Teaching case Ecorobotix: Embracing sustainability through AI-solutions

Part I: From a vision to a prototype

The Ecorobotix story begins with its founding in 2014. The two founders, Aurélien and Steve, connected through their shared passion for sustainability. Aurélien, hailing from an entrepreneurial family, has a background in business and economics, while Steve, engineer by training, grew up on a farm and has strong memories of days clearing weeds from the family fields while his classmates enjoyed afternoons off. Their commitment to environmental protection stems from personal interests combined with the realization that many social and ecological challenges are almost impossible to solve using existing approaches. Together, they envisioned a future where sustainability and technology could work hand in hand.

The founders believed that there had to be a better way than pulling weeds by hand. Yet, they were even more dissatisfied with the common approach of spraying herbicide widely across fields. Weeds are estimated to be responsible for one third of crop losses, making their control essential. However, indiscriminated broadcast spraying not only heavily pollutes soil and the environment, but also harms the crop itself. The founders began brainstorming ideas and developing prototypes, asking themselves: What if weeds could be sprayed in a targeted manner?

In order to deal with this question in more depth, they received a loan from the Foundation for Technological Innovation which allowed them to move from initial ideas to the start-up phase. A few trials later the answer became clear: artificial intelligence (AI) could make this vision possible. The AI would provide the capability to recognize and differentiate between weeds and crops. To achieve this, AI had to be trained with many images containing human-annotated crop plants and weeds. Field images had to be taken under consistent conditions, manually labeled, and used to teach the AI how to distinguish weeds from crops effectively. This process required extensive manual oversight and evaluation. For example, the AI needed to account for the varying stages of crop growth. Self-recorded training data had to include every phase of plant development; otherwise, the AI might mistakenly identify young crops as weeds, potentially causing unintended consequences.

Moreover, the AI had to seamlessly integrate with a physical device capable of targeting and spraying weeds with precision in real-time. This combination of advanced AI and physical functionality became the foundation for Ecorobotix’s innovation, paving the way for a more sustainable and efficient approach to weed control.

The first product developed by Ecorobotix in 2016 was a prototype of a self-driving solar-powered robot designed to independently traverse entire fields and apply herbicides very specific (see Appendix B). This means that the robot would use AI to scan the field and apply herbicides in a targeted manner in real-time. Using AI, the robot could scan fields and treat crops with herbicides in real-time, providing a severe contrast to conventional methods where farmers either broad-sprayed herbicides across fields using tractors, or manually removed weeds. The founders, supported by a small team of employees, spent 4 years developing this prototype. By 2018, a dozen of these robots were produced and deployed across Europe during two seasons to perform autonomous weeding on pilot projects in a few different crops.

Throughout this process, Ecorobotix faced several challenges. First, they encountered efficiency problems because the self-driving device was only 2 m wide and could only operate very slowly due to the sequential operation of its robotic spot spraying arms. This hampered the throughput and made it less viable for large-scale use. Second, there were legal issues, with concerns about autonomous and self-driving machines in several countries. Third, the training of the AI was an enormous effort. The main issues were on the one hand to have sufficient annotated images representing the extremely varying conditions of a cultivated field, such as various kind of soils, weeds, and natural illumination conditions. On the other hand, the training of the AI had to be done manually. In addition to the hurdle of developing such an AI, a physical device with high reliability and robustness needed to be developed to carry out the treatment. Lastly, the self-driving vehicles had high investment costs, creating an imbalance between expenses and potential revenue, rendering the business model difficult for the farmer.

These limitations forced the Ecorobotix team to make a first important pivot on their product during Summer 2018, which was possible through the support of private investors and a research project supported by the Swiss Confederation. In order to increase the machine throughput, they abandoned the robotic arm for a boom with a high density of nozzles, allowing keeping maximal vehicle speed even with high weed densities. They doubled the speed of the robot and decided to use a controlled illumination system and cameras under a cover to get rid of the variations of natural light that prevented efficient AI performance. The lessons learned with the first generation of vision and AI system led to a much more robust system with very good plant recognition rates. However, even if this second generation of robots was much faster, the fundamental problems of autonomous robots remained: safety issues, high costs, complex to use. In June 2020, after more than 6 years of exhausting technical and commercial development, the adventure was on the brink of failure.

