SFTW Interview: AI for legacy companies with Feroz Sheikh
Data is the new tractor, data is the new microscope, data is the new shipping container
Welcome to another edition of SFTW Interviews. These interviews are normally available to paid subscribers only, but the current edition is being made available to everyone. I will be publishing a couple of SFTW interviews every month. With the holiday season coming up, you are getting two SFTW interviews within 7 days.
The current edition of SFTW Interviews is my conversation with Feroz Sheikh, Group Chief Information and Digital Officer (CIO and CDO) at Syngenta Group.
Programming Note: The next two interviews include a senior representative from an innovative company in Brazil, and a senior US representative from a major OEM. If you would like to read future SFTW interviews, you can subscribe to the newsletter. There is a 15% discount, if you sign up for an annual plan before the end of the year.
Feroz is one of the most thoughtful technology leaders in the agriculture industry. He has been at the forefront of digital, and AI initiatives at the Syngenta Group. Feroz has been with the Syngenta Group for more than six years. Prior to joining Syngenta, Feroz has been an entrepreneur in the education space.
My conversation with Feroz focused on AI initiatives at the Syngenta Group, the role of AI within legacy industries like agriculture, capabilities needed to deploy value creating AI solutions, organizational mindset and the role of leadership, Syngenta’s AI manifesto, their engagement with the ecosystem, and democratization through AI capabilities.
Finally, we talk about some predictions on how AI will change farming practices and farming (Feroz has some really interesting thoughts on it).
Now onto the lightly edited transcript of our conversation.
Rhishi Pethe (RP): Hey, Feroz, thanks for joining. And today I want to specifically focus on some of the AI work that Syngenta has been doing. One of the things I'm noticing is that legacy industries have a unique challenge, and I want to get your perspective on incorporating technologies like AI.
Do you agree with the premise that it's a different challenge for legacy industries given something like AI? And if yes, what are those challenges? How is Syngenta and how are you thinking about them?
Feroz Sheikh (FS): For sure. Legacy organizations that have been around for a long time. They have been operating in a certain way and digital transformation has been a buzzword for quite some time. So companies are in various phases of their digital maturity. And you can almost say that data and digitalization is the necessary prerequisite for how well a legacy organization is able to ride the AI wave.
And those organizations that are leading, that have been leading in the digital transformation that were able to tighten their socks in the last few years would be a little better suited to implement AI for the organization as well as for their customers.
So both in the R &D or for their own usage, as well as the products that they bring to the customers. And the ones who probably were caught sleeping will have a much harder cliff to climb.
Thankfully at Syngenta, we've been on that journey for a few years. Our data is in order. Our R &D teams have been using AI for quite some time as part of an active data-driven discovery process, as well as, with our digital tools and the growers as well.
So from that point of view, we found ourselves in a position where given the data and given the coverage of our digital tools, we were in a leading position to be able to build AI and bring that value to the customers, to the farmers.
And we've done that with the launch of Cropwise AI, both for smallholder markets as well as big, large markets like the US or Brazil. We have dipped our feet. It's still early days to find out which way this will go, the adoption, the success, and the learnings along the way. But the initial feedback has been very positive so far.
Future of Farming Imagined (Artwork by Eeshan Inamdar)
RP: Yeah, that makes sense. AI is not new. It's been around for some time. But you mentioned data, you mentioned other infrastructure issues. What is the portfolio of capabilities somebody should have to be in a better position to do this?
FS: The first and foremost, I would say is probably a bit of a mindset that the organization is willing to experiment with this new technology, the new way of doing business and a new way of interacting with their customers and bringing innovation to the customer.
And if we don't have the right mindset, we can either run into issues and struggle with bringing this kind of a change or not. In fact, in a way, it's a standard change management problem. It's no different when it comes to AI. The second thing, of course, is data. So when you have the data, it is at the end of the day, the data that feeds the AI.
