00:00:00 Introduction
00:01:28 What is wrong with gut instinct?
00:04:01 The challenge of biological variability
00:06:15 Is agribusiness uniquely uncertain?
00:09:06 Long agricultural decision horizons
00:11:50 Lessons from Brazil’s 2021 drought
00:14:05 Building decisions resilient to multiple scenarios
00:18:17 Minimum order quantities and flexibility
00:19:24 Models as superpowers for human experts
00:21:58 Why traditional S&OP is not enough
00:24:53 A practical decision-making thought experiment
00:27:06 Moving beyond average-case planning
00:28:18 Conclusion
Summary
Conor Doherty interviews André Margotto (Global Supply Chain Director (Juices) at Louis Dreyfus Company), alongside Joannes Vermorel, about decision-making under uncertainty in agribusiness. They explain why biological variability, long production cycles, thin margins, and volatile markets make intuition and single-number forecasts insufficient. Through examples involving droughts, factory capacity, packaging contracts, and supplier portfolios, they show how probabilistic models can complement human expertise. The discussion argues for resilient decisions across multiple scenarios, better risk sharing, and moving beyond average-case planning.
Full Transcript
Conor Doherty: Welcome back to Lokad. Today, Joannes and I will be talking to André Margotto. He’s a Global Supply Chain Director (Juices) at Louis Dreyfus Company and he has over 15 years of experience in agricultural supply chain. Today he will be talking to us about decision-making under uncertainty in agricultural supply chains, a paper he and I wrote together.
He will also explain to Joannes and me all the practical reasons why agricultural supply chains are too important to base expensive decisions on gut instinct alone. Now before we get started, follow Lokad on LinkedIn. And while you’re there, connect with Joannes, André, and me. And with that out of the way, it is my pleasure to welcome to LokadTV, my friend André.
You and I worked together recently on a paper decision-making under uncertainty in agricultural supply chains. And I think the foundation of that before we even talk about like probabilistic forecasting or anything like that is you and I had a discussion and you talked about how important it was for people to move away from just making decisions based primarily on gut instinct and embracing much more technology. So I think that’s probably a good place to start. After 15 years of experience, what is wrong with gut instinct?
André Margotto: First of all intuition is not a problem. Intuition, I truly believe, is a tool, an asset when you use it wisely. I have a very nice example. My father is a farmer, a coffee grower.
He’s a farmer of a very small plot and he grows coffee in the same place for decades. So he knows everything about that small surface. And this year, he told me that he didn’t see enough bees on the flowers. So, it triggered a concern.
And he decided to water his coffee trees a little bit more than usual because when you don’t have bees, you don’t have enough grains, you know, the number of grains. So you have to compensate a little bit by giving more water so the grains grow bigger and then the final volume is a little bit compensated. He can do that because his intuition is in fact a decision model that was trained in his head for decades and he can connect bees to flowers to grains. Of course he’s a single guy working on very small surface for decades.
In real life when we talk about a broader surface, that is not the case because the surface is bigger and the variables are much numerous and volatile. So and people still work as my father does on surfaces that are thousands of hectares and this is where I truly believe intuition is not enough. This is where we have to bring more tools to people to make decisions on top of the intuition that it still plays a role in the overall scene.
Conor Doherty: One of the big again we’re talking about experience and heuristics. One of the things and I actually do want to read from the paper that we wrote together André and again I’m quoting you. You said, “In agriculture you do not fully control the raw material. You can plan, you can invest, you can use technology, but the crop still has a biological nature.” You have 15 years of experience with this.
What is so problematic about the biological nature of what you do and why does that then mean that you sort of have to embrace new technologies, new decision-making? So what does gut instinct leave on the table when you’re dealing with biological materials?
André Margotto: In agriculture we have a very interesting structure because normally we have the volatility coming from the demand side the classical problem like the bullwhip effect and then it converts to the upstream side of the chain. In agriculture we have that happening in both sides. We have nature bringing noise from the supply side and the demand, the market bringing noise on the other side. And it happens because basically the raw material is alive.
It’s responding to its environment, weather, diseases, and we cannot fully control it. And what brings a lot of challenges is this double volatility coming from both sides converging to the middle of the chain where you have to take the real decisions how much to purchase when to purchase what to process what mix of production how do we distribute that what size of packaging and this setup in agriculture brings challenges that normally are not addressed in normal supply chain conversations. And it’s very funny because when we go to the supermarket, we take for granted that the food is there and oh my god my favorite apple is not there so I buy another apple. But why my favorite apple is not there this week.
