it's obvious because the users usually care about a specific location, at least our users. We understand that for national security or aviation, they require a model that provides coverage over a wider area, like sometimes continents. But for agricultural users, they care about their farms, and the farms will not move in space. So -- CHAD: Well, technically, they are moving in space; it's just the weather goes along with it. CARLOS: [laughs] So yeah, I guess that it's just a different way of tackling the problem. We focus on doing these forecasts to each specific location instead of having a forecast done for the whole globe that could be used in many different locations or for many different industries, but it's not necessarily tailored to any industry-specific or location-specific. CHAD: Yeah, that's great. Mid-Roll Ad I wanted to tell you all about something I've been working on quietly for the past year or so, and that's AgencyU. AgencyU is a membership-based program where I work one-on-one with a small group of agency founders and leaders toward their business goals. We do one-on-one coaching sessions and also monthly group meetings. We start with goal setting, advice, and problem-solving based on my experiences over the last 18 years of running thoughtbot. As we progress as a group, we all get to know each other more. And many of the AgencyU members are now working on client projects together and even referring work to each other. Whether you're struggling to grow an agency, taking it to the next level and having growing pains, or a solo founder who just needs someone to talk to, in my 18 years of leading and growing thoughtbot, I've seen and learned from a lot of different situations, and I'd be happy to work with you. Learn more and sign up today at thoughtbot.com/agencyu. That's A-G-E-N-C-Y, the letter U. CHAD: So have you managed to bring it full circle now, and are there coffee growers in Colombia that are using your solution? CARLOS: [laughs] I hope so. We have talked with coffee growers for sure. They care about temperature gradients. And I really think that going to Colombia as we scale will make the whole platform easier to use. I think that we can go full circle soon, sooner rather than later, into Colombia. We got support from the World Trade Center here in San Diego to do commercialization assistance to translate our solution from English to other languages. So we will be tackling Spanish, French, Italian in the very near future because it's important to offer the forecast also in a way that they could interact natively without having to have the limitation of using an English language platform into their day-to-day life. But yeah, full circle probably we’ll be going full circle soon. CHAD: So language is one barrier to scaling and to adoption. Are there other ones that are typical barriers of adoption for your customers? CARLOS: We are very competitive here in the North American market, the European markets. Our prices are in dollars. But that by itself is a problem for emerging economies; for example, you know, $100 here is not the same thing as $100 in other countries. We have to take into consideration exchange rates or the amount of disposable income that they will have for their operations. CHAD: And I'm not super educated about it, but I know that there are certain industries in agriculture where the growers are particularly pressed for margins, and coffee is one of them, right? CARLOS: Exactly. So, fortunately, in many ways, for the bigger crops, specialty crops they are traded, and the prices are linked to U.S. dollars so that can be translated, our services can be absorbed, let's say. For the smaller crops that are not traded or that just stay locally, the price is not linked to the U.S. exchange; then it's definitely a bigger barrier for them. But hopefully, we will get to a point if we have a sufficiently large adoption in North America and the developed world; these technologies could be subsidized or made more accessible in other economies. CHAD: What are some of the concerns that growers have? Take the specialty crops, for example, is it a matter of are they doing this because they want to make the best product possible, or is it because they want to prevent crop loss? CARLOS: It is both, actually. The uses of weather information in agriculture varies, as you said. There are many different applications; one is to get more actionable alerts. For example, we saw what happened in Burgundy last year where a substantial part of their region lost their crops, close to 80% maybe. I don't remember the number, but it was definitely substantial. And so, having more accurate forecasts and alerts gives them an opportunity to adapt better, to get cover, protect their fields to a certain extent. Weather information affects also pests and disease models, so application of fertilizer with spraying is also affected by local conditions. In many ways, for the operations that are very, let's say, sophisticated, some of them even link the sugar content on the fruit to weather conditions. And understanding how these weather conditions affect sugars could tell them when is the optimal time for them to, let's say, harvest? And the difference in the sugar content might determine the difference between higher margins or so-so margins [laughs] for their yield. So yeah, it's a combination of quality of the