Neatleaf is like a Terminator for plants: it never sleeps, it never eats, and it will never ever stop monitoring your plants to make sure they stay alive. If that was good enough for Sarah Connor, it should be good enough for weed growers, but what are we even talking about here? The Terminator is a time-traveling cybernetic organism programmed to follow orders without fail, but is it also able to grow great weed? The answer is, not quite yet, but the promise of the Neatleaf technology – essentially a data monitoring and collection system on steroids – is so vast that few ideas are off the table. In the meantime, the Neatleaf Spyder is proving its worth in actual grows, where its eye-in-the-sky robot effortlessly delivers reams of data that heretofore was collected by fallible human beings.
Elmar Mair, Neatleaf CEOAs explained by Neatleaf CEO Elmar Mair during a call with Cannabis Business Executive, the value the Spyder offers to cultivators today is immediate and evident, and it’s only just getting started. We spoke on the heels of an announcement that Neatleaf has partnered with iAnthus Capital Holdings, owner of 37 retail cannabis dispensaries in seven U.S. states, “to bring Neatleaf’s AI technology to the company’s Pleasantville, New Jersey cultivation facility, which supports the company’s MPX NJ brand.”
Mair noted right off that integration of the system had begun far before the announcement. “We basically have been running our system in the New Jersey facility at MPX since last December,” he said. “They wanted to try it, they were intrigued by it, and so we started with one system, and they saw the significant value that it provides. They had some great results in the following growth cycles by being able to measure microclimates, monitor issues, quantify plant stress, measure bud sizes and such, and last spring they rolled out the system in the whole facility. Every room, every zone, has got their own system, and now we’re talking about rolling it out more broadly.”
He described the system as basically a cable-based robot. “It’s like a football stadium camera, a sky-cam for greenhouses and indoor environments,” said Mair. “You mount four corner pieces in the corner of a greenhouse bay or an indoor cultivation room, and then you have the cables and at the center there’s a box with sensors. We measure all environmental conditions – temperature, humidity, CO2 – we measure plant growth by measuring plant height, we measure leaf temperature, which is key because it tells you how quickly the plant is growing and transpiring, we measure light intensity, and we also measure images. The platform moves around the canopy constantly and keeps an eye on your plant. It identifies each yellow leaf, necrotic leaves, and detects buds and measures flower sizes, and so it quantifies your cultivation.
“What we know from any other production line is the first thing you do if you want to optimize production is measure the items, and you detect quality by identifying issues, and you do all that in automatic fashion,” he added. “In cultivation, we have been relying on humans so far to go in, look at the plants, understand what’s going on, summarize it, somehow report it, and it’s all very error prone and tedious and not consistent at all. So, the system automates that so the cultivators can focus on what really matters, which is dealing with issues rather than identifying issues.”
A Robot for All Seasons
I asked Mair how the system incorporates different growing techniques, and how it knows what to look for in a particular grow. “That’s the whole point,” he responded. “Everyone has their technique, different substrates, different nutrients, different knowledge and expertise, and it’s all based on people’s subjective experience. That’s how we have been operating, but it’s not data driven, which causes so much friction. At the same time, you have to nail it every day, and if you screw up one day, it has a huge impact on the yield, so there’s a lot of anxiety around ensuring that everything works smoothly. You need many devices and lots of infrastructure – lights, fans, HVAC, and humidifiers – and they all have to work reliably, and they don’t always do that, so there’s a lot of stress there, too.
“But the Neatleaf system itself does not need to know about any of that,” he added. “All it does is measure everything, so independent of how you operate and what the conditions are, we move around and measure all these parameters in a spatial way. We create maps over time so you know what’s happening across your zone, and if something goes wrong, you can immediately measure it.”
If a data point comes back that’s problematic, does the system know that it’s problematic, or does a human with that knowledge still need to interpret all of the data points coming in as well as the actionable solutions that need to be done? Is that where the AI comes in? “There are different aspects to AI,” replied Mair. “One is, obviously you have certain set points. That’s why you say, ‘I want my environment to be at a certain temperature, humidity, light, and whatever.’ The system now measures the microclimates and if a corner is not according to set point, it notifies you.