We don’t know. And this is simply why there was a disruption between these converging volatilities that happens to collapse in the middle of the chain, you know, and sometimes you have too much, sometimes you have too little and this is what brings extra challenge when we talk about meat, milk, fruits, sugar, coffee, everything that comes from the fields.
Conor Doherty: Joannes, is there something Like obviously you’ve worked with Lokad’s been what 18 years you’ve worked with a variety of different verticals. Is there something uniquely uncertain about agriculture or is it comparable to things like aerospace etc.
Joannes Vermorel: Again there is probably not things that are unique in the sense that you will never see anything like it but I think what makes really the agribusiness very challenging is that volumes tend to be very large and margins tend to be very thin. You see you have for example pharma they do have uncertainty on the yield but what they are selling is they are selling things at potentially $100 per gram. So there are classes of things that are almost irrelevant like transport is easy. Packaging is comparatively easy.
There are plenty. So you have here in I would say in the agribusiness you end up with massive changes in prices for businesses that typically don’t have luxury margins. The margins are pretty tight. So that’s that makes I would say this sort of business very challenging because even modest variation in price can really massively you know undermine your capacity to make any profit.
André Margotto: Just connecting with real life it happens often in France you see for example milk producers going to the streets with their trucks and open the valve, saying, “I can’t sell my milk for this price.”. What really happens? They had to make decisions months or sometimes years.
Joannes Vermorel: Yeah.
André Margotto: Before they knew the outcome.
Joannes Vermorel: Exactly.
André Margotto: And the just to add an extra layer of challenge on top of what Joannes said.
Joannes Vermorel: Yeah.
André Margotto: The capital needed to run these supply chains is very high.
Joannes Vermorel: Yes.
André Margotto: To purchase cattle, to plant one hectare, to buy inputs, to set up a watering system, you know, it’s a lot of money to put and the margin is very compressed. So, anything that happens along this long way before between decision taken and outcome happening, it just destroys your margin. Sometimes you lose a lot of money. There are ways to try to compensate that but the exposure is very real and we see that in the news sometimes that people gets angry because the price the current price doesn’t pay the cost that they built for sometimes two years or even three years depending on the business you are into.
Conor Doherty: I think and it’s something we mentioned in the paper the role of time as well is a huge one. When you were talking about the sort of lock in effect that you have to deal with on a daily basis. So for example again in a previous conversation you mentioned a single glass of orange juice on the table that’s the result of a 12 to 18 month series or life cycle of decisions in sugar cane or harvesting sugar that could be six years I mean depending on how often you harvest. Agricultural decisions have a time horizon that is just enormous.
I don’t think most people fully grasp that. So, could you elaborate a bit on that, please? Yes. The time horizon is maybe the thing that makes it most difficult to manage this supply and demand connection.
There are ways to compensate that as I said and risk sharing is one of them. You can share risk along the way. With the different players in the chain using for example contracts that are indexed to weather to yield to price. So when the volatility happens everybody is not completely exposed even insurance is based on weather there are there are tools to mitigate that but the mitigation means that the problem will be shared.
It will still be there. So instead of one single part of the chain dealing with everything, you can find long-term partners to play together this long-term game. Agriculture is not about the year, it’s about five years, 10 years, 15 years and you have to make sure that your moving average is going well. If in 10 years you miss nine and win one, you are in a big problem.
But you have to try always the opposite, you know, to make sense to keep going. On that note, I mean, one of the good things about having you join us, André, is you have a huge amount of experience. And in fact, I think it was what I think it was 2018, you shared with me a paper that you wrote on integrated planning. Last week we published a paper together on a more quantitative approach to decision-making.
So, combining your experience with what we’ve written, what exactly changed over the last few years to make you rethink the way you want to make decisions.
André Margotto: I like one cliché from Eisenhower that says, “Plans are useless. Planning is essential.” Back then I was looking for the perfect plan, you know, low variation, good accuracy, and in 2021, I learned that this is not the target. In 2021, we had the worst drought ever in Brazil and we were not able to forecast the yields just because that thing never happened before. It was like a really black swan event and I even have my numbers from this year showing our forecast month by month and how wrong we were every single month.
It was like a disaster and we had the best people in the room. Not only agronomists but data scientists, statisticians, everybody. And nobody was able to know what would happen. And we focused so much in trying to anticipate the number.
We forgot what really matters. What decisions hold regardless of the outcome. And by the end of the year, we failed both in anticipating the number and taking the right decisions. For example, close one of the factories sooner or something like that there.
And it’s very hard to take these decisions because it becomes emotional because when you are losing the game, you say, “Okay, I can’t fail. I have to keep going because I have to win.” But sometimes saying stop is the best answer, but it takes discipline, it takes method, and it takes courage. And when people get emotional, it’s even harder even to have this kind of discussions.
