Taking Flight: A Case Study on Designing Ag Drone Services for Smallholders in Emerging Markets
Introduction
The field work forming the basis of this case study took place in 2019, and it was commissioned by a social business that creates and delivers affordable, income-boosting products and services for smallholder farmers in Myanmar. They have been serving rural families in Myanmar since 2004, across their platform of farm technology, agronomy services and farm finance.
Myanmar is ranked third on the Global Climate Risk Index, making it prone to erratic weather patterns and rising temperatures which increasingly pose new threats to farm livelihoods. As farming becomes more complex and risky with climate change shocks and new pest outbreaks, Myanmar’s 30+ million farmers have few places to turn to for accurate cropping knowledge.
With these challenges in mind, a hybrid team of inhouse designers and Gyula Simonyi as external consultant conducted research and design work in 2018-19 aimed at improving the resiliency of Myanmar’s smallholder farmers. Included in this work was an exploration of the potential of offering agricultural drone services. Although in the end the decision was not to move forward with launching any drone services in Myanmar, we wanted to offer an inside look into our human-centered design process and share our experiences and key learnings from the project. With this case study we hope to inspire, inform and connect with all those who have an interest in agricultural services for smallholders.
“However, we were surprisingly unable to find any counterparts around the globe who were beyond the proof of concept stage with smallholder drone services. We realized that our work in Myanmar would be relatively uncharted territory.”
“Our team’s recommendation was not to pursue drone spraying further, with the gap to viability proven too wide.”
“Perhaps most importantly, we hope that by sharing our experiences through this case study we can help and connect with other companies and organizations supporting agricultural drone initiatives for smallholder farmers. We encourage anyone who is interested in learning more to reach out to us, and we look forward to continuing the conversation.”
Background Research
In early 2018, the team led a foundational research project to better understand farmers’ experiences with pest, disease and soil problems. The ultimate goal of the project was to understand customer needs and identify opportunity areas for offering new, impactful products and services. Members from our project team conducted nearly 100 in-depth interviews with smallholder farmers across 4 townships in Myanmar, exploring topics such as detection of pest and disease problems, relationships with local input shops, and experiences using pesticides and other agrochemicals.
There were many useful findings from this research, but some of the most interesting included:
Farmers rely heavily on input shops to help them solve pest and disease problems, but the advice they receive is often ineffective, inappropriate or involves excessive chemical use;
Most farmers wear little or no protective equipment when spraying pesticides and other harmful chemicals; and
The frequency and quality of crop monitoring vary widely among farmers.
Ideation
With the core research complete and the key findings documented, our team was ready to move on to the ideation stage of the design process. We began by developing a list of problem statements that reflected the core customer needs that we uncovered during our research. As an example, one of the problem statements was, “There is a need for more accurate and accessible diagnoses and treatment recommendations.” We then used these problem statements to ‘ideate’ or brainstorm solutions to the various problems. Some of the most important rules of brainstorming include ‘quantity over quality’ and ‘no bad ideas’, so the initial rounds of brainstorming resulted in a huge number and range of ideas. We clustered similar ideas together, prioritized solutions with the most potential, and then further developed the most promising ideas.
Two of the solutions that emerged from this process involved agricultural drones. First, we hypothesized that a drone spraying service could offer farmers a safe, efficient and effective alternative to manual spraying of pesticides and other agrochemicals. And second, we wanted to explore how we could use aerial imaging drones to offer valuable crop health information and high-quality advice to farmers.
After some light prototyping with farmers to stress test the potential of these agricultural drone solutions, we assembled a core design team of one local and two international designers. Our brief was to conduct technology and service model prototyping trials with agricultural drones in order to design an impactful, scalable, and financially sustainable service for smallholder paddy farmers.
Preparations
Our team’s first step was to conduct desk research to better understand the technology ecosystem, regulatory environment and relevant drone projects ongoing in Myanmar and other developing countries.