“And there’s another aspect to that,” he added. “These are the set points for environmental, but we also have the image feed, which is great for humans to be able to go back and compare and see how the plants looked a day ago, two days ago, a year ago, when they grew that one specific cultivar the last time. The AI actually can pick up on every yellow leaf, every necrotic leaf, and now you can say things like, ‘I have 5 percent more yellowing,’ which a human will never be able to say, because you can’t go in there and count that. It has to be very drastic and significant so that you can pick up on that signal and that change, and then it’s often too late. So, being able to pick up on these things very early on allows you to act long before a human could actually do it, so it saves you a lot of money and improves your performance.”
What about seeing below the canopy? How does the system evaluate all the stuff going on underneath the upper leaves? “The system has a top-down view as it is right now, but it gets different viewpoints,” answered Mair. “It doesn’t just look at it from one spot. Now, the one thing that you have to keep in mind is that plants are like trees. They grow in a way that exposes as many leaves to the light as possible. So, as it grows in this tree-like structure, you actually can see very, very deep into the canopy already. You might not see every leaf, but you see most of them, which is something we were very excited about when we tested the system for the first time.
“There’s one more aspect I want to highlight, which is really key to these early detections,” he added. “It’s our leaf temperature sensor, which measures leaf temperature and compares it to the air temperature, so you know how much the leaf is cooling. The leaf cools because it transpires. It pulls the liquids with the nutrients from the substrate and uses them to produce biomass and to photosynthesize, and as it does so it basically transpires and cools the leaf. So, the bigger the gradient of the difference between a leaf and air temperature, the more the plant is operating. That’s a signal no one has access to – not in full-canopy coverage fashion – and now you can use the signal to tweak your environmental parameters, depending on what genetics you grow, etc., to really have the plant optimally grow. The first thing that happens if the plant is stressed for whatever reason – if it’s a pest or virus, or some nutrients issues, or whatever – is that it stops transpiring.”
Does the system alert operators to the changes? “Correct,” said Mair. “It measures those values that you want to optimize, and it is not reactive cultivation anymore. Now, it’s reactive where these plants are stressed, they are super yellow already, and the damage is done at that point.”
Can the system be linked directly to automated operations, like watering systems, HVAC, or dehumidifiers? “Currently, the focus is on analysis, monitoring, and reporting with automated reports on what’s going on,” said Mair. “The next step, which we are rolling out in September, is the integration of substrate sensors. You’re going to be able to measure what’s going on in the substrate as well, and then you will have the full picture of what really matters for the plant. The following step is tapping into the control protocols and becoming like a source of knowledge about input for the environmental control systems so that you can start automating and tweaking those set points in the most optimal way.”
As far as adjusting the cultivation environment, that is currently done manually, added Mair, but the system has additional advantages. “The huge value-add for iAnthus and all other clients is that they can measure things that they couldn’t measure before, and they can quantify and compare across growth cycles,” he explained. “But the big MSOs also have different experts at different facilities, and those directors of cultivation are spending a lot of time in airplanes flying from one facility to the other to talk to the teams, trying to figure out what’s going on, and they’re relying on what the team reports to them or sends them via pictures on WhatsApp, and they report every day that it looks great, and then it’s harvest day and it doesn’t look so great.
“It’s very challenging and that’s why they love this system,” he added. “They even say that it is going to save their marriage because they don’t have to fly that much anymore. Now, they can remotely log in and see the plants, they can go back in time and understand what happened to the plant, and they can provide guidance. They can also comment on every element, every leaf, every heat map, every line chart, and tag people and say, ‘Hey, Joe, can you check this out and do a lab test here?’ It becomes a very, very useful tool for them to actually operate a cultivation.”
This is the point where the possibilities take flight. It sounds as though at a certain point in time not that far in the future grows will be able to speak to one another without a human being involved. The reason this is an issue is because cannabis flower production is notorious for its lack of uniformity and consistency harvest to harvest. Even with the same strains under the same conditions, results vary, even with the same cultivators in charge. One could argue that’s a part of the plant’s charm, but if the goal here to increase consistency and uniformity over time, which most all CPG brands seem to want, this technology would seem to augur such a future.
