What we learned about AI adoption in agriculture, and why we’re putting those lessons to work across Brazil’s grain value chain.
A few months ago, I wrote about what I called the “pilot gap”: the distance between testing a promising new technology and actually putting it to work inside an agribusiness.
My conclusion at the time was that agriculture probably didn’t need more AI pilots, better-prepared ones.
Since then, our team at The Yield Lab Latam has been speaking with companies across the agrifood sector as we developed our first AI Readiness Program. We wanted to better understand what was stopping companies from moving from interest in AI to adoption, and what could be done differently before a pilot even began.
We expected to hear a lot about technology and data, and we did. But some of the more interesting obstacles had little to do with either. Companies struggled with deciding where to start. Pilots lacked clear measures of success. Projects lost momentum when people changed roles. Internal processes weren’t always ready for the technology being tested. And, quite often, when we started a conversation about AI, the immediate response was to bring in the technology or IT team.
AI is obviously a technology, but that doesn’t necessarily make it a technology department problem. Many of the decisions that determine whether it creates value are really management decisions that need to be made much earlier.
In the Weeds
Before discussing models, data architecture or startups, someone needs to ask what the company is actually trying to improve. Where is it losing time or money? Which decisions could be made better or faster? Where could a relatively small improvement have a meaningful financial or operational impact?
One of the best challenges we heard during our conversations barely mentioned technology. The company essentially knew the result it wanted: it had a process that took a certain amount of time and wanted to achieve the same outcome in a fraction of it.
That gave us something more useful to work with than simply asking where AI could be applied.
Management needs to look across the business before getting into the technological weeds. Not every problem needs AI, and some processes may not yet be ready for it. In other cases, a relatively simple application could already deliver value today. Working out where those opportunities sit requires a good understanding of the business before getting into the technology.
This is made more difficult by how quickly AI itself is moving. A use case that looked unrealistic a year ago may be relatively straightforward today, while something we dismiss now could become viable next year. A rigid three-year AI roadmap could become outdated surprisingly quickly.
Companies need a clear idea of what they want to improve, but an open mind about how they get there.
Pilot Purgatory
Some of the reasons projects stalled during our conversations were surprisingly mundane. A pilot could have a champion, gain approval and demonstrate that the technology worked, only to lose momentum when the person responsible moved to another role. Another could remain alive indefinitely because nobody had established a clear measure for deciding whether it had succeeded or failed.
One company had a particularly good name for these lingering projects: monstrinhos, little monsters that nobody quite knows what to do with.
Other projects became bogged down in contracts, security requirements or integration issues that could have been anticipated earlier. Sometimes technology was introduced before the underlying business process was mature enough, effectively turning an expensive technology pilot into a process-diagnosis exercise.
In many of these cases, the technology itself had worked. The project stalled somewhere else, whether because of ownership, internal alignment, integration or simply because nobody had agreed beforehand what success should look like.
And this wasn’t limited to companies at the beginning of their technology journey. Some of the companies we spoke with already had substantial internal data, analytics and machine-learning capabilities.
A company can have excellent data scientists, sophisticated systems and years of proprietary data and still struggle with a much more basic question: with so many possible applications of AI, where should we focus first?
The opposite is also true. Companies don’t necessarily need perfect data or a large AI team before they begin. They need to understand the business problem well enough to work out what data and capabilities will actually be required. Being technologically advanced and being ready to put AI to work aren’t necessarily the same thing.
Over the Fence
The best solution may not exist inside the company.
AI is developing too quickly for even large technology teams to follow every new model, startup and application. In many cases, somebody outside the company may already have spent several years solving exactly the problem you have just identified.
This is why I think AI is particularly well suited to open innovation. Internal technology teams remain essential, but expecting them to build everything themselves makes less and less sense when the range of external solutions is expanding so quickly.
At the same time, we frequently heard some version of: “Send me the list of startups.”
The temptation is understandable, but a long list of AI startups isn’t much use if you don’t know what you want them to solve. Without a clearly defined problem, it can quickly become a technology shopping exercise, where an impressive solution gets matched to a problem because it happens to be available.
Before looking outside, the company needs to be able to explain what it wants to improve in terms that someone outside the organization can understand. How does the process work today? What isn’t working well enough? What would a materially better result look like? Once those questions have answers, scouting becomes much more useful.
Sometimes it will lead to a startup. Sometimes the answer may already exist with an established provider or inside the company itself. And sometimes, after looking at the problem properly, AI simply won’t be the right answer for now.
That is the thinking we are now putting into practice. We start by identifying and prioritizing real business challenges, turn the most promising into something measurable, and only then look globally for solutions that could fit. The end point is a pilot designed with clear technical, operational and business criteria, including the possibility that the right decision is not to proceed.
For our first AI Readiness Program, we are starting with Brazil’s grain value chain, with my old haunt CME Group as Founding Partner of the program.
Grains are a natural place to start. The value chain stretches from inputs and production through origination, logistics, finance, risk management and trading, creating plenty of problems worth solving and enormous amounts of data to work with.
It is also where much of the AgTech ecosystem has historically concentrated its attention, giving us a deeper pool of existing solutions and startups to explore rather than assuming everything needs to be built from scratch.
Over the coming months, we will put the approach into practice with a small group of companies from across the grain value chain. The aim isn’t to apply AI everywhere. It is to get better at identifying where it can make a meaningful difference today, while keeping an eye on what is coming next.
At the speed AI is developing, none of us knows exactly what will be possible a year from now. Companies don’t need to predict every application. They need to understand their business well enough to know where to start, be willing to look beyond their own walls for answers, and remain open to changing course as new possibilities emerge.
If you’re interested in learning more about the AI Readiness Program or participating in the first program focused on Brazil’s grain value chain, please get in touch.
Thanks for reading.
KFG
Kieran Finbar Gartlan is an Irish native with more than 30 years’ experience living and working in Brazil. He is Managing Partner at The Yield Lab Latam, a leading venture capital firm investing in AgriFood and Climate Tech startups in Latin America.