On the technology side, we mapped out the key features, limitations and prices of a wide range of agricultural spraying drones, imaging drones, multispectral sensors and software systems. The goal was to find the best-fit technology for our purposes in Myanmar and the best value for money, while also taking care not to put ourselves at odds with any regulatory bodies in Myanmar. For the drone spraying, we decided to purchase a locally built agri-sprayer drone similar in size and quality to the DJI Agras MG-1. On the imaging side, we decided on a DJI Phantom 4 Pro drone equipped with a Sentera NDVI Single Sensor.
We also held calls with universities, companies and organizations working on similar agricultural drone projects for smallholders, in the hopes of learning from their experiences. However, we were surprisingly unable to find any counterparts around the globe who were beyond the proof of concept stage with smallholder drone services. We realized that our work in Myanmar would be relatively uncharted territory.
With the decisions made on technology purchases, we began to make preparations for the field trials. One fundamental prerequisite was that we needed people who could actually fly the drones. There were essentially zero professional drone pilots in the entire country (partially due to the Myanmar government’s cool and ambiguous stance on drones), so this meant hiring and training staff from scratch. We recruited and hired three new staff (two external and one internal agronomist) to join our team as drone pilots, and sent them to two weeks of flight school hosted by a local aerospace engineering university. We also scouted locations and made a rough schedule of activities for the field trials.
Technology Prototyping
The 3 weeks of technology prototyping field trials began in May 2019 in Madaya township of Myanmar’s central Dry Zone. The objective of this phase was to test both the drone spraying and drone imaging technologies with smallholder paddy farmers and generate insights that would inform the next phase of service model prototyping.
With our final team of 8 designers, researchers, drone pilots and agronomists, we kicked off the field trials by collecting some crucial data from the villages where we would be operating. We met with farmers to conduct a special type of interview that we had designed, called ‘year on your farm’ sessions. During each such YOYF session we collected 1) basic demographic information of the farmer (name, age, contact information), 2) plot-specific information (the locations and geographic outlines of each unique plot), 3) detailed crop cycle information (weekly farm events throughout the year), and 4) the farmer’s interest level in participating in our drone spraying or imaging tests.
This data served several key purposes for our project. First, it gave us leads to farmers who were interested in having us test our drone spraying or imaging activities on their plots. Second, it let us know how to reach those farmers and where their plots were located. And finally, it provided a rich dataset that allowed us to better understand the nuanced seasonality of farm activities and the potential market demand of various ‘drone-assistable’ events.
At last, our team was ready for our first test flights on farmers’ plots. As mentioned earlier, our drone pilots had attended 2 weeks of flight school, but this training offered only the basics - how drones work, how to fly them, how to stay safe. The practical realities of flying drones (especially large agricultural spraying drones) on smallholder plots in rural Myanmar were much more complex and challenging than what could be taught in a basic training course. Therefore, the field trials were not just a test of market fit for drone technologies, but also an active and ongoing training ground for learning how to operate in this complex and challenging environment.
Our first flights were difficult, to say the least. Our drone pilots were not yet confident in their abilities, and our processes and procedures were severely immature. This would serve as the starting point for our long learning process in the field. Each day we would typically split into two teams (one spraying and one imaging), and at the end of the day debrief on what had occurred, what we had learned, and how we can improve. There were two key tools that we used to track our progress and improve our operations. One was a simple ‘daily’ document where we recorded each day’s activities as well as any process, technology or customer learnings. The second was an operations checklist (actually two - one for spraying, one for imaging) that served as a detailed step-by-step guide for the drone pilots to follow. At the end of each day we would update the checklists and print out the new versions to be used the following morning.
Drone Spraying
The drone spraying operations were extremely intense and challenging. With 6 propellers and a payload of 10 liters, the drone was large and difficult to operate. It required a light truck to transport all the equipment, and required two people to carry the drone box. While we only sprayed water throughout the field trials, we developed and followed strict guidelines as if we were actually working with dangerous agrochemicals. We always operated in flight crews of two people (one pilot, one assistant), plus at least one other team member to observe. When we sprayed farmers’ plots, we tried to simulate as if we were servicing real, paying customers - not only so our pilots could practice, but also so we could understand how long each step would take and continue to improve the process.