Part II: Pivot switch to a «final» product

The Ecorobotix team decided to make a second pivot: abandon the concept of an autonomous machine, keeping only the AI and spot spraying technology, and putting it on a machine behind a tractor looking like a standard agricultural implement. While the core concept of using AI to scan fields and treat weeds with ultra-precise spot spraying herbicides in real-time remained unchanged, this pivot allowed the team to leverage their existing AI model and maximize the resources already invested. This solution significantly reduced costs, improved compatibility, and addressed many of the initial inefficiencies, providing a more practical and scalable alternative. Later on, this product change proved to be a turning point. The changes did not go unnoticed by investment observers, in fact, venture capitalists gained interest:

«I had already looked at Ecorobotix years ago. At that time, there was only one product, an autonomous robot that resembled a spider, very slow and otherwise unconvincing. The CEO took on board the investor feedback and did a pivot – that impresses me, because the ability to listen is very important in a founder.» (venture capitalist in Bilanz, 2022)

In the following year, Ecorobotix focused on refining its product. This new product replaces the traditional methods of either employing a team of people to manually remove weeds or using a conventional trailer equipped with sprayers along its entire length to blanket the field with herbicide, affecting both the crops and the weeds. The new implement, with its high-precision spot sprayers and AI, specifically targets weeds while leaving a small protection zone around the crops, ensuring they remain unaffected by the chemicals.

The implement operates in two parts. The first part scans the ground using cameras, collecting real-time field images and employing AI to identify weeds to be eliminated and crops that must be protected from herbicide exposure. Its AI determines whether each pixel represents a crop, soil, or weed. The second part features a series of high-precision electrically activated nozzles arranged in a row, which are connected to the tank. These nozzles target the weeds with centimeter-level accuracy, marginally touching the crop plants, or even avoiding them entirely depending on the chosen safety distance (Figure 1).

OPEN IN VIEWER

Figure 1.

Ecorobotix’s AI-based product, identifying crops and weeds.

Farmers, the end-users of the Ecorobotix trailer, are provided with a tablet connected to the implement, allowing them to customize its operation. Through this tablet, farmers can select the type of crop being cultivated and adjust the safety distance the sprayers should maintain around the plants. They can also modify the height of the nozzles from ground and control how closely they operate near the crops. This customization ensures the device can be tailored to specific field conditions, offering flexibility and improved efficiency. Moreover, farmers can monitor the treatment on their tablet, observing in real-time what the AI has identified as weeds (Figure 2).

OPEN IN VIEWER

Figure 2.

Illustration of the tablet enabling farmers to make customizations.

During the company’s initial years, developing the product presented several challenges. When Ecorobotix abandoned its original concept of an autonomous robot, it became clear that understanding and addressing the interests of the farmers—the end customers—was essential. Farmers were not interested in a self-driving device for several reasons, including cost, complexity, and regulatory concerns. Instead, they expressed a preference for a solution that could be attached to their existing tractors. This second pivot not only aligned with the practical needs of farmers but also supported Ecorobotix’s sustainability proposition. The new trailer significantly reduces the use of herbicides by targeting weeds with precision, addressing environmental concerns while protecting the interests of the farmers. By adapting to these needs, Ecorobotix succeeded in developing a product that balances sustainability with usability and affordability, ensuring its appeal to a broader customer base.

However, the product presented several challenges that had to be addressed. First, Ecorobotix encountered a familiar issue in AI development: data collection and processing. The AI used in the device was newly developed, and gathering the necessary data proved to be a major hurdle. The team needed to capture images from various fields using the same cameras under consistent conditions, which was rarely practical. Once collected, the data had to be labeled, and the AI needed extensive training. However, data collection was just one part of the challenge; another was implementing and validating the AI in real-world field conditions. Even standardizing the AI’s performance for the same crop across different fields proved difficult. The variability of soil conditions, field locations, and weather further complicated the AI’s performance. A trained model did not guarantee consistent results across diverse conditions. As a result, Ecorobotix had to develop several prototypes, conducting numerous test series and optimizations to ensure the AI’s reliability. The integration of data, AI, and the physical product required fine-tuning to achieve a seamless interaction. The early AI prototypes, including those for the abandoned self-driving robot, were significantly less precise compared to the current product.