product. It's a combination of preventing loss of the product. And it's also labor scheduling and activities, for example, that are regulated by OSHA that prevent farm operations to maybe don't, let's say if they are like temperatures above 95 Fahrenheit or 100 Fahrenheit. So having that extra information in alerts will also help them with farm management operations. CHAD: So can you give me a sense of the stage you're at or the scale you're at now with the business and where you see your next stages of growth being? CARLOS: Thank you. Yeah, great. So we are fortunate to have scaled this solution beyond California. We are now a global platform. We are providing forecast to Spain. Recently, we got contacted by some growers in South America, so we are testing for avocado growers in Brazil and Colombia, for example. So I'm not serving yet coffee growers in Colombia, but the avocado growers in Colombia, it seems that they got a hold on what we do, [laughs] so it is getting there. And now we have the resources, the ability to go global and offer this anywhere in the world that is connected with an IoT device. So it's fully operational. And we are now in the midst of fundraising to scale the team, provide the customer success operations, and to support growers in different geographies, to support growers of different crops. And I think that if we are going to be successful globally, it starts with customer support, customer success, and understanding your users' needs, so they don't feel that, again, they will receive a one size fits all vanilla-like solution and that we really care about why specialty crops are special. CHAD: So when you were just starting out, who was the first team member that you added to the team? CARLOS: Oh, it was great. So in many ways, I thank the Economic Development Council of San Diego for funding a set of interns in data science, weather analytics, and business development. So our first hires, in many ways, were supported thanks to the Economic Development Council. We were the two founders, and then we got support in business development to understand which, for example, specialty crops really care about weather. Then some data science interns, data scientists that helped us with grants that we did for the National Science Foundation, and NASA that we got...we supported one of the grants. During COVID times, we participated in a very interesting opportunity to know the effect of COVID on forest fires, for example, and that was in collaboration with NASA. So first hires were interns, entry-level positions in data science, in back-end engineering, and then front-end business development. Now we are very excited to be expanding the team. We recently hired a Chief Product Officer with ten years of experience in Bloomberg, experience with visualizations, and talking to customers and users. So I think that for us, it's very important to, again, I reiterate, to have the ability to provide a great user experience, to provide meaningful information for specialty crops so they feel that they are special. CHAD: You mentioned that you got some business development help using those grants. But right now, is the actual sales work being done by the founding team? CARLOS: Yeah, at the beginning, as a founding team in a small startup, you have to wear multiple hats. So yeah, it's very common, and in many ways, I appreciate that we didn't rush to hire in terms of sales too early because it's important that the founding team understands the user perspectives, their needs, what they call the pain points to understand how to steer product into that direction. And then sales will follow once you have a solution that is highly needed, that users really like and that it can be shown that it can be scaled globally. So we are working on scaling, on accuracy of the forecasts. And yeah, next hires will be to get somebody that will help us in sales and can bring us to the next level. CHAD: What does the sales cycle look like for the kinds of customers you have now? Do they tend to be smaller, or do they tend to be larger enterprise customers? CARLOS: So, in the beginning, we worked with smaller enterprises to understand how to use the data, for example, connect the data from one or five sensors transmitted online. So dealing with smaller enterprises, farmers was optimal at that point as a company. And now, we are focusing more on businesses, farm managers, or management companies that have hundreds, sometimes thousands of sensors on their management. So we deal with more like business to business instead of going direct to grower at this stage because, as we were mentioning earlier, we're a small company, and going direct to grower requires lots of support and dedication in terms of dedicated agents and sales teams. CHAD: Do those companies tend to have long sales cycles? CARLOS: The bigger ones, yes. If you are talking about publicly traded companies, they will want to start with pilots then validate them. And you can move at different timescales with them that are not necessarily aligned with the startups at this stage. But there are some farm managers that have a way higher frequency of decision making. So their sale cycle could be one month, two months instead of having to build a relationship for years. CHAD: You mentioned the pilots, and you mentioned earlier telling the story about a customer that said, you know, "If you