With each day our operations checklist became more robust and sophisticated, and our drone pilots became more confident. Every mistake was a learning opportunity (including one day in which the drone crashed into a tree). On certain days we ran more specific experiments - for example, our pilots successfully ran a night spraying operation (which was at the same time exciting and terrifying). We also used this time to get a better understanding of our customers and the environment in which they operated. We conducted in-depth interviews with farmers to hear their feedback on the drone spraying service, but also spoke to other actors in the ecosystem such as machine owners, brokers and laborers. By the end of more than two weeks, the iterations on the checklist had slowed significantly, and we felt that we had built a solid understanding of the key challenges and opportunities around offering a drone spraying service to smallholders.
Key Learnings on Drone Spraying
Operations
Every plot is different. Paddy plots in Myanmar are small (1-3 acres) and irregular-shaped, with all sorts of obstacles including trees, houses, fences and even other crops. Some have good landing areas for drones, while others do not.
Drone pilots must be strong across a range of skill sets. They not only need technical skills, but also strong critical thinking, problem-solving and communication skills to face the unique challenges that each spray service job presents. Therefore, finding and retaining the right talent would be both crucial and extremely difficult.
Plot access is a major limitation. Given the size and amount of equipment required, it is very difficult to operate from any takeoff point that is more than 50 meters or so from the road. Given the poor quality of the road network in rural areas, this means that a huge percentage of plots are unserviceable, especially during the rainy season.
Seasonality is a major obstacle to financial viability. From our ‘year on your farm’ sessions, we found that the biggest peaks in demand for spraying occur early in each growing season due to the ubiquity of herbicide spraying. Fully servicing this period of peak demand would mean a huge excess in capacity during the slower periods of the year.
Using drones for broadcasting will not mitigate seasonal effects. Our team explored the possibility of using the same drones for broadcasting of seeds and fertilizer in order to increase the overall utilization throughout the year. However, the peak demand for broadcasting also occurs at the beginning of each growing season, so this would only exacerbate the seasonal effects of the business.
Maximizing daily utilization is crucial to operating a viable business. Spraying of pesticides and other agrochemicals is not advisable during midday hours due to the harmful effect on pollinators and the rapid evaporation of liquids. Possible ways to improve daily utilization include spraying at night and reducing transportation time by organizing nearby customers to be serviced on the same day.
Night spraying is possible, but risky. We successfully conducted night spraying operations and found it to offer certain advantages, such as cooler temperatures, smaller crowds and greater effectiveness against pests that come out at night. However, it is more risky and requires more skill from the pilots.
Reactive spraying is most impactful but operationally difficult. Reactive spraying (i.e. spraying to solve an unexpected pest or disease outbreak) has a much bigger effect on the farmer’s yield compared to preventive spraying (which is actually often more harmful than helpful). However, reactive spraying is difficult to offer because you cannot predict when or where the pest or disease outbreaks will occur, and therefore cannot plan ahead.
Crowds attracted by the novelty of drones can pose a safety concern. Drone operators need to account for the possibility that their activities will attract a crowd, and they should take safety precautions such as setting a perimeter to help control the crowd. One solution we explored was involving the customer farmer in the task, thus gaining extra capacity while making them feel more part of the team.
Technology
The technology is much more immature than we had thought. The DJI MG flight software is buggy and unreliable, making it difficult to consistently control flight parameters. These software issues, combined with certain fundamental design features of the drone and the spraying nozzles, result in a suboptimal quality of spraying.
Autopilot mode is very useful, but still requires skilled drone pilots. Drone pilots can use the DJI MG app to relatively easily create and save autopilot flight plans by walking the perimeter of the plots they are looking to spray. However, the pilots must always land the drone manually and be ready to switch to manual flight mode if there are any issues during the flight.
Agronomy
Spraying whichever agrochemicals the farmer asks may be counterproductive. Due to the poor advice farmers often receive, they may request to spray chemicals that are excessively strong or inappropriate for solving a particular problem. Therefore, a drone spraying service would be much more impactful if it were combined with high-quality agronomy advice.