Conor Doherty: Well, one of my favorite quotes and I actually put it in a callout section was, “In agribusiness, your decisions need to be resilient across multiple future scenarios.” Essentially, that’s what we’re talking about and we’re talking about risk. Are there any examples that would again concrete examples from your world that would illustrate that point for people listening?
André Margotto: Okay. One of the decisions that you have to take is the level of commitment that you make before you have all the variables in your in your hand. I have one example very interesting. We had a crop that was outside the normal seasonality because nature gave us this gift.
So the raw material was there but the factory was not there and we had to go to a smaller factory to process that raw material and we could make more money good money from that but it all depended on the volume that this smaller crop would provide. So what we did we made a sensitivity analysis with a break-even scenario that says, “Guys, if this small crop goes beyond this volume it’s a good idea to lease this small factory to process this raw material because it makes good money below this threshold it doesn’t make sense.” And guess what? We were below the threshold and people were still asking, “Shouldn’t we do that anyway?” Because in their hearts it was a shame because the raw material was there. So we need to do something but guys are we there to process or make financial results from that you know and it’s very interesting because the emotion says, “Guys, let’s do it anyway.” No, stop.
Are we okay to lose money doing that? If you are okay, we go forward and easy to say nobody signed the losses related to that processing we did not do because we took a decision based on a structured approach instead of our good feeling or emotion. When you close contracts for packaging, you don’t know the size of your crop. So you have to commit with the supplier for a certain volume of package to pack a crop over which there is uncertainty.
So you can go 100% fixed price, which normally is the economically obvious choice because the unit cost will be lower but you can go a part of the contract fixed price and a part of the contract variable price and the variable price of course is higher and it’s very hard to sell for a CFO or for a financial director that the best choice means a higher cost.
Joannes Vermorel: Yes.
André Margotto: Because if you commit with this low cost and the crop goes lower, what you have a lot of package staying in your warehouse for one year waiting for the next crop. So selling probability-based decisions are very hard because our brains are deterministic for evolutionary purposes. We have to hunt. We have to take fruits from the tree to survive for millennia.
And now two centuries ago, we had to change the way we think. But our biology is deterministic. And how do you sell a decision that’s based on probability? This is one of the biggest struggles that I face.
Storytelling, yes, is essential there because you have to translate these ethereal concepts into something tangible to the decision making to feel comfortable on that. And this package example is very interesting because of course you can run scenarios, you can crush data, but how do you convince a financial director that a higher cost is the wiser decision? I don’t have an answer for that a definitive answer but this is a practical challenge that we have to guide decisions under uncertainty
Joannes Vermorel: Absolutely and I have my own similar situation that Lokad faces with so many companies is the minimum order quantities it’s the purchasing team will just negotiate the lowest price per unit possible and this will come with a gigantic minimal order quantity and the reality is that it’s exactly for the same reason as your packaging example. It is very inflexible and usually a much better negotiation would be to have like a I would say a ladder of price breaks that is very well thought out. But a ladder of price breaks is indeed optically it’s going to be a higher price per unit than a hard, sky-high MOQ that represent the highest possible reasonable volume that the company can afford. It is exactly the same sort of thing and indeed it is quite difficult to have the director of purchasing to embrace something that I would say superficially looks less efficient.
André Margotto: I always like to say to my team, “Models bring superpowers to humans.” I like to use that analogy, you know, because I truly believe when you couple wisely, both worlds you take the most from it because models can rely on human expertise to run well and human expertise can be used it to challenge the models itself. And I have a nice example from that.
Conor Doherty: Please.
André Margotto: I had a very smart data scientist back in the time that learned the Markowitz algorithm portfolio optimization of investments. So how much of your money should you put on bonds in stocks, or in Bitcoin? There is a mathematical approach to optimize this boundary and take your perfect mix. So what was the wise decision for a junior data scientist?
Let’s apply that in our supplier database to show okay this supplier should run on a fixed contract a fixed price contract. This one should run on a variable price contract. And when I plot that I have the perfect curve the perfect portfolio. There is a lot of money.
The first expert that checked the model said, “Guys, nice approach, but let me tell you something. You see this data point here where you say it should be a variable price. I know this guy. I’ve been to his birthday party.
I know which soccer team he supports. I know him. He’s a fixed price guy.” On the other hand, you see the this data point here. I know this guy.
He is a doctor. He doesn’t care about his farm because he inherited it from his father. He is a variable price guy. So the expert took the model, challenged it, re-fed it, ran it again and we had a new version of the optimized portfolio but combining both worlds together.