Part III: Shaping the business model

For each crop, Ecorobotix began with internal tests in an alpha phase. After 2 years of product development and refinement, the company produced the first pilot machines, which were tested internationally in 2020. These pilot tests allowed Ecorobotix to gather extensive data, including images of various crops, to further train its algorithms. This data was used to enhance the beta version of the algorithm, which was then provided to customers free of charge for smaller scale testing. To ensure safety and accuracy, many test runs were conducted using water instead of herbicides. This allowed Ecorobotix to verify that the safety distance around crops was sufficient, ensuring that only weeds were sprayed.

When the device including herbicides was tested in real-world conditions, results typically became apparent within 2 weeks, showing whether the device was functioning as intended. Since the sustainability aspect of Ecorobotix’s product is not a self-fulfilling prophecy, they began conducting concrete tests and measurements on fields. Their spot-spray treatments resulted in a reduction in weed killer usage of approximately 90% compared to traditional broad-spray methods.

Continuous feedback from beta-test customers played a critical role in the product’s evolution. Farmers’ experiences and insights were relayed directly to Ecorobotix, enabling crop product manager Simon and his colleagues to improve the device iteratively. This collaborative approach helped ensure the product met both technical and practical requirements, gradually refining it into a reliable and effective solution.

In the process of developing a fully operational product, the business case became increasingly clear (see Figure 3). The main customer group of Ecorobotix are farmers. They aim to perform their work more efficiently and move away from traditional methods, such as employing farm workers to remove weeds manually or spraying entire crops. Several factors drive this shift. Some herbicides will be banned over time and will no longer be permitted for use, necessitating alternative solutions. The increasing cost of herbicides places a significant financial burden on farmers, particularly in countries with lower price levels. Farmers face challenges in finding enough farm workers to perform manual labor, making automation and efficiency crucial. For some farmers, personal attitudes and values around sustainability play a role, motivating them to adopt greener practices. However, the primary factors influencing their decisions are economic reasons. To enable farmers to transition to more efficient field management, it is essential that the new solution remains affordable and accessible.

«What fascinates me is that you don’t enter the stage of the crop, but only the crop itself. The software then independently recognizes all the shapes, leaf stages, colors of the different soils, and so on. There is a remarkable amount of technology in the device, and it is constantly evolving.» (farmer and customer of Ecorobotix in Agrarheute, 2024)

OPEN IN VIEWER

Figure 3.

Ecorobotix’s value proposition embedded in a viable business case.

Among other industry players, Ecorobotix introduces a new business model in the agricultural sector, transforming a traditionally physical market by leveraging digital opportunities through its AI-based solution. In addition to the purchase price of the device, which is comparable to that of a high-end tractor model, farmers pay an annual license fee for algorithm updates, following a subscription-based model. There may also be additional one-time costs for adapting the system to new crops, depending on the farmer’s objectives.

Beyond ecological efficiency and the fulfillment of Steve and Aurélien’s initial vision, the Ecorobotix team also developed a viable business case. Their device has a lifespan of at least 10 years, and calculations based on cultivated onion crops indicate that it pays for itself within approximately 5 years on a 20 ha area. In addition to its ecological and economic advantages, harvests are often more abundant.

Government subsidies can also have a significant impact on the business model. Regulators have the potential to subsidize Ecorobotix’s product to encourage its adoption, as seen in France, where authorities covered around 40% of the device’s purchase price, thereby influencing the business case for farmers. It is not uncommon for subsidies to require such products to be ready for series production. The primary goals of regulators are to maintain oversight of technologies and herbicides in use while promoting sustainable, long-term agricultural practices.

In other European countries, subsidies for semi-autonomous precision devices are also increasing, further improving the business case for farmers and accelerating the adoption of sustainable solutions. For example, in Switzerland, subsidies of 10% for semi-autonomous devices will be available starting in 2025.