Safe and effective chemical concentrations for drone spraying are unclear. Drones spray about 10-12x less liquid per acre compared to manual spraying, meaning that if we were to use the same amount of agrochemicals, the concentration would be 10-12x higher and therefore extremely dangerous. We could not find any clear research available on the appropriate chemical amounts or concentrations for drone spraying.
Customers
Labor is becoming more scarce, but it is still cheap. Rural migration has resulted in labor shortages in many rural areas, but labor prices are still very low - the average price of a laborer to spray one acre of paddy is about MMK 2,000 (~$1.30). This is the price point with which any drone spraying service would need to compete.
The health effects of manual spraying are not a major pain point from farmers’ perspectives. Since the most serious problems associated with pesticide and other chemical exposure are long-term (such as cancer, birth defects and neurological effects), the vast majority of farmers in Myanmar are unaware or unconvinced of the health consequences of manual spraying. Thus farmers see little value in one of the key benefits of drone spraying.
Drone Imaging
Concurrently with the drone spraying operations, we were also testing out the drone imaging technology. The key challenges were altogether different from those of spraying. The biggest obstacles were not operational, as the imaging drone was small, flying at high altitude away from obstacles, relatively easy to operate, and could cover over 100 acres in a single flight. Rather, the challenge was to hack together a process flow suitable for serving smallholder farmers using hardware and software that were designed for large farms in the developed world. Specifically, we needed to figure out a process in which we 1) use the drone and sensor to collect crop health information, 2) identify and ‘ground truth’ problem areas, and 3) successfully distribute the data to the respective farmers.
The first step of the process, collecting crop health information, was by comparison the most straightforward. NDVI, or normalized difference vegetation index, has been used for decades to measure crop health from satellites, and more recently from drones. It is a simple measure of plant vitality based on how the plant reflects light at certain frequencies. Our Sentera NDVI sensor could capture the near-infrared (NIR) images required to calculate NDVI. We used Sentera FieldAgent software, which included a flight planning app as well as an offline desktop program that would ‘stitch’ together the NIR images into a cohesive NDVI map of the flight area. The Sentera tech certainly had its limitations and frustrations, but overall it seemed to meet our basic requirements.
The next step was to make sense of the crop health information that we had collected. NDVI is useful for roughly identifying which areas might have a pest, disease or soil problem, but it cannot tell you what the specific problem is. Therefore, it is required to ‘ground truth’ the problem areas in order to give a specific diagnosis. Farms in developed countries are typically large, homogenous and well-organized, so ground truthing problem areas in person is a worthwhile exercise. However in Myanmar, a 100-acre flight can capture information on the plots of over 20 unique farmers, and accessing each of these plots can be extremely difficult and time-consuming. Therefore, our team needed to come up with a viable alternative to in-person ground truthing.
The solution we developed was to use the drone’s normal RGB camera to take up-close photos of the problem areas that we could then have agronomists use to give remote diagnoses of the specific problems. But this would require flying the drone to areas that were well beyond the line of sight of the drone pilots, which is quite risky. The FieldAgent flight planning software did not have any features that could help us with this task, so we needed to search for other, more flexible flight software programs. We eventually discovered Litchi, a flight software with which we could set points of interest (POIs) on the map where we wanted to investigate problem areas. Our pilots could then manually fly the drone to each POI at a high altitude (50+ meters) to avoid obstacles, use the drone camera to safely descend and take up-close photos of the plants, and then move on to the next POI. After the flights, the pilots would upload the photos to the cloud, and an agronomist could look through and give a diagnosis for each problem area.
The final step was to find a way to package and distribute all this data to farmers in a way that was simple and actionable. There were a range of different methods and channels that could have been used to distribute the data, but we decided to prototype with a printed ‘Crop Health Report’ that we would give to farmers in person. The report included the NDVI map of the farmer’s plot, the up-close photos of the plants, a diagnosis of the problem(s) in the plot, and detailed recommendations on how to solve the problems and prevent them in the future. We had time to share three different rounds of Crop Health Reports with farmers, each time receiving feedback from the farmers and improving the format and content of the reports for the next round.
