The Center for Data Innovation spoke to Yasir Khokhar, co-founder and chief executive officer of Connecterra, a Dutch firm that uses AI and sensors to track the health and fertility of dairy cows. Khokhar talked about how AI can address some of the classic problems facing farmers, as well as help improve efficiency to meet future food needs.
Nick Wallace: When I first heard about what Connecterra does, it was described to me as “fitbits for cows.” What gave you the idea to try something like that, and can you tell us a little about how Ida, your AI platform, tracks cows’ health?
Yasir Khokhar: The idea came up almost by accident. I’ve been in the tech industry for a while, I used to work at Microsoft. I moved to Holland about four years ago. At the time I was interested in what was happening with sensors and machine learning, and I just started playing around with this in my free time.
As it so happened, I was living on a dairy farm at the time, in a little village south of Amsterdam. I needed to put my sensor on something that moved. I talked to the farmer and a friend who was also a dairy farmer, and I asked, “hey, what if I put my sensor on a cow?” So they said, “well, what can you tell us about the cow? Can you tell us where she is, how much she walks, how she eats?” And I said, “I don’t know, it’s just a simple sensor that I’m mounting on a cow, but I’ll try to figure it out.”
Agriculture is a fairly inefficient process, if you think about it, and technology can really help with some of the problems, and it just kind of took off from there, really, with that one “a-ha” moment of saying, “what happens if you put a sensor on a cow?” It turned out to be a very important space that sensors, machine learning, and artificial intelligence can help with. That was about three years ago. We were given some research grants by the European Union to continue our work, and that led to the development of Ida.
But Connecterra as a company is really focused on trying to solve some of the bigger problems that we have in front of us today, using AI and sensor data. That’s our core focus. We run a full-on AI platform, with learning algorithms both on the farm and in the cloud. So the sensor that goes onto the cow transmits its data to a base station that stays on the farm, which runs a portion of our AI algorithms.
We’re the only company today that can detect seven unique behaviors of a cow, like eating, drinking, walking, standing, laying idle, or chewing. But that’s not very interesting for farmers: most customers don’t really care about the data, they care about what it means. That’s where the second set of our AI algorithms comes into play, where we are now using the movements of the cow to detect and predict health issues, like preclinical mastitis [an infection of the udders], or the age-old problem of detecting when a cow is fertile.
Wallace: Some analysts like to talk about today’s new technologies ushering in what they call a “fourth industrial revolution,” 200 years after the first. But the first agricultural revolution was 10,000 years ago. What problems are you trying to solve: new ones, or problems that have gone unsolved for all that time?
Khokhar: The fundamental problem that we’re trying to solve is that of scale. So if you take a couple of steps back, the food question is essentially asking how you feed four billion more people in the next 30 odd years when you’re not going to have any increase in land or water. You need a 60 percent increase in food production, but how is that going to happen? That basically points to a need to make agriculture more efficient. There are obviously other aspects to that story where alternate sources of food are being investigated, but by and large, the bigger problem is really going to be about making food production more efficient.
In that respect, we’re going after the efficiency problem and the sustainability problem. The ultimate goal of Ida is to figure out how to run the world’s most efficient dairy farm, and then branch out from there to looking at how to make farming more efficient in general, whether it’s a dairy farm or an arable farm.
The kind of problems that you need to solve to make farming more efficient are mostly well known. The challenges are around scale, particularly when farms get bigger and farmers get older. Or you have farming countries like the Netherlands, where farming is quite efficient—though it could be more efficient—and countries like the United States, or developing countries, where farming isn’t as efficient, even though the cows and what you’re farming is basically the same.
The difference between inefficient and efficient farms is usually around operational practices and the skills of the farmer, and that’s where the technology comes into play. By using Ida you can see a cow’s milk production go from ten liters a day to thirty liters a day, simply by knowing what’s wrong with her at the right time.
Having said that, we’re also discovering new things. I’m not at liberty to discuss those in detail right now, but because something like this has never been done before, we are discovering new patterns of behavior, we are looking at new insight models, which are quite unique, and it’s only by using optics like Ida that you’ll really be able to get those insights out.
Wallace: What are the typical benefits to an individual farmer of implementing this kind of technology?
Khokhar: Today, Ida does three clusters of insights. The first is around fertility. Cows only produce milk when they’ve just given birth. So getting a cow pregnant, at the peak of her ovulation cycle, is really important for a farmer. Most farmers will miss that for up to three cycles. That’s about 60 days, or 60 to 100 days in some cases, where you’ve got a cow that’s eating, but not giving milk.
What you’ve essentially got at this point is a sunk cost, because feed is expensive—it’s 70 percent of the operational cost of a dairy farm. So in this instance, if you can reduce the time lost through human error in predicting the peak of the ovulation cycle, you can get cows pregnant earlier. As a consequence, they will start producing milk earlier, and also be more productive because they’re not just eating without producing milk.
That’s step one, and it’s a really important metric for a dairy farmer. We can help reduce that by a significant amount. We’ve seen some cases where the cycles come down by 20, 30, or even 40 percent: it really depends on the farm and the farmer.