«I think that saving pesticides will become a necessity for many companies in the next few years. Small quantities are not enough to achieve the political reduction target, especially in sensitive areas.» (farmer and customer of Ecorobotix in Agrarheute, 2024)

By supporting innovations like Ecorobotix’s solution, regulators can help drive a shift toward more sustainable farming while addressing broader environmental and social objectives. In many cases, state subsidies played a crucial role in making these technologies more accessible and economically viable for farmers.

Since the purchase price of the device represents a significant investment for farmers, some opted to buy the equipment and rent it out to other farmers. While this approach does not align with the intended business model, it provides an alternative means of making the investment more financially viable for farmers.

Part IV: Hurdles of a scale-up

As Ecorobotix progressed toward developing a fully functional product capable of treating crops and ensuring that the spray targeted weeds without affecting the crops with a satisfactory degree of precision, it continued to face challenges. Ecorobotix reached a significant milestone by entering markets in multiple countries. It completed a funding round of $52 million with several venture investors in 2023. Nearly a decade after its founding, the establishment of a subsidiary in the United States marked a major step in its expansion. Today in 2025, with over 180 employees in Switzerland, the company is experiencing rapid growth. However, Ecorobotix is still grappling with the challenge of extending the applicability of its AI to other crops and contexts.

This expansion underscores the context-dependent nature of the AI-solution. For example, in the United States, vegetables are grown on ridges having different geometries than the ones found in Europe. This difference affects the product’s operation. When weeds at the top of a ridge are treated, the distance from the sprayer to the weeds is much shorter than when weeds at the bottom of the ridge are targeted. These variations became apparent during initial tests, necessitating specific adjustments. First, the detection of the weeds, whether they are located at the top of the ridge or below, had to be adjusted. Second, the device’s timing for triggering the sprayers had to be adapted to account for the varying distances, depending on the location of the weeds. Although these adjustments might seem minor, they vary significantly across regions, presenting new challenges for Ecorobotix with every expansion and scale-up. The AI is so specific that it has to be trained differently for each type of crop and cannot be applied across countries.

Ecorobotix’s expansion in multiple countries and its rapid growth introduced both opportunities and challenges. With a significantly larger workforce compared to its early days, the company is now better equipped to address error reports from sold products more efficiently. The increased personnel resources enable faster adaptation of existing algorithms to new contexts and the development of new algorithms. This improvement accelerates critical processes such as data labeling, which is essential for efficiently implementing AI-driven projects.

OPEN IN VIEWER

Figure 4.

Growth of Ecorobotix’s employees over time.

Over time, the company grew substantially in terms of sales and employees (see Figure 4). Ecorobotix developed specialized expertise in retraining their AI for different crops and handling error reports effectively. Through years of experience working with AI in this specific context—integrating it with a physical product—they built a deep, knowledge base on how to train and adapt the AI to new contexts. This accumulated expertise makes the Ecorobotix team faster and more efficient.

However, this growth also necessitated the professionalization of the organization. With a growing number of employees, the company had to develop and establish formal processes, requiring greater coordination to ensure smooth operations. While this added structure provides scalability and efficiency, it also adds layers of complexity to decision-making and daily operations.

«In the past, you could discuss it over coffee, but now you have to attend a lot more meetings and also develop theoretical concepts.» (product manager Ecorobotix, interview)

Due to significant growth, the company has expanded its workforce across various areas. As a result, employees now come from diverse professional backgrounds, not necessarily from agriculture. To ensure that all employees understand Ecorobotix’s core competencies, history, origins, and priorities, the company organized internal training days. During these sessions, all employees have the opportunity to sit in the cab of a tractor and operate the company’s product in a real field setting. This hands-on experience helps them internalize the company’s mission and values.

Challenge 1: Digital business related to sustainability

The founders of Ecorobotix prioritized sustainability as their primary goal. However, operating a digital business centered around sustainability creates tensions. On the one hand, Ecorobotix must generate profit to grow and scale its solution effectively. On the other hand, the company must remain true to its core proposition and market their product with the features that promote sustainability. Farmers, the end customers, have mixed interests regarding sustainability. While some farmers value green products and align with sustainability goals, the majority are more focused on economic benefits. As agriculture is their livelihood, these farmers prioritize solutions that save costs and increase yields. To fulfill its sustainability mission, Ecorobotix must appeal to this larger group by emphasizing the economic benefits of its product.