Thus we had successfully developed a rough but viable process for collecting, analyzing and sharing valuable crop health information with smallholders.
Key Learnings on Drone Imaging
Operations
A viable service will only be possible if we can serve clusters of nearby plots. Offering a crop monitoring service to farmers on an individual basis would be operationally untenable given the time and effort required for each flight. In practice in the Myanmar context, this would mean entire villages organizing and signing up for the service together. An alternative approach would be to use Crop Health Reports as a free customer acquisition tool in a portfolio of monetized products and services.
Agronomy
Certain problems are easier to diagnose with drone photos than others. The confidence levels of our agronomists were very high for certain problems such as weeds, while certain pest and disease problems were more difficult to confidently diagnose remotely.
Crop monitoring flights for paddy are most useful 30-70 days after planting. This is the period when the most serious pest and disease problems occur. It is also optimal for NDVI flights because the plants are tall enough to create a canopy that reduces the amount of sunlight reflecting off the water in the paddy fields.
Technology
The existing software environment does not match our needs well. Ideally we would have been able to use a single software program on a single device to perform flight planning, NDVI processing and ground truth flights. However, completing these tasks with the existing software environment required multiple software programs across multiple devices, using several manual bridges.
Cloud-based software platforms popular in developed countries are not suitable in developing, rural areas. Given the spotty coverage, slow speed and high costs of cellular data services in rural Myanmar, we had to find software that allowed for locally hosted, offline stitching of drone images and calculation of NDVI maps. This increased both the operational complexity and the skills required of the operators.
Customers
Farmers highly value the recommendations included in the Crop Health Reports. They appreciated receiving the reports in general, but in particular valued the recommendations for how to solve their problems and how to prevent them in the future. The recommendations were detailed and included options for biological treatments, which differentiated them from other advice farmers typically receive.
Conclusions (Technology Prototyping)
Service Model Prototyping
Given the promising potential of offering a crop monitoring service to farmers, our team began to prepare for our next round of field trials. This time they would include 2 weeks in Minbu township, still in Myanmar’s central Dry Zone. Our objective was to develop 2-3 service models to mid-fidelity. Whereas the previous field trials focused on understanding the technology and operations, these trials would try to answer the question, “how can we actually offer this as a viable service?”
Our team kicked off this new phase with a concepting workshop to review the key insights from the previous field trials, develop key problem statements, and brainstorm different service model ideas to test. In the end we decided on three models to test, which we called ‘Low Touch,’ ‘Mid Touch’ and ‘High Touch’. As the names might suggest, we wanted to test service models that would vary in the level of engagement with farmers and the relative cost to serve.
Service Models Overview
Low Touch
Collect NDVI only (no ground truth photos).
Use NDVI to estimate whether there might be any problems in each farmer’s plot.
Send SMS alert asking farmers to check their plots for problems and to call our agronomist if they find anything.
Mid Touch
Collect NDVI plus ground truth photos.
Give specific diagnosis of problems, and recommendations to solve.
Call individual farmers to share the advice, followed by a summary SMS message.
High Touch
Collect NDVI plus ground truth photos.
Give specific diagnosis of problems, and recommendations to solve.
Compile printed Crop Health Report for each farmer and share in a group setting.
We selected villages in Minbu that were far enough apart from one another that we could test each service model in relative isolation. Similar to the previous field trials, the first step was to collect basic information from the villages where we would be operating. We called this process ‘base data collection’ - similar to the ‘year on your farm’ sessions with farmers, but much quicker because we were not collecting detailed crop cycle information. In other words, we collected basic information about each unique plot and its owner, and mapped out the geographic outline of each plot in Google Maps. It was necessary to collect the base data before we started our operations because we needed to know whose plots we would be flying over and how to get in touch with the respective farmers.
We also needed to set up a system to store and manage the base data as well as the data generated from each of the monitoring flights. For the purposes of the field trials, we built a very basic data storage system using Google Sheets and Google Drive. We implemented a strict naming scheme to keep track of the various forms of data, including flight plans, NDVI maps and ground truth photos. This system was sufficient to meet our needs during the field trials, but it was admittedly very manual and prone to human error.
