Our Learning

Precision Development (PxD) leverages scale and the design of two-way information systems to learn, innovate and iterate for impact. 

At scale, our projects demonstrate improved economies of scale associated with reduced marginal costs. Critically, from a learning perspective, PxD’s work at scale gives us access to richer user data which – when complemented by rigorous experimentation and testing – contributes to more efficient learning. 

We learn and innovate by intentionally drawing on and synthesizing economic theory, empirical evidence, local knowledge and user feedback. By continuously harnessing feedback loops, we progressively iterate and improve the quality and impact of our advisory content, the quality and efficacy of our service delivery, and impact. 

Our research is a global public good which contributes to the knowledge base for all practitioners and researchers active in this sector. We use our findings to improve the quality of our services, as well as to inform the services that our partners offer.

Wherever possible, we make our learnings and research outputs publicly available to spark further innovation and research in the sector, thereby contributing to the generation of larger aggregate impacts for the poor.

How We Work

PxD’s work is informed by our user experience and rigorous research. Our interventions, and the improvements we make to those interventions, are evidence-led and evidence-based. PxD continuously monitors our programs: 

  • At every stage of service design and delivery, we generate insights on how to serve our users better, and rigorously test these insights. At scale, these systems allow us to gather large amounts of direct feedback from users in order to continuously improve our performance.
  • We conduct A/B testing – comparing two or more service design options to assess which is preferred or more effective – to inform rapid upgrades to our content and service delivery. To date, PxD has conducted over 60 A/B tests. Drawing on insights from our program data, external behavioral science evidence, and best practices among peer organizations to design these tests, we are able to iterate improvements for user experience and deliver more appropriate, accessible and customized information. 
  • We deploy human-centered design practices to continuously improve our understanding of our users, and to design products and interfaces that meet our users where they are – including design for low literacy, technology-constrained, and very low-income users. 
  • We use experiments and trials to systematically understand impact and feed this information back into our model to refine it over time. 
  • We engage with, learn from, and contribute to leading academic and policy research in the behavioral sciences, economics, and other social sciences.

What We Know About Digital Extension

Research demonstrates that mobile phone-based agricultural extension services can increase the adoption of optimized farming practices and improve productivity and yields. 

Well-designed messages change farmer behavior. 

SMS-Extension and Farmer Behavior: Lessons from Six RCTs in East Africa, Fabregas et al, 2019a

A meta-analysis of six studies (four of the six studies analysed were undertaken by PxD) in Kenya and Rwanda found that, on average, farmers who received advisory text messages promoting the use of agricultural lime adopted the input at a rate 11.3% higher than farmers who did not (Fabregas et al., 2019). 

Mobile’izing Agricultural Advice: Technology Adoption, Diffusion, and Sustainability, Cole et al., 2020

In Gujarat, India, cotton farmers were more likely to use fertilizers after receiving visual aids and weekly push calls to promote customized recommendations on soil fertility. Analysis suggests that this intervention significantly increased the likelihood that farmers used recommended inputs, most notably doubling the proportion of farmers who adopted recommended application of UREA and MOP, compared to farmers in the control group (Cole, et al. 2020). 

CABI and OAF Trials, 2020

SMS messages to Kenyan farmers, tweaked on the basis of evidence generated by various A/B tests, increased the self-reported likelihood of farmers adopting recommended practices to address Fall Armyworm by 5 to 23% (2020, Kenya, CABI and OAF trials).

Changes in farmer behavior can increase yields. 

Realizing the Potential of Digital Development: The Case for Agricultural Advice, Fabregas et al 2019b

Evidence suggests that changes in farmer behavior can lead to yield increases across a variety of settings. A meta analysis of seven studies in Africa and India demonstrates a 4% average yield gain associated with digital agriculture programs (Fabregas, Kremer, Schilbach. 2019). This increase is an average effect among all farmers to whom messages were sent, including farmers who did not open or engage with the content. As a consequence, yield gains are likely to be concentrated among a smaller group of farmers who do adopt recommendations, and the per-farmer impact for farmers engaging with our content is likely to be significantly higher.1 

At scale, our interventions are very cost-effective due to low-costs. 

