IVR and text message interventions to provide fertilizer information to farmers — experiments from India
One of the biggest challenges farmers in India face is ensuring that their soil is healthy — enabling sustained high crop yields. India’s total fertiliser consumption is less than the global average but it is the second largest consumer of urea (a nitrogen based fertilizer) in the world. To address this, the government of India launched the Soil Health Card (SHC) Scheme, providing every farmer in India with a customised set of crop-specific fertiliser recommendations that would ensure the long-term productivity of their soil. These recommendations are printed on cards and distributed by agricultural extension workers who are expected to explain the general purpose and recommendations of the card to farmers.
However, in a qualitative study conducted in 2018 covering about 450 farmers, we found that across eight districts, overburdened extension workers are unable to adequately explain soil health card recommendations to farmers. Rather than individually visiting each farmer, extension workers often distribute cards en masse in villages.
Experimental evidence from Gujarat, authored by Shawn Cole and Garima Sharma of Precision Agriculture for Development (PAD), suggests that supplementing an in-person explanation of the card with either audio recordings or video clips may enhance farmers’ understanding. Compared to in-person extension, both audio and video performed equally well in terms of improving comprehension and eliciting farmers’ trust in the recommendations. This is insight is particularly useful because in-person interactions are more expensive than remote options.
Recently, IDinsight conducted several ICT-based campaigns with farmers. We found that pick-up rates among farmers spike at particular times over the course of the day (ranging from 9.30 AM to 8.30 PM) and that call-backs at these times are effective at flipping almost one-third of non-respondents. We also found that farmers were highly unlikely to respond or interact with messages that required keypress or voice responses. We have incorporated these results into the design of an ICT-based intervention, the impact of which will be assessed through an Randomized Controlled Trial later this year. In this intervention, we will be sending farmers non-interactive voice blasts at these specific times of the day.
Our processes and results are outlined in detail below.
II. Our Objective: Using SMSes and IVRs to explain soil health cards
To educate farmers on balanced fertilizer usage, we explored the feasibility of communication channels like text messages, automated calls, videos etc. During our initial scoping, we found that smartphone usage was low among the farmers limiting the set of feasible channels to text messages and automated calls.
In January and February 2019, we designed and recorded a set of educational messages on the soil health card scheme, that were sent to farmers through mobile phones. These were designed to be feasible for the government to implement through their existing systems of mobile-based extension.
Two channels were considered for this exercise:
- Interactive Voice Response (IVR): a method of automatically calling respondents with pre-recorded message at $0.028 per minute per call in India. A drawback of this medium is that farmers who are unfamiliar with it, may have limited willingness to listen and interact with the calls.
- Short message services or text messages (SMS): SMSes are more familiar to farmers, can be re-read as needed, and are inexpensive ($0.015 per SMS in India). Two drawbacks — there is a challenge reaching illiterate farmers and also making the messages brief.
To develop the content to disseminate, we used findings from our study on farmers’ knowledge and attitudes towards the soil health card scheme. The messages had three objectives:
- Inform and remind farmers about the Soil Health Card scheme given that many are unaware they had received a soil health card, unable to recognise it, or had forgotten that they had received it.
- Highlight how the card can help farmers, and what benefits they can expect if they use the card.
- Provide farmers with some broad instructions on how to read and use the information on their card.
The team wrote scripts of messages to be sent through IVR and SMSes, which were then translated into Hindi and recorded using a simple smartphone recorder. We also produced a literature review of best practices around ICT interventions in rural settings, and interviewed other organisations with experience using IVR calls in rural India.
III. Testing ICT interventions
We ran two pilot tests of ICT interventions. Each pilot included multiple rounds of campaigns. With these pilots, we wanted to identify: a) the viability of the technology platforms for the intervention, b) farmer engagement with touch or voice response requests and the length of message farmers are most likely to listen to, c) the best time to contact farmers, and d) the optimal call-back protocol for the intervention.
In the pilot, we tested various interventions, including ICT messages (IVR and text):
- Simple recorded message explaining how to read the Soil Health Card, and giving contact information for the Kisan Call Centre if farmers had any questions
- Touchtone response asking farmers to press a key on their phone if they wished to hear more information. The content remained the same as the recorded message.
- Voice response asking farmers to say something if they wanted the message to continue
- Only SMS introducing farmers to the Soil Health Card and giving contact information of the Kisan Call Centre.
IV. Pilot results
First campaign: To assess response rates for different types of interventions (simple recorded message, touchstone response, voice response, and only SMS).