can provide us with better data," but I think companies as they scale or as they talk to potential customers, you also don't want to take on too much work that you should be charging for to be able to do that pilot. How do you strike that balance? CARLOS: It's a fascinating question. And I think that from a founding member perspective, let's say, it goes as a function of the stage of the company and what other, not necessarily monetary, benefits you can get from these pilots. We have been even recommended to not have unpaid pilots anymore, for example. I think that it's important at the beginning to get access to the information that you need to validate the technology with users that really care about what you're building. And sometimes, there are different ways that these pilots can be structured in a way that the final user might give you a reference or might spend time with you doing the quality control, quality check, saying what kind of features they like, so that's also very important as a young startup. As you grow, probably once you have that validation, there is no need necessarily to take into endeavors that will lead to unpaid pilots that you don’t know if there's a clear end to that. And you can move to a more structured pilot program that has clear deliverables, and at the end of window, a decision will be made depending on the set of topics that were agreed between the companies. CHAD: You might even be able to get away without pilots if you can make a strong case by showing other case studies that are relevant to that potential customer or where you explain, oh, you know, these people had a similar situation to you and here's how it's solved, and here's the success that they had. CARLOS: Totally. You nailed it. It's in many ways to sometimes build credibility, find analogues in the sector, or a use case that can be comparable to the pain point that another user might have. And it could be, let's start with the avocado growers in Brazil, and they have probably the same pain points that they have with avocado growers in Colombia. Once we have that sorted out, then we probably can go and talk with avocado growers here in California or Mexico, Central America and tell them, "Hey, this is the value that we've unlocked in Brazil. Do you have a similar problem?" CHAD: What I have found is that this is one of the important reasons why you have to have a good product which is part of what you've been saying all along, you know, you really wanted to focus on making sure the product was working and that it was good. Because when you do, then you can also use referrals, you know, not referrals, but like, hey, you want to talk to this avocado grower, and they'll be happy to talk with another potential customer because they're excited about what you've done for them and been able to do with them. CARLOS: Totally, totally. And agriculture is always open to new technologies, but they are traditional in many ways. And it's a small circle, and I think that it is very important to build products right and really care about what you're doing and your end-users. Build together. Don't come necessarily with assumptions saying, "Hey, here agricultural grower A, I have a solution that will change your life," without knowing necessarily where are they coming from and their life experiences, and how they interact with products before. So yeah, I totally see the benefit of referrals. Word of mouth is very big, going to conferences with agricultural growers. There are big networking events that could help us more than just going and doing a Google ad campaign, for example, at this stage. CHAD: I think that's probably an important lesson that not only applies in agriculture but in a lot of industries. And I really appreciate you stopping by to share with us. And I really wish you the best of luck as you progress in your journey at Benchmark. CARLOS: Oh, thank you very much. I really appreciate it, and I hope that we can continue the conversation here. Just count with us anytime that you need to talk about weather, agriculture, IoT sensors. Happy to help the audience too, and always discuss what's out there to help the Giant Robots community. [laughs] CHAD: Carlos, if people want to get in touch with you or find out more about the company, where are the best places for them to do that? CARLOS: Go to benchmarklabs.com and then fill out a form there. And we will definitely be in touch with all of you. I will personally answer all the queries. I'm very, very happy to share our technology, share what we are building. And we are so excited because by having this technology, you can help save water, energy, and even pesticide use, and that's a huge contribution to the environment as we move forward. So yeah, thank you very much again for the invitation, and I'm here; count with me as a future resource. CHAD: Wonderful. And you can subscribe to the show and find notes and links along with an entire transcript for this episode at giantrobots.fm. If you have questions or comments, email us at hosts@giantrobots.fm. And you can find me on Twitter at @cpytel. This podcast is brought to you by thoughtbot and produced and edited by Mandy Moore. Thanks for listening, and see you next time. ANNOUNCER: This podcast was brought to you by thoughtbot. thoughtbot is your expert design and development partner. Let's make your product and team a success.