India has hundreds of thousands of independent farmers working in hard-to-reach areas, and it can be difficult for these agriculture producers to gather the information they need to maximize crop yields and profits. Organizations like Digital Green are dedicated to using technology to provide farmers with training, information about market prices, and other extension services. Traditionally, they have done so using video, but as cellular penetration increases, there is an opportunity to give farmers customized information via mobile phone. Digital Green is in the process of developing Farmstack — a platform that helps organizations share data with one another. This system can be used to share customized information and services with farmers, using multiple models of communication. The advisory and research organization IDinsight worked with Digital Green to assess the effectiveness of this platform. Farmstack is still under development, and Digital Green is testing out different use-cases of the system in various settings. IDinsight studied a particular Farmstack use-case that involved cashew farmers in Andhra Pradesh, in which farmers were given customized advisory information through video, Integrated Voice Response (IVR) and SMS. A Farmer Producer Group gathers to demonstrate the preparation of an organic fertilizer in Visakhapatnam, Andhra Pradesh. ©IDinsight/William Slotznick Combining videos with customized information leads to higher technology adoption Cashew farmers in Andhra Pradesh generally have low yields, due to inclement weather and nutrient-deficient soil. To promote productivity, the government of Andhra Pradesh has encouraged farmers to adopt “Community-Based Natural Farming” (CBNF previously known as ZBNF) techniques. CBNF involves farmers using natural ingredients (such as cow dung and urine) to create chemical-free fertilizers and pesticides, which can help them achieve higher production at a low cost. The information provided by Digital Green helped disseminate these CBNF techniques developed and endorsed by the Agriculture Department. IDinsight ran a randomized controlled trial, in which one group of farmers (“video-only”) were invited to watch informational CBNF-focused videos every two weeks. Another group of farmers received an implementation of the “Farmstack” system, in which Digital Green delivered supplementary communication to farmers on their phone, along with the in-person video screenings. These messages reinforced content that appeared in the videos and also provided targeted information on weather and soil quality. In addition, IDinsight’s study included a non-random group of cashew farmers from outside the project area who did not receive any training from Digital Green. This “non-intervention” group serves as a useful comparator to the two program arms — although we cannot use it to make causal claims about the effectiveness of the video-only treatment, since these farmers were selected through a different process. Unexpectedly, India’s COVID-19 lockdown in March disrupted video dissemination in the midst of the cashew-growing season. While video sessions paused, farmers were still able to access SMS and IVR messages. In light of these changes, IDinsight conducted all data collection via phone-based surveying. A farmer shows the research team cashew seeds that were stored from the prior harvest in Visakhapatnam, Andhra Pradesh. ©IDinsight/William Slotznick What did we find? First, we found that phone-based advisories effectively reached cashew farmers in Andhra Pradesh. Analyzing the administrative data from the IVR system, IDinsight found that 81.6 per cent of recipients received at least one IVR and listened to, on average, 63 per cent of the message content. Phone survey data indicates that 68 per cent of respondents report attending a video screening, and 59 per cent report receiving at least one IVR call. As the phone is a shared asset in most of the households, it’s possible that a call was actually received by other family members, making the success rate even higher than reported. Farmers also report that, in general, the IVR messages were easy to understand and relevant to their needs. Farmers showed increased knowledge of how and when to apply the CBNF practices. These effects are illustrated in the adoption data as well. For four key natural production practices, farmers in the Farmstack group consistently reported higher adoption by 5–8 percentage points with respect to the video-only arm (where adoption ranged from 13–30 per cent, depending on practice). This may not seem like much, but it’s important to note that Farmstack is inexpensive — adding a marginal farmer to the system costs approximately $3.40 over the course of a season. It’s notable that farmers in non-intervention areas saw very low levels of adoption for any of the CBNF practices. This indicates that the video-based approach likely increased adoption as well. Among the 46 per cent of farmers in the Farmstack group who did not adopt any of the practices, the most cited barrier was “poor access to the requisite inputs,’’ followed by “too high effort” — suggesting that farmers understood the messages but faced other barriers to adopting new practices. Even though CBNF is designed to be easy for farmers to adopt (meaning information is the key constraint to adoption), it’s clear that farmers faced other hurdles to adoption, even after they learned about the CBNF techniques. Adoption rates for four key CBNF practices across arms. While Farmstack farmers did report somewhat higher yields, our yield measurement is very noisy, so the difference between our two treatment groups is not significant. What are some lessons for the future? While Digital Green’s video dissemination model leverages a field monitoring system to record viewership and adoption, like in this use case, many other mobile phone-based extension models relied on phone logs for tracking access. While these logs are useful for assessing reach, they do not reveal the actual adoption of practices and results in the field. Going forward, Digital Green is considering tracking data from multiple hybrid channels — such as layering mobile data collection onto the Farmstack system — in order to better monitor success. Secondly, given that the FarmStack platform supports data integration from multiple sources, there is a need to get the incentives for data-sharing right and ensure that organizations trust that their data, and its uses, will be secure and controlled. Digital Green plans to include peer-to-peer connectors and usage policies, which will help organizations control who, when, and for what purpose, others may use this data. What does this mean for policy? Well, it’s another great sign that reaching farmers via additional digital channels (like mobile phones and video) is a successful and scalable strategy. This case had a couple of unique features when thinking about extrapolation: the IVR messages were delivered on top of videos (at least early in the season before COVID-19), which provided messages that reinforced what farmers had recently learned via video. We do not know if these messages would have worked without the videos. Also, the Farmstack system provided customized weather advisories and soil information. This is an added layer of complexity and cost to the system but also potentially makes it more effective. Future investigations could help us understand exactly what kinds of personalized information is valuable to farmers, which would help develop the most optimal system possible.