Design: We sent 508 farmers one of four randomly assigned messages, all of which contained the same content in terms of soil health card advice. The first group received a simple uninterrupted voice recording; the second received requests for keypresses to continue the message; the third was asked for voice response to continue the message; and the fourth received only an SMS. The first three groups also received an SMS after the call ended. The farmers that did not pick up were called back twice.
Results: The table below highlights our findings. This excludes numbers registered on the DND database. After all callbacks, 116 calls were answered, giving a total pickup rate of 41.1 percent.
As previously mentioned, we confirmed through this pilot that farmers seldom engage with interactions requests. Only 8 out of 87 farmers made any keypress responses, and even fewer responded to prompts for voice interactions.
Decision: Due to the low interaction with keypress or voice requests, we proceeded with simple voice recordings and SMSes.
Second campaign: To test different call timings to determine which time of day results in the highest pickup rate and to judge the success of two different message types by comparing call duration (or, the amount of time farmers stayed on the phone).
Design: We sent two versions of a one-minute recorded message to a total of 535 farmers . We limited the messages to a minute based on the findings of our second campaign. Both messages tested contained the same information but in a different order. We randomly assigned farmers’ phone numbers to different groups, and these farmers received calls at different times of day. Initial calls were made between 9:00 AM and 8:30 PM, on 13–15 February 2019.
Results: Selected results with high pick-up rates shown in the table below. Calls made at 12:30 AM, 5:30 PM, 7.30 PM and 8:30 PM were the most successful in terms of rates of pick-up and call duration.
The first of the two messages performed slightly better in terms of call duration, but this result was not uniform across time. On subsequent calls, the first message was chosen, since it prioritised the information we most wish to deliver about soil health cards.
We redialed numbers that were not reached in the first round. Of the 121 numbers we called back, 39 answered during at least one callback.
Decision: Based on these results, we selected the highest-performing time slot for call scheduled for the upcoming intervention/RCT. We also set the duration of the recorded message to under ~75 seconds.
Third campaign: To assess response rates to call backs that were spaced at different intervals from an initial call.
Design: We called 30 farmers in March 2019. These farmers had not been contacted during the previous pilot and were expected to have similar response rates.
Farmers received a 75 second recorded message over the phone. The farmers who did not respond to the initial call received call backs at differing times. Half of the farmers received call backs every 30 minutes, while the other half received call backs every 4 hours.
Results: We found that calls at 9:30 am resulted in the highest response rate and duration. While the overall sample size is too small to make broad inferences, we find that the timing of the call backs is more important than their duration with which they’re spaced out- i.e, there are periods of time where farmers are more likely to pick up calls, irrespective of what time the first call was made to them.
- It is essential to design message-based interventions around the work schedules of the target respondents. Calls at 12:30 AM, 5:30 PM, 7.30 PM and 8:30 PM have the highest chance of being answered by farmers — likely matching the farmers’ work schedules, which start early in the morning and include breaks in the late morning, afternoon and night. Unfortunately, we were not permitted to make IVR calls before 9:00 AM, so we were unable to test the success of early morning calls. Without express permission from the Telecommunications Ministry, calls and SMS can only be delivered between 6am and 6pm
- With three callbacks (made 4, 24 and 48 hours after the first call), our final conversion rate was 32.2 percent. This suggests that it is possible to flip some non-respondents with low-cost call-backs, but that these returns diminish after 2–3 callback attempts.
- Farmers were highly unlikely to respond or interact with messages either keypress or voice responses. Anecdotal evidence from a separate field study suggest that farmers are unfamiliar with the format of recorded IVR calls, and are thus unlikely to understand instructions requiring response.
- An unexpectedly high number of farmers are registered on the “Do Not Disturb” database — of the 750 numbers we scheduled for calls during the third pilot, nearly 26 percent were registered on this call list, which prohibits broadcast messages and mass campaigns from directly contacting them without their written consent.
While some experience and lessons from this pilot can be used in designing and testing phone-based intervention for other sectors, the most variance in these results will come from changes in the target population, since farmers are a very specific sub-group, with unique daily rhythms and habits. Within the agriculture sector, these results are likely to vary by context (like mobile usage, geography and season), as the likelihood of picking up calls is presumably linked with the intensity of agricultural activity in a given time and space. Future research is required to see whether this result can be replicated under different agronomic conditions.
We have incorporated these results into the design of an ICT-based intervention, the impact of which will be assessed through an RCT later this year. We will be sending farmers non-interactive voice blasts at the most successful time slots. We plan to share the results from this RCT with the readers after the analysis is complete.