“100 percent,” said Mair. “It is a huge issue, and if you talk to those MSOs, they say the biggest challenge is that their sales team sells something, or they get some orders in, and they asked for 10 percent strain A and 50 percent strain B, and then a couple of months later, once they actually have the results, they get 30 percent strain A and 5 percent strain B, and it’s a huge problem for them to actually operate.
“It’s unlike in any other domain, if you think about the marketing effort, inventory effort, and the whole logistics effort,” he added. “It’s always data driven, and the focus is always on trying to show consistency. You have these crazy differences every cycle, and that’s what has to change. The reason why we have that is because it’s such a complex space that has to be navigated, and we rely on humans to look at everything and digest everything and make these decisions, which is just very, very challenging.”
Is the system also expected to improve quality? “Understanding where you can save and where it makes sense to save is one aspect, but quality is obviously the next one, and consistency always blends into quality,” said Mair. “But also, pushing the limits of the plant and being able to ensure that the plant can operate at its utmost performance. Each plant is a production line for itself, so it’s about really understanding that every genetic is different. It’s something where I think it’s just scratching the surface of what’s possible.”
One would also assume that utilizing such a system would reduce or remove the need for flower remediation post-harvest. “You can prevent those things by detecting them early, because then you can actually do something about it,” agreed Mair. “Whether it’s powdery mildew or any other thing, when it happens, you can prevent the conditions that actually led to it, and then you don’t need any of those things. The key is ensuring that you can measure those things consistently and reliably across the full canopy, and all those measures should not be necessary.”
Intriguingly, the Neatleaf system is designed to deliver actionable data immediately and only gets better at doing it. “We have these systems running now in Europe, in Canada, in the US, in so many different facilities, and the AI, which we have trained on the data which we have collected so far, generalizes across facilities,” said Mair. “As soon as we install it, you get quantified plant stress and all those things out of the box. The AI will adjust to your environment and become better and better, but it’s already providing a very solid quantification of it, and at the end of the day, the actual absolute numbers, especially around plant stress, are not as critical. No one needs to know that I have 400 yellow leaves or 405 yellow leaves. What’s important is to know that I have 3 percent more yellow leaves today than yesterday, and I have 5 percent fewer fallen leaves than yesterday, so that you can see the trend. You see basically how the plant reacts to your actions and your input, and that’s the key.”
Data Driven
The entire system is built on data, of course, and the cumulative value of that data is potentially enormous. So, who owns the data? “We have collected a lot of data over the two-and-a-half years that we have been running these systems,” said Mair. “The way we think about it, and I think the only way to think about it, is that the people who generate the data are the ones who own it. That basically means our clients own the data that you capture in their environment. But we have the right then to train our AI on it, so we can train our plant stress models, our detection models, all the things on top of that, which does not cause any IP concerns because it’s agnostic to any operator. Being able to pick up on yellow leaves doesn’t harm anyone, and that’s our IP.
“Obviously,” he added, “those are our models, our proprietary AI, and the way we look at it is that we don’t know every environment. As you said, every cultivator operates in a very different way, and there’s no holy grail in terms of how you want to operate a plant, because it really depends on what the environment looks like, what HVAC system is used, what humidifier is mounted, what irrigation system you have, what water you can use. So, I don’t think that there’s a holy grail, the one recipe which works everywhere, and that’s not the goal. The goal is to have a tool that allows you to find the optimal setup, the optimal parameters, the optimal processes, to operate in the most efficient and the most optimal way in your environment, with your devices, and with your input. That’s why there’s no risk of giving something away, because all of that is very specific to what you grow and how you grow it, and the tool allows you to optimize it for your environment.”
That said, one can easily imagine scenarios ten years away where AI-generated plant data has market implications and value that’s market-disruptive. “Definitely,” responded Mair, “and that’s why we are very excited to work with a lot of breeders, which you know are constantly hunting for new phenotypes and changing the genetics. The idea is that as you develop a new genetic, you can actually then run our system in that environment and identify the most optimal way to grow that one specific genetic, and then you can sell your genetic together with the recipe which needs a certain performance. It’s a win-win-win situation.”
Mair said Neatleaf has around 60 systems in the field and is installing new ones almost every week. “And it definitely keeps evolving,” he added. “We are introducing a new system, the next generation, probably at MJBizCon, that’s going to allow us to scale even faster. I can’t talk too much about it, but we are very excited about it.”