«The environmental aspect, which was the original idea of the business and its founders, is now achieved through the leveraging of economic efficiency.» (product manager Ecorobotix, interview)

For example, Ecorobotix’s product reduces pesticide usage, minimizing damage to crops and mitigating the harmful effects of excessive pesticide application. Healthier plants can lead to improved yields in subsequent years, aligning both economic and environmental interests.

«In most cases, you can see two weeks after application that our part of the field is growing better than the untreated part.» (product manager Ecorobotix, interview)

Country-specific regulators also play a critical role in this dynamic. In some countries, subsidies are available for Ecorobotix’s product, encouraging its adoption and reinforcing sustainability goals. However, in other countries, subsidies are unavailable, often because the product is the only one of its kind, and regulators avoid favoring a single company. Furthermore, regulatory frameworks and policies vary widely across regions and are subject to frequent changes, requiring Ecorobotix to continuously monitor and adapt to these shifting conditions.

Challenge 2: Scaling an AI-solution

One of the big challenges for Ecorobotix is that they do not have a static, finished product for sustainability that they can simply manufacture and sell. Because the AI is highly context-dependent and specific, it cannot be easily transferred to a new crop or environment. New images must be collected, and the AI retrained for each new crop, a process that still involves manual steps such as labeling images and monitoring the AI’s learning progress.

However, this challenge extends beyond new crops, even existing algorithms for certain crops also sometimes need to be readjusted when scaling up to a new country. Soil conditions, cultivation methods, and herbicide use can vary significantly between regions, and these factors all influence the product’s effectiveness. Tests and controls are essential with every expansion to ensure the device does not mistakenly spray the crop instead of the weeds.

A key tension arises in the development of the algorithm. On the one hand, more devices are needed to improve the algorithm’s performance through real-world data collection. On the other hand, devices can only be sold if the algorithm is already functional and reliable.

«We have to be good enough with the first beta version of the algorithm to satisfy certain customers who will then buy the device. This is how you efficiently advance to the stage of a commercial algorithm that is really good.» (product manager Ecorobotix, interview)

Ecorobotix is constantly faced with the reality that they can never be 100% certain of perfect quality. Even if the device has been successfully used to treat a crop for one season, a small update can cause certain weeds to be misidentified or treated incorrectly. Consequently, the device’s functions and algorithms must undergo continuous testing, even for what is considered the final product.

Challenge 3: Usage of the product

Ecorobotix makes efforts to train its customers as effectively as possible. Before making a sale, the company allows farmers to test the device in their fields, often with a water-filled tank to simulate operation. This enables farmers to see exactly what the device would have sprayed. Following this, Ecorobotix provides training and detailed explanations about how to use the device effectively.

Farmers value the ability to customize the device’s settings to meet their specific needs. For example, using the tablet interface, they can adjust the height of the sprayers or change the safety distance the sprayers should maintain around the crop.

«We quickly realized that you can’t just try it out. A lot of expertise is needed and the driver has to know what he’s doing.» (farmer and customer of Ecorobotix in Agrarheute, 2024)

However, this flexibility introduces another area of tension. Farmers are primarily interested in maximizing their yields and may customize the settings to achieve the highest possible output. Unfortunately, these adjustments may not always align with Ecorobotix’s sustainability goals. Tensions arise for Ecorobotix because the company has no control over how farmers ultimately use the device. This lack of oversight means that some users might prioritize economic gains over sustainability, potentially counteracting Ecorobotix’s proposition.

«For edamame, for example, there was no algorithm yet. So we took a chance and used the program for soybeans [for edamame]. Here, 30% of the reporting plants were not recognized.» (farmer and customer of Ecorobotix in Agrarheute, 2024)

Case questions

- How and why did Ecorobotix adapt its AI-based solution over time? Please identify the stages of development and the reasons for adaptation.