And if we have data from our research or from the customers and so on. That data is what makes the AI smarter to train and to offer the services and value to the customers. It's been a journey.
And companies who had the data locked up in various legacy systems, there have been efforts to bring it into a data lake and things like that. So you gradually climb the data maturity ladder. And once you have the data in a place where it's clean, consistent, ready to use, the shift to go from statistical predictions to feeding an AI model is relatively simpler and easier.
And if we don't have it, if the data is scattered everywhere, if the data is either noisy or not clean then it takes a whole lot of effort just to first make it ready before you can even start to feed it to the AI. So that I would say is the second ingredient. And the third is the algorithm or the service to have clarity on exactly what we are going to offer and how.
If we don't have those kinds of digital tools, then the AI at best can help you to make better decisions, but then, when we need to offer those as a service to our customers, we need a vehicle, we need a digital tool through which we would offer that service to the customer.
And those companies who had that kind of a medium or a channel would find it easy to add AI on top of the existing services that they have. And others will probably start from scratch and they will struggle with customer acquisition and scale up for growth.
There are various ways in which we look at it, but these days the internal belief that we have is an equation of 10, 20, 70.
So we say it's 10 % data, 20 % the algorithm and the digital tools, and 70 % about the organization and the change management problem.
And our readiness to be able to drive this kind of an impact. So the 10, 20, 70 is the formula that we talk of.
RP: Wow, that is surprising. Everybody talks about data, how data has gravity. Models will move to where the data is. But you're saying that is only 10% of the equation. Has your view on that changed over the last two years? Or you've been consistent with this view, the 10, 20, 70?
FS: I mean it has changed to the extent that I realized the increasing importance of the organization mindset. So, what you just said, a few years ago, I would have said the same thing, it's all about the data. And if we don't have the data, and the data is not clean and in order, we will not be able to build anything and take it out. Working through the transformation at Syngenta, what we realized is that the role that the larger part of the organization plays is equally important. Once you have the data and the algorithm, if the organization is on board with that change, then you can bring out that innovation and services to the customers. And if the organization is still stuck in a bit of legacy, then no matter what kind of data or tools we provide to them, they're not going to use it.
It's kind of like saying that you have a calculator and there are those who are willing to use the calculator and there are those who probably would have hesitated initially and still do pen and paper.
RP: Just because everybody has access to Excel, that doesn't mean everybody is a financial analyst, right?
FS: An even better, more relevant example is probably the self-driving capability in today's cars. Initially I was very hesitant to put my car on autopilot and I always felt that I could drive it. I feel safer when I'm in control and over time I learned to give control to the autopilot so it's basically that it's a change in mindset. The fact that Autopilot is available in the car doesn't automatically mean anything different.
RP: Rightly or wrongly, this is the flak that a lot of the legacy companies take, that you don't change fast enough. You're stuck in your old ways. So what are you and your teams and what is Syngenta doing to move this big ship? How do you get them to think differently?
FS: This is where I would say that, in my role at Syngenta, I've been positively surprised by the organization's willingness to change, experiment and try new things. Now, sure, there is a part of the portfolio where we are very risk averse.
Because, you don't want to provide a negative recommendation to the farmer, which will, let's say, impact the crop, for instance. So there is a balance between where we are willing to take risks and where we are willing to experiment as an organization.
One of the things which I have consistently been doing over a few years is to find like-minded leaders and work with them to create some success stories. And those success stories then bring in more people and more use cases on top of the technology.
The other positive thing which has happened recently is that we have a new CEO. Jeff Rowe took over in the role of CEO from the start of this year. He is, I would say, luckily a very tech savvy CEO. He's a fifth generation farmer himself. He has thousands of hectares of farmland outside of Chicago, but he is also a very passionate techie.
Internally, he coined this term for us, that data is the new tractor and data is the new microscope and data is the new shipping container.
So when your CEO makes this kind of statement, then you know that the organization is ready and willing to go down this path.