The actual system is also being scaled. “The current version is designed to cover up to 5000 square feet, and the next version is going to be able to cover up to 10,000 square feet,” noted Mair. “It doesn’t matter about the environment. The one thing is that it’s designed for single gear, so not multi-gear; that’s just how it’s designed right now. Otherwise, it’s very flexible and easy to install. In general, we install it ourselves, but we have clients who have installed the system by themselves.
“What we offer right now is that if there’s any interest from a new customer, we will install the first system for free,” he added. “They can just rent the hardware, and there is no commitment necessary. They basically can try it for as long as they want. We really believe in the product. We don’t want to sell a contract. I hate it when people make you sign a one-year contract. Why? Don’t you have trust in your quality?”
What about using the system itself, managing all the data points? Is it a steep learning curve? “With our system, the software is like a dashboard on the web that you can open up from any device, like a mobile phone, tablet, or your desktop,” said Mair. “Then we have an in-house data science and plant science team that will work with your onsite cultivation team to show them how the system works, highlighting certain things you can observe in the space, showing experiments you can conduct and how you can optimize and push the limits of your cultivation. So far, this has been very, very successful, and teams have embraced that they can really push the needle in terms of what they can achieve with yields, not just labor saving. So, yes, it comes with a whole package, which is part of that deal.”
I asked Mair what he is hearing in terms of feedback from people who have been using the system. “What they really love is that the system basically reports any anomalies or anything suspicious that is seen in the environment, anything out of norm that they can act on,” he replied. “It takes away all the anxiety for cultivators to constantly look at what’s going on with the plants. They can rely on the fact that there’s a second pair of eyes, or more, looking at it in an automated way. Every leaf is looked at multiple times a day, so that’s really, really helpful for them in terms of confidence that they know things are trending right. And the leaf temperature is really helpful, because that allows them to optimize the performance of the plant.
“There are other aspects to it,” he added. “To have an archive of all the data so that you can go back and see how the plants looked at the beginning, and when they started to deteriorate. ‘Oh, there’s a one leaf that is burned in a certain way, like in necrosis, which is a certain pattern. How did that happen?’ We try to figure out when this happens, but we’re also able to figure out what caused it. It allows you to troubleshoot and compare it to other growth cycles and compare it to other facilities. Why does this genetic perform in this way here, and it performs very differently in my other facility in another state? Now, you can overlay the growth data, the plant growth and development, if you have numbers to talk about, and that’s what I really love.
“What they’re also calling out is the lack of substrate information,” he noted. “That was the one missing piece, understanding what’s happening in the soil. That’s why we have been integrating substrate sensors and will be rolling that out in September. It’s very exciting.”
I observed that the system is obviously valuable to large grows, but it’s easy to see how it could also be of value to smaller craft grows that want to create a very high-quality product; not more of it, but of higher quality. “The smaller cultivators really enjoy the system, because they get an understanding of the microclimates,” said Mair. “They ensure that they can achieve this same consistency over and over again. I think that’s one of the aspects, but we can also loop in other experts, friends, and colleagues. If there’s an issue, if something is going on, it can give them access and they can see the plants, how they evolve, and can provide feedback. It makes it very easy, because often, if you are smaller, you don’t have all the expertise for every task or issue in-house, so being able to access that knowledge is incredibly helpful. You may have an in-house team to help with that, but they can also move anyone into their network.”
As far as interest in the system, “We are definitely expanding and growing,” noted Mair. “We see a lot of excitement in the market about this system from big clients and many others who see the value of what this can mean for cultivation, and how they can solve a lot of the pain when it comes to operations. We are installing more and more systems, which are very easy to install, so the idea is to work with facility teams so they can install the systems themselves, which makes it far more scalable.”
Financing the company’s expansion is also underway, as are other growth initiatives beyond cannabis. “Thus far investments have come from venture funds, and we are currently doing another round of fundraising because we are seeing a lot of demand, and we want to fulfill that,” explained Mair. “We are also getting outside of cannabis, using the system with other crops, which is very exciting. The system is not necessarily tied to any specific crop, it’s a monitoring tool. But cannabis is our beachhead and a very exciting market for us, because the plant itself is incredibly interesting, one of the fastest growing crops, but very, very challenging to grow, and there’s very little research. It’s basically like driving a race car in the mountains without having a map of what turns are ahead.”