During the previous field trials we tested our crop monitoring process on just a handful of plots and farmers, but now we needed to add more structure and detail to the process. To start, we divided the plots into ‘clusters’ comprising about 100 acres and 20-25 unique plots (since the drone could cover about 100 acres in one flight). This helped to streamline certain parts of the operations because we could save and reuse the flight plans for each cluster, and also use the clusters as a way to group farmers together for data distribution.
We also added more structure to the process of giving remote diagnoses and recommendations. Though both of our agronomists were remotely diagnosing problems with a high level of confidence, we surprisingly discovered that there was actually a high number of disagreements between them when assessing the same ground truth photos. This finding led us to later conduct more structured tests with other agronomists to examine how closely each agronomist’s remote and in-person diagnoses matched.
With the base data collected, the data storage system running and the process flow refined, we proceeded to test each of the service models in the respective villages. Each day the team was typically split into one field team that would fly to collect NDVI and ground truth photos, and one base team that would remain at the popup studio to manage the data, generate Crop Health Reports, call farmers, etc. While our team spent about 80% of our time in the field during the previous field trials, we were now spending about 80% of our time at the studio. This fact reflects the shift in focus from technology and field operations to back-end operations and customer engagement.
For each service model, we conducted 2-3 rounds of collecting, analyzing and distributing the crop health data to farmers. We measured the relative time and effort required for each step of each service model, and conducted follow-up calls and in-person interviews with the farmers to receive their feedback. By the end of two weeks, our team had collected an abundance of operational data and qualitative insights that we would use to give our final assessment of the three crop monitoring service models.
Key Learnings on Crop Monitoring Service
Technology
NDVI is a good indicator of problem areas. Despite some particular challenges of using NDVI with rice, we were able to use it as an effective indicator of which areas we should investigate further. We successfully found pest, disease or soil problems in 87% of the problem areas that we identified with NDVI.
Customers
It is very difficult to consistently reach Myanmar farmers through any channel. Farmers rarely check their SMS messages (because they receive so much spam from telco providers) and frequently purchase new SIM cards (therefore changing phone numbers). Gathering a specific group of farmers for an in-person meeting can also be challenging as it requires notifying each of the individual farmers, either directly or through a ‘contact farmer’ in the village.
It is difficult to effectively communicate plot locations with farmers. With no nationwide geo-referencing system to use and limited map literacy, farmers typically describe their plot locations in reference to local landmarks or their neighbors’ plots. With this in mind, we helped farmers understand the NDVI maps in the Crop Health Reports by labeling the surrounding plots with the names of the plot owners.
Most farmers already know when their plot has a problem. Detection of problems was much less valuable to farmers than the specific diagnoses and recommendations.
Farmers seem to be willing to pay for Crop Health Reports. Compared to when they received crop health information over the phone, farmers really seemed to value the Crop Health Reports as a more tangible and trustworthy product. On average, farmers said they were willing to pay about MMK 5,000 (~$3.30) per report.
Operations
Base data collection is a critical prerequisite to operations. It is a significant but worthwhile one-time investment that would require a trained, disciplined team to collect data at scale. Without collecting the base data, it would be impossible to offer a crop monitoring service to our customers.
A more robust data storage and management system is required to scale. Crop monitoring operations generate huge amounts of data that would need to be carefully stored and managed at scale.
An ‘assembly line’ approach can help to streamline imaging operations. Certain steps of the aerial imaging process, such as assessing NDVI and plotting ground truth points in Litchi, could be performed remotely by operations support staff. This would allow the drone pilots to focus their time on flying, thereby increasing the daily utilization of the drones.
Agronomy
The accuracy level of remote diagnoses is disappointingly low. When we conducted tests with agronomists to measure how closely their remote and in-person diagnoses matched, the results were disappointing, with an accuracy level below 50%. There could be ways to improve the accuracy level in the future, such as offering agronomists specific training on remote diagnosis or improving the quality of our drone photography, but it is unclear how big of an effect these steps would have.