Realizing the Potential of Digital Development:The Case of Agricultural Advice, Fabregas et al 2019b

Providing digital advisory services is extremely cheap and systems are able to reach large numbers of users with ease. Estimates suggest that PxD’s benefit-cost ratio may be as high as  10:1 (Fabregas et al, 2019). PxD’s cost per farmer has been steadily decreasing with each year of operation and in 2020 is approximately $1.25 per farmer per year.2In comparison, in person farmer field days cost between $35 and $153 per farmer reached.3nbsp;

A/B tests generate insights for increasing user engagement. 

Fall Armyworm Message Design, PxD 2019

PxD has effectively increased farmer engagement with digital content by testing small tweaks in messaging through A/B trials. For example, in Kenya, PxD found that small, meaningful changes in the words and language used (for example, using urgent language) in messages can increase engagement by three percentage points. 

Gujarat Call Length A/B Test, PxD 2017

Results from A/B testing on customization features found that we could nudge farmers in India to listen to 28.8% more content if we gave them the option to choose between two short calls a week or a single, long call. 

Ethiopia, Rotating IVR Menu Options, PxD 2018

Tweaking message framing – for example, by including a user’s previous best quiz score in a messaged invitation to participate in a quiz about Fall Armyworm in Kenya – increased response rates by 11 percentage points over the control group. Rotating the ordering of agricultural content on a hotline audio menu in Ethiopia so that the most relevant option is always #1, resulted in a two-fold increase in farmers reaching the most timely content. 

Increased user engagement leads to improved knowledge and comprehension. 

PxD SMS Lime Trials, 2016

Improved knowledge empowers farmers to make more informed and optimal decisions for their farms. Evaluations of PxD’s Kenya program found that maize farmers who received customized SMS messages on the use of agricultural lime for soil health and healthy plant growth had 21% to 28% better knowledge on the topic compared to farmers who did not receive messages (2016-17 Kenya, PxD SMS Lime Trials). 

The Promise and Challenges of Implementing ICT in Indian Agriculture, Cole & Sharma, 2017

In Gujarat, India, PxD supplemented the state’s Soil Health Cards (SHC) for fertilizer recommendations with several different types of digital extension to assist farmers with comprehension. An evaluation of the program found that farmers who received a customized SHC together with an audio supplement, demonstrated comprehension of the fertilizer recommendations 37 percentage points higher than farmers who received the SHC without supplemental materials (Cole & Sharma 2017).4


Expanding the Evidence Base

The true impact of PxD’s work is likely to be larger and should increase over time. 

SMS-Extension and Farmer Behavior: Lessons from Six RCTs in East Africa, Fabregas et al, 2019a

Impact estimates do not take into account spillover effects and in many instances, non-beneficiaries can learn and benefit from our services. For example, we have observed that SMS messages to particular farmers enrolled in farmer groups in Rwanda increased lime adoption for all farmers within the group, including those who did not receive messages (Fabregas et al, 2019). 

Crony Capitalism, Collective Action, and ICT: Evidence from Kenya Contract Farming, Casaburi et al, 2019

The involvement of other players in value chains can also enhance impacts. For example,  the implementation of a two-way hotline for Kenyan sugarcane farmers to report fertilizer delivery delays to the contract farming company reduced delivery delays by 22% (Casaburi, Kremer, Ramrattan. 2019). Over time we anticipate that the adoption of more advanced technologies by farmers, such as smartphones, and the integration of more accurate and scalable technology to our services, such as remote sensing, will further improve the quality of advice we can give. Continued A/B testing and iterations on our messages will continue to increase impacts over time. As we generate more evidence and share our learning and insights, other implementers and policymakers will scale successful programs and discontinue unsuccessful ones with benefits accruing to the community of development practitioners and beneficiaries at-large.