So when he spoke about these, he was referring to the use of data and AI in the field.
Data is the new tractor, informing new decisions and better precision. Data is the new microscope inside of our R &D labs, helping with discovery. Then data is the new shipping container, essentially looking at all of our manufacturing, supply chain and logistics.
Again, being informed by data and using AI to do something as simple as optimize the routes or predict the inventory demand and so on.
RP: It doesn't sound very different from any other big technology shift that happens. Find early successes, get buy-in from the top leadership, find like-minded people. You can look at past playbooks and say, how did we do this?
When you incorporate AI within Cropwise, you want to do better product placement, recommendation, and you talk about more informed decisions. What are the metrics that you're looking for to see if it is working?
FS: First and foremost, our focus currently is on the feedback loop. So when we give a recommendation to the user, we also seek active feedback. Did we make the right recommendation? In some cases, the recommendation we made is accepted and implemented. In some cases, the user says, you gotta be joking, right? And they do something else. So that tells you how many times you're right, how many times you're incorrect.
Our current stance when it comes to GenAI is that it is still early days of that technology. And therefore, positioning in the market is to offer this to the advisor, allow the advisor to have the superpower to offer that as a service to the end-grower.
Instead of giving it to the grower directly, the advisor is able to provide that human oversight on the recommendation and also give us the feedback. So that feedback is one important element. And then the engagement, how many times they are using the system, how many times they are getting the answer that they were looking for. And how often do we end up replying that, I don't understand, can you tell me more, with the question, for example.
RP: What has been the feedback from the advisors? Because one trope is that AI will take away all our jobs and kill all of us in the next couple of years. And yes, it is going to create some shifts in the labor market. Do they see this as a threat, as a tool?
FS: And I'm 100% aligned with you that this is not a case where AI will replace the advisor. In fact, our positioning has been that this is about giving better tools to the advisor so that they can make better recommendations.
I like to take an example from the work which I was doing before this in education. So when you have a teacher teaching a class of students, it's about the teacher to student ratio. So you have one teacher and a class of 20 or 50 students. And the work we were doing before was about using data and AI to create digital learning tools so that kids can learn better back in India. And we used to have the same question all the time: Will AI replace the teacher? No.
It's all about disrupting that student teacher ratio. So, the same teacher, when they have the data and they can get inputs on the weaknesses of every child, the learning gaps they have, they can prescribe what content children with this kind of learning app should have. And, finally focus on the ones which really need the personal attention of the teacher. You're disrupting the ratio.
And you can change the ratio by an order of magnitude. So the same teacher can now address a class of 500 students. Bringing that analogy back into agriculture, it is simply about the same equation that the agronomist or advisor will now perhaps be able to advise more growers, more hectares, and give them better and more personalized advice when they have the power of data and AI in their hands. So it's really about disrupting that ratio and making the advisor more effective rather than thinking that the AI will replace the advisor.
RP: Talking about change management again, you guys released an AI manifesto. It sounds great in practice, butI want to get your perspective on one or two of these. What are the trade-offs you're trying to make? What do you mean by fairness? And what's a real example where you are saying, we need to do something a little bit differently because we want to be fair.
FS: Yeah, the manifesto is something which is very, very close to our heart as an organization, as a leadership team here and me personally, because we were very clear from the beginning that when it comes to technology like this, we want to do it the right way. There will be several cases where we will be faced with choices.
And we want to make the right choices when it comes to the ethical considerations or taking accountability for the AI and so on.
Fairness, one very simple example is let's say we are running a marketing campaign, we go to some content creators and they create a video for us. Today AI is capable of translating that video into many different languages and we start using it for that.
So the question of fairness comes back to this: do we pay the content creator for one campaign or do we pay for 10 campaigns where we use the content and translate it by AI? The right answer is perhaps somewhere in the middle so that it creates a win-win. The company is not spending money and resources on 10 times content creation.