The second cluster is around feeding. We can detect problems of efficiency in the cow converting food to milk—so how much does she eat? Is she eating healthily? When there’s a problem, usually what that means is the farmer needs to change the type of feed that he’s giving to the cow. Today there’s no way for farmers to figure that out on the level of individual cows. Usually they give the same meal to all their cows, and what that means is cows that are predisposed to produce less milk with a certain type of feed will be less efficient. We can give those insights, and also give early warnings if there are digestive problems—again, really important for a farmer.
The third one is really around health. Looking at infections like preclinical mastitis, which is a very typical issue that happens. Just to give you an idea: if you have a cow with a serious infection, and her milk gets added to the rest of the herd’s milk, that entire milk tank needs to be then discarded, because you cannot have infected milk go to the purchaser. So being able to detect health problems early not only saves the life of the cow—because when it gets too late, usually they have to be culled—but also saves the farmer the headache and the significant financial loss of discarding an entire tank of milk.
We’re also working on new issues, some of which we haven’t really announced yet, but they have to do with a lot of other issues that happen on a farm, such as lameness, for example. Calvic distress is another issue: when a cow is giving birth she can get into a very complicated labor, so how do you alert the farmer that she is in distress?
But more importantly, Ida, today—and this is completely unique to us—gives recommendations to the farmer. We’re working on this whole AI based approach: so you’ve detected the problem, but how do you solve the problem? And has the problem been solved? So based on what Ida is learning, we can then recommend solutions to farmers as well. We can say to the farmer, “we’ve detected this problem, take these two-three steps,” and then we’re monitoring to see if the problem is going away. And if it does, then farmer gets an indication from the algoritm saying “yes, this problem has now gone away.” We see that work really well with our current customers. It’s a pretty unique approach to help farmers become more efficient. And all of this comes just from monitoring movement. That’s the magical part of it all.
Wallace: Rural broadband still lags behind urban broadband in a great many places. What steps do you take to work around that, and what might the rise of 5G allow you to do?
Khokhar: We’re currently operating in markets in Europe and North America where we do have access to broadband. Now having said that, in this new world where you’ve got sensors and the Internet of Things, there’s an obviously push from many companies saying, “send all your data to the cloud” because that’s good for them—the more data that’s there, the more money a lot of these companies will make, both on their network infrastructure and on the cloud side. But we process a lot of information on-farm, and we only send what we need to the cloud.
There’s also this concept of “fog computing,” where you’re trying to push computing power closer to where the action is happening, and that’s an approach that we take. The first step is to get only the data that you need. For that, we don’t need very advanced Internet connections—we can work quite happily with a 3G connection for a farm of up to a couple of hundred cows.
Obviously if you have a farm with thousands of cows, such as those that you have in the United States with four or five or ten thousand cows, obviously you will need a more serious Internet connection. And typically, those farms do already have that. If you go to very remote, small farms, then usually as long as you have cellular access, that’s fine—at least, that’s the case for us.
Now, improved broadband connectivity is obviously good for us. It’s good for everybody. You can push more services, and you can push better quality of service down to the customer. We’ve built a lot of resilience into the product, to sustain long outages. We use a lot of data compression techniques, which are really trying to work around the problem of poor broadband, but when 5G comes, I think everybody will benefit. It is going to stimulate the economy a lot more.
Wallace: Does this stop with cows, or do you have ideas about what this product could do with other animals? And while we’re on the subject, what made you choose cows to start with?
Khokhar: We’re unlikely to go into other animal breeds in the foreseeable future. But what we are looking at is, like I said earlier, to make dairy farming and farming in general more efficient. Apart from the cow on a dairy farm, you also have feed, you have land, and you have lots of mechanical operations on a farm, like sorting gates. So if you want to run an efficient farm—a highly efficient farm—then you also need to know what’s happening with your feed. Is it in a good state, is it healthy, is anything happening to it? You want to have a look at your field, where you grow grass, are you growing it at the right rate to be consumed by your cows?
We’re really looking at the whole farm, and from there we want to look at other aspects of farming to make food more efficient. Again, using the framework of sensors and machine learning.
The reason why dairy cows are really interesting is they’re one of very few farm animals that are farmed for their health, and not for their meat. You don’t want to kill a dairy cow as soon as possible, you want it to live as long as possible. With most other animals, you will grow them so that they can be converted into food. That really changes the dynamic of what you need to do, and the return on investment for the farmer. A healthy cow that lives long is in the long term more efficient, whereas if you did the same for beef cattle, the farmers are mostly interested in getting them to a certain weight, at which point they can be sent to McDonalds. That’s one of the reasons why cows are pretty unique when it comes to the farming space, and that’s one of the reasons that we chose them.
Secondly, when we looked at what dairy farmers have to work with, the tech is out of date. It’s at least 20 years behind the curve. Dairy farming is also a huge market, and it has a fairly significant ecological footprint. All of those factors combined just make a lot of sense to say this is a really interesting space to try and start from.