Conclusions (Service Model Prototyping)
After completing the field trials, our core team spent another 2 weeks synthesizing the data, building a financial model and preparing our report and recommendations.
We found that the Low Touch model had little to no impact. Few farmers even noticed or read the SMS messages that we sent out. Further, the messages only included a generic alert that there might be a problem in the farmer’s plot - without the specific diagnosis and recommendation of the problem, the value of the messages was very limited.
The Mid Touch model was much more impactful because our agronomists shared the specific diagnoses and recommendations with farmers over the phone. Though reaching farmers by phone was still a hassle, the Mid Touch model required less time per customer than High Touch because we did not need to spend time generating and distributing the physical Crop Health Reports. However, farmers found much less value in the Mid Touch model, as they were only willing to pay MMK 1,000 (~$0.65) or less for the service.
Most farmers found the Crop Health Reports to be very valuable, given the positive feedback we received and the indicative willingness to pay of about MMK 5,000 (~$3.30). Though the High Touch model required more time per customer, the potential revenue earned from the Crop Health Reports would more than make up for it. In addition, farmers who received the High Touch service were more likely to follow our recommendations compared to those who received the Mid Touch service, according to our follow-up surveys. Thus, High Touch proved to be the most promising service model in terms of impact, scale and financial viability.
Initially our team’s recommendation was to run a further pilot project to test a version of the High Touch model with real customers across an entire growing season. However there were two key concerns that emerged.
The first concern was related to scale and financial viability. Our financial model showed that the service could be profitable, but profitability was highly dependent on the price at which we could sell the Crop Health Reports and the rate at which we could convert farmers into paying customers. The service would need to scale in a linear fashion with the hiring of a significant number of new staff, so this could pose a serious risk if the service were to end up being unprofitable.
The second concern was around the accuracy of remote diagnoses. The results from our accuracy testing were very disappointing and made us question the level of positive impact that our service could have. We would have been much more comfortable with an accuracy level of 70% or higher, but we were not confident that we could reach that level with any incremental steps for improvement.
Unfortunately, our team’s final recommendation was to pause the project and only move forward with a crop monitoring service if there were significant improvements to the diagnosis accuracy.
Where do we go from here?
In the end our social enterprise client did not launch a drone service, but there were a host of other positive outcomes from the project. Many of the learnings from our customer research were used to improve their existing crop protection services. We also generated many useful insights into the challenges and opportunities of ‘farming-as-a-service’, which is a space that will likely grow in importance in Myanmar over the next few years. And it was a great learning opportunity for the project team members, using a design process that in many ways could be replicated for future projects.
Perhaps most importantly, we hope that by sharing our experiences through this case study we can help and connect with other companies and organizations supporting agricultural drone initiatives for smallholder farmers. We encourage anyone who is interested in learning more to reach out to us, and we look forward to continuing the conversation.
We were very proud of the amount of learning that our team was able to accomplish in less than three weeks in the field. We had drastically moved up on the learning curves of both drone spraying and imaging, in terms of the technology, process and operations. And we had generated key insights that we could use to assess the viability of offering agricultural drone services in rural Myanmar.
Though drone spraying was an exciting opportunity with certain potential for impact, offering this type of service to smallholders in Myanmar is simply not financially or operationally viable. It would require significant investment in training highly skilled staff who would need to operate under extremely difficult conditions (lack of georeferencing databases, low map literacy of customers, poor road network, unreliable power grid, spotty cellular coverage in the field, etc.), while competing against cheap manual labor in a highly seasonal business. Further, the drone spraying service did not solve any major customer pain points, at least from their perspective. Our team’s recommendation was not to pursue drone spraying further, with the gap to viability proven too wide.
We saw more long-term potential in using imaging drones to offer a crop monitoring service to farmers. Though the process we developed during the field trials was not perfect, it was enough to give us confidence that we could offer real value to farmers regarding the health of their crops. We were especially encouraged by farmers’ positive responses to receiving the Crop Health Reports. Our team recommended to move forward with a second round of field trials to prototype different crop monitoring service models with customers.