And the content creator is probably making more than just the one content that they created. So as long as we are transparent about what we are we using it for? How many languages are we going to launch? And what is the agreed remuneration for that content creation? That's an example of being fair about it, being aware that there is an implication or dilemma like this that exists, and then taking the right decision in the interest of both parties.
RP: You talked about how Syngenta has access to millions of hectares of data. Do those situations create information asymmetries and how do you deal with those? Or you do a lot of business with smallholder farmers. There's a concern that technology like AI will widen the gap between people who have access to it and people who don't. I know that's a little bit different from fairness, but what do you think about that?
FS: No, but that's an important part of the question. The first part where we talk about Syngenta having access to the data, I would correct that statement that it's not that we have access to the data. We are not looking at that data. The data ownership is still with the grower and whatever data that they upload in the systems is used to offer services and advice to them.
Coming back from the manifesto point of view, it's one of the points which is important for us to keep reiterating and reminding ourselves that just because a grower uses our tool doesn't automatically give us access to their data.
Unless the grower gives the consent, they anonymize the data, then we use it for training the algorithm to give a better recommendation back to them. So as long as we are very clear on what data we are allowed to see and maintain that the ownership still belongs to the grower.
It is at the end of the day, all about using that to offer better services to the grower. And think about it from this point of view that the grower uses Cropwise and they upload their data for that reason, that it's my data and the system will use this to make a personalized recommendation to me. But we've actually drawn a line in the sand internally that this data is not something which we would use for any other purpose other than to give that service back to the grower.
Then to your other part of the question about the democratization of access to AI and technology. It's an important consideration and it is a risk that the developing countries run. Our stand again is that what we will see is adoption based on let's say micro entrepreneurs.
So when we launched drone based spray applications in India, for example, it is very clear that the smallholder farmers will not be the ones who will use that technology themselves. But what we did was we created an ecosystem around the grower, of people who are willing to invest in that kind of technology and who can operate it on behalf of the group.
The other thing which happens with GenAI is that it's also very capable of the whole conversational aspect of it in the local language. I think that is actually fundamental to democratization. It could be a very sophisticated algorithm behind the scenes. It could be a whole lot of data.
But if I can explain it in simple terms to the grower in their native language to a grower in Maharashtra in the Marathi language, right? That this is what it means for you. This is what you need to do. Then you're essentially taking away the complexity of that interpretation from the grower. They are no longer looking at fancy charts and fancy heat maps and trying to figure out what it means for my farm.
The system is telling them in simple terms in their local language that you need to perhaps spray your application because there is a risk of disease, right? So I think in fact, GenAI creates a better democratization for smallholder markets than pre-GenAI days. So it makes the entire power of everything that we have before now more accessible.
RP: So going back to the capabilities, the 10, 20, 70 ratio. What do you think about the build versus partner versus buy decision? What are the types of capabilities where you would look outside and why? What role can startups play? How can they engage with somebody like a Syngenta when it comes to AI capabilities?
FS: I think first and foremost, we always say that we are open to collaboration across the entire spectrum of technology domain, not just in AI. So if there are startups who have an interesting idea or interesting innovation in technology, we are more than willing to engage and learn more and work together.
No matter how big we are, recognize that agriculture is a complex domain and the sheer size of the problem across the planet, different languages, different types of growers, different crops, different agro-economic zones. We cannot solve all of it ourselves. So we are very open to collaboration.
Specifically, when it comes to the space of GenAI, I think it's a new and emerging skill. We have the data science team and the computational agronomy team, and they've been working in this space for a while. But when we think about scaling these solutions, it's a new and emerging skill. And both the engineers who are working in this space from a data science, neural network, and deep learning point of view are welcome to collaborate with us.
And then there is also a dimension of how you deploy AI assisted technology and offers in the market. And again, the analogy that I use here is not everybody needs to know how to build a car, but you do need to know how to drive the car. So that's the part where we want to make sure we up-skill our team in learning how to drive this AI car, how far this will go, what fuel to put in and how do you plan your route, for example.