PxD’s Research Going Forward!

PxD’s research agenda will focus on mechanisms to achieve improved outcomes for farmers and other users:

  • PxD will continue to strengthen our evidence base on outcomes of interest, such as increased yields, averted crop losses, improved farm profitability, and increasing cost-effectiveness of our services. 
  • We also aim to deepen our understanding of the mechanisms through which our programs achieve impact. For example, we are studying how our programs can facilitate social learning, for example  when users of our services share knowledge with others and positively influence the likelihood that non-users  adopt recommended practices. 
  • We will continue research on how to efficiently and cost-effectively collect accurate information from users which can then be used to improve recommendations, for example, crowdsourcing information on the incidence of pests or input availability. We will also explore how our programs can improve the effectiveness of traditional agricultural extension by motivating employees or sharing knowledge. 
  • We  plan to investigate how our programs can reduce information asymmetry in markets through interventions like providing farmers with agro-dealer contact information, as well as input and price availability. 
  • Lastly, to address the effects of climate change, we will study how our programs can provide and enable farmers to practice, inter alia, regenerative agriculture, climate-smart agriculture, agroecology, and organic farming. These practices present clear opportunities to improve upon the conventional focus on yields and food security toward a more holistic consideration of soil health, environmental sustainability and farmer resilience and independence.   

References

Selected publications and working papers

Casaburi, Lorenzo, Michael Kremer, and Ravindra Ramrattan. 2019. “Crony Capitalism, Collective Action, and ICT: Evidence from Kenyan Contract Farming.” Working paper.

Cole, Shawn A., and A. Nilesh Fernando. 2020. “‘Mobile’izing Agricultural Advice: Technology Adoption, Diffusion, and Sustainability.” Working Paper, March.

Cole, Shawn, Tomoko Harigaya, Grady Killeen, and Aparna Krishna. 2020. “Using Satellites and Phones to Evaluate and Promote Agricultural Technology Adoption: Evidence from Smallholder Farms in India.” Working Paper, September.

Cole, Shawn, and Garima Sharma. 2017. “The Promise and Challenges of Implementing ICT in Indian Agriculture.” India Policy Forum 2017-2018. New Delhi.

Fabregas, Raissa, Michael Kremer, and Frank Schilbach. 2019. “Realizing the Potential of Digital Development: The Case of Agricultural Advice.” Science 366 (6471): 30–38. https://doi.org/10.1126/science.aay3038.

Fabregas, Raissa, Michael Kremer, Matthew Lowes, Robert On, and Giulia Zane. 2019. “SMS-Extension and Farmer Behavior: Lessons from Six RCTs in East Africa.” Working Paper, September.

  1. The 4% yield gain estimates the “Intent to Treat – ITT”. Estimates of “Treatment on the Treated – TOT” would likely show higher gains but that data was not available for this study.
  2. Calculated including all fixed costs and overheads (annual all-in costs/total farmers reached per year)
  3. Based on evidence from Uganda, Mozambique, Indonesia, Bangladesh from a) Low and Thiele. “Understanding innovation: The development and scaling of orange-fleshed sweetpotato in major African food systems.” Agricultural Systems, 2020; b) Quizon, Feder, and Murgai. “Fiscal sustainability of agricultural extension: The case of the farmer field school approach.” Journal of International Agricultural and Extension Education. 2001.; c) Ricker-Gilbert et al. “Cost-Effectiveness of Alternative Integrated Pest Management Extension Methods: An Example from Bangladesh.” Review of Agricultural Economics, 2008.
  4. Furthermore, with a video supplement, a farmer’s comprehension was on average 41 percentage points higher, only slightly below the 48 percentage point comprehension increase associated with in-person meeting with an agronomist.