The Promise of AI
Looking ahead, it is within that unknown aspect of cultivation where artificial intelligence would appear to truly flex its muscles. “There are a million AI companies out there, and they all tweak one or the other language models,” said Mair of the Neatleaf approach to AI. “That’s not the differentiator. If you think about it, what is gold these days are the picks and the shovels, being able to access data and having a unique data set which no one else has. That’s our system. We have that, and we can do things and offer features that no one else can. You can summarize the performance, show the growth cycles, business analytics, forecasting, yield, all those things that no one else can do, and that’s where our system is very different.
“I worked on AI for different other projects, very complex systems, like self-driving cars,” continued Mair. “What I learned is that you have to have the right data, and you also have to have a comprehensive set of data. And with respect to the right data, one thing I noticed when working on AI and machine learning is that it isn’t sufficient to have a stationary camera somewhere, just staring at the plant. You need to get different viewpoints of the objects that you want to learn about, which is also how a human operates. They don’t go in there, look at a plant, and then just stand there and stare at the plant and try to assess what’s going on. No, they go in there and try to really figure it out, and that’s how you get reliability and robustness in your models.
“The other thing is that you need a comprehensive idea of what’s happening in that space,” he added. “If you have gaps, if you don’t know how the plant looks, if you don’t know what’s going on in the substrate, you will never be able to exclude things, and you’re not going to be able to figure out what’s happening, and the AI can’t converge. With our system, we now have that, and we can create models that can understand what the root cause of an issue is rather than just the symptoms. We’re at the stage where we can identify symptoms, but the next step is for you to wake up in the morning, you get a summary of your daily crop in the email, and it says, ‘In these 10 zones, eight are okay, but in one, there is a 70 percent chance for nutrient deficiency or a 10 percent chance for a virus. Joe is available. He can go and do a lab test for that virus.’”
I told Mair that can imagine an experienced cultivator getting his morning report and then going back to the system and asking it, ‘Give me your 10 top reasons why this is happening,’ but he also has his experience and knowledge to layer on top of that. “Totally, and that’s a revolution,” enthused Mair. “It’s going to take some time, but one thing we have probably all learned in the last years with ChatGPT and the impact of that is that AI is really, really, really good at making sense out of a complex space, out of a multi-hyper-dimensional space, where you have all these different things that are correlated and involved, and that’s what cultivation is.
“You have all these factors – temperature, humidity, CO2, airflow, nutrients, irrigation, light – all the things that are related to each other,” he added. “If you change one, you have to change everything, and currently, humans have to ensure that they know how to do that for every plant, every genetic, etcetera. It’s just impossible. I am amazed how cultivators are actually navigating this space, but it’s also crazy the pressure and risk they’re exposed to.”
Amazingly, when you spend time with legacy cultivators with 20 or more years under their belt, they seem to know in their bones what the system offers. “Exactly, and that’s what triggered me,” said Mair. “I talked to this one expert cultivator, and he was like, ‘So how do you operate? What do you do if you come to a room and there’s an issue. I used to go in and just feel that something is off.’ And I was like, ‘What you mean, you feel it?’ And what that actually means is that there are cues which they pick up on visually or by the skin or by smell, which the brain tells them, ‘These are not consistent, it doesn’t make sense, something is off.’ But even they can’t say what the cues are because it’s so complex and they are so minor, and that’s where AI has shown itself to be really, really good at understanding those decision boundaries.”
Before we rang off, I asked Mair if Neatleaf, despite starting with cannabis, is ambitious enough to have the entire agriculture industry in its sights. “I mean, the system is agnostic to the crop, and cannabis is a very, very exciting crop for us,” he reiterated. “But as you develop a new technology, just think about it: Tesla started with the highest margin market, the Model 3, so you start with the high-margin markets, and you work your way down. It’s the same way we looked at the horticulture space. What’s the highest margin – it’s obviously cannabis – but we are already exploring the next markets, and it is very, very exciting.”
Copyright
© Cannabis Business Executive