And this is where I see it’s the beauty of that the model gave the purchaser superpowers to enhance his knowledge on an optimized portfolio that was used to drive the approach to the market. And this is a very nice example to see how everything well managed together brings extraordinary results.
Conor Doherty: André, I just want to present the position of somebody who’s listening to this and might not necessarily agree and I’ll come to you André first for the practical side of this that someone might listen to this and say, “Well, yeah, I agree with all of that, but I have my S&OP process. I have my IBP process. Like, I have systems in place. Why do I need to adopt what you’re saying?
Because I already have experts who can do all of these things. I have meetings, I have schedules, I have all of this already in place.”
André Margotto: I would ask this person a question. Are you happy with what you have? Are you using it properly? I know more use cases of S&OP and models that don’t work than the ones that work.
And everything comes down to the way you use it because a good tool in the hands of someone who cannot handle it well will not produce the outcome that you are expecting. So it’s not because the governance is there that it means that the value is there. So why should we try to pursue what we are talking about here right now to ensure that by the end of the year you are happy with the results that you got regardless of what happened along the way. It could be more, it could be less.
But did you achieve a good level of results with the decisions that you took along the way? Because sometimes you are lucky and people try to not agree with that. A very good result of the year can be just something that you did not anticipate and we were in the right place at the right time. In my perspective, this is too bad because in the next year it can be the opposite and double anticipation and decision under uncertainty is about the decisions that are good enough regardless of what happens.
And I truly believe that the usual S&OP processes and models do not take that into consideration. Normally it’s one baseline, one number, one forecast, a lot of execution around that and in the end the result is not what you expected. It can be positive or negative but still you are not at the wheel you know you were a passenger, not a pilot and it’s very dangerous very dangerous
Conor Doherty: I wanted to ask a basic thought experiment that I think takes all the ideas that have been discussed and makes it very concrete for everyone. So and I’ve written this down so I’ll just pitch it to you and then André I’ll come to you first. So imagine the following scenario, André. So we’re working in agricultural supply chain.
You must commit or maybe not significant money today. The raw material outcome remains uncertain. André, as you described, prices may move. In fact, they almost certainly will.
Processing capacity is limited. Several markets are available, but obviously they differ in margin. They differ in risk. Before you commit any resources, what are the practical questions that you and your team must first answer?
André Margotto: I keep a good practice always to run a crash-test scenario or worst-case scenario. This should be the threshold between the level that you can commit on a deeper level and the one that you should discuss because I like what Joannes said a little bit further the latter, right?
Joannes Vermorel: Yeah.
André Margotto: You don’t have to be binary, white and black. You can define shades of grey.
Joannes Vermorel: Yes.
André Margotto: And the worst-case scenario should be the threshold where you draw the first frontier. From where you can be more let’s say fixed more less flexible and from which threshold you should start discussing the tradeoffs between flexibility and efficiency you know so running threshold scenario worst-case scenario you can call whatever you want it’s a very good practice to start discussing this balance between commitment and exposure.
Conor Doherty: All right. Well, it’s been a great conversation, but I have only one closing question, and because you’re the guest, André, I’ll give it to you. Given all your experience, giving everything that you’ve learned both on the job and then theoretically, given everything you’ve written, what is one belief or one practice in agribusiness that you wish you could remove or one that you wish you could add?
André Margotto: I would love to remove one that’s very simple to explain, very hard to address to eliminate the average case planning approach. Average can kill any ambition, any dream. It’s a one-dimensional view of a multi-dimensional problem and people have to move away from averages and start talking about distributions. It’s talking about risks, talking about probability.
The reality is probabilistic. Our plans, our minds, are deterministic. How do we bridge this current state of working to the one that I truly believe is the way? I don’t have the answer.
I just have the belief that we should pursue that with models, with people, with knowledge, with discipline, with new ways to think about the same problems. You know, there are so many dimensions to address this very simple ambition. But the ambition is there. Stop averages.
Talk more about distributions.
Conor Doherty: Well, I don’t have any further questions. André, I approve. Joannes approves. There you go.
You got the thumbs up validation. There you go. Seal of approval. I don’t have any other questions.
André, it’s been a pleasure having you here. I was really glad you were able to join us. Thank you so much for all your time and for working with me on this paper. And to everyone else, if you’re interested in continuing the conversation, please reach out and connect with Joannes, André, and me on LinkedIn.
We’re happy to talk. In fact, the link to the paper that André and I wrote will be in the top comment under the video on our website and also on YouTube. And with that, I have nothing else to say. We’ll see you next time and get back to work.