And in that journey, there are several kinds of partnerships along the way. But when it comes to the actual building of the car, building of the neural network, and so on, there is obviously the build versus buy question of partnering with people, startups and technology organizations, as well as continuing to invest in our own data science capability to build some of the competitive advantages, if you will.
RP: And last question, what are some of the things in which you think farming is going to change in a 5 or 10 year horizon? Most probably we'll get it wrong. But what is your current perspective?
FS: Yeah, none of us have the crystal ball and one of our presidents used to say that it will change slower than what the techies think and it will change faster than what the legacy thinks will happen.
But if I imagine a future I think we will start to see some new agronomy protocols emerge. And one such example is the work we do with Inner Plant and John Deere, where we are incubating the genetics in the plant that can show when the plant is under stress. You have the treatment for that stress, fungicide, and then you have the see and spray cameras and machinery that can apply on that.
In 5 to 10 years from now, it will change the protocols that farmers follow. All of this involves computer vision, genetic engineering, and the chemistry incubated along with it.
But how well the plant, the machine, and the chemistry operate with each other will change.
So I think we will start to see these novel agronomy practices.
One of the other things I think is going to happen is AI will start to make the machines more autonomous. So we will start to see a little more of autonomy at precision. And when you take away the human factor, when the machines are operating autonomously, then you can imagine a sci-fi world of a swarm of robots operating in the field 24 x 7.
Then you don't need to have a very heavy machine that goes out in the field once every four weeks and has to spray a broadcast at that time. The autonomous machines can operate in the field and take an action instantly when they see a thing. So I think some of those can possibly be the future where we'll see novel practices, novel agronomy come into practice enabled through technology.
Some of it is already starting to happen in some pockets and some of it will take some time. I think the question is how quickly will these technologies scale? So you have the machinery or you have the genetics, but how long will it take for it to penetrate in the market? And that's where I think all of us can probably go wrong in our predictions.
RP: It almost sounds like an Uber moment. You had mobile, you had GPS, you had payments, you had people being comfortable with trusting your app to do something. I think that's what you're talking about, a convergence of different factors, which can create new things.
FS: And this convergence is then coupled with the willingness of the farmers to adopt. And that's where we will see a ramp up curve that can possibly look like a slow ramp up or a hockey stick. One of the realizations that I had recently, and that's my thesis, is that if you think about some of the biggest digital players in agriculture, John Deere, FieldView and CropWise. Between these three put together, we almost have one billion acres digitally connected today.
So it's no longer about more technology. It's really about this ecosystem of various factors coming together in creating grounds for adoption and impact that the farmers see. And that would also be my call to action for innovators, policymakers, that we all need to create the grounds for farmers to adopt and change the agronomy practices, sustainable practices, and so on. We need to create those incentives. We are reaching a point where it's no longer a technology problem. The technology is there.
RP: Thank you, Feroz!
Key Takeaways from this SFTW Interview
Legacy industries face unique challenges in AI adoption.
Data and digitalization are prerequisites for AI success.
A supportive mindset is crucial for implementing AI.
The 10-20-70 formula highlights the importance of change management.
AI enhances the capabilities of agricultural advisors.
Feedback loops are essential for AI effectiveness.
Fairness in AI involves ethical considerations in content creation.
Data ownership remains with the grower, ensuring privacy.
AI democratizes access to technology for smallholder farmers.
The future of agriculture will see more autonomous technologies.
Relevant SFTW editions from the archives
Data has gravity (Oct 6, 2024)
Are we thinking about AI the right way? (Sept 30, 2024)
The Fusion Frontier (May 5, 2024)
Laggards and Overshooters (Feb 11, 2024)
CCAs or LLMs? (Apr 30, 2023)
Infrastructure, Platforms, and Knowledge Graphs (Sep 11, 2022)