not-registered Login to view full entry:

Inbound engagement: remote training

IND -20 -1415

    Basic Information

  • Experiment rationale and/or abstract
    PxD operates a free agriculture information service delivered over mobile phones, in collaboration with a State Government Department of Agriculture in India using a two-way IVR platform with "outbound" push calls and "inbound" hotline service.

    This experiment builds on previous rounds of A/B tests that seek to increase engagement with the inbound service. In this experiment, we extend phone-based remote training to a subset of farmers who received reminder and instructional messages about using the inbound hotline. We explore the impact of training on driving engagement, and in particular on asking a valid question, when compared with a pure control group or a group which only received reminders.
  • Status
    Completed
  • Start date
    Q4 Oct 2020
  • End date
    Q1 Jan 2021
  • Experiment Location
    India
  • Partner Organization
    _N/A
  • Agricultural season
    Kharif, Rabi
  • Research Design

  • Experiment type
    A/B test
  • Sample frame / target population
    Farmers on the service
  • Sample size
    8,502
  • Outcome type
    Information access, Service engagement
  • Method of Measurement
    Platform administrative data
  • Research question(s)/hypotheses
    Is a combined treatment of reminder messages and remote training effective in increasing the number of farmers that ask a valid question? What type of user is the training most effective for?
  • Research theme
    Communication technology, Service design
  • Research design notes

    There are three rounds of these inbound tests, the first two revolved around reminder messages. In Round 1 we sent reminders to roughly 4000 farmers. In Round 2 we sent a series of reminders to 15,914 farmers. 11,419 of these farmers were in the treatment group and 4,000 are in the control group and did not receive any messages.

    In this third round, we build on Round 2 and randomly assign the treatment group farmers to one of three new treatment arms (including the training arm), stratifying along original treatment status (i.e. the type of reminder messages that farmers received), geography, and their engagement with the inbound service during the 10 weeks of the experiment (i.e. (i) if they didn’t call in at all (ii) if they called in but were unsuccessful (blank calls, calls under 60 seconds and invalid questions) (iii) if they called and were successful (feature accessed for over 60 seconds or valid question recorded).

    The treatment arms are:
    1. Control (T0): Farmers receive no treatment
    2. Pest (T1): Farmers receive a series of reminder messages + two additional “use-case” messages that encourage farmers to record a question about pests
    3. Training (T2): Farmers receive a series of reminder messages + remote training.
    4. Reminder only (T3): Farmers only received a set of reminder and instructional messages in the first stage

    Note that we define a valid question as a farmer asking a query related to agriculture.

  • Results

  • Results
    We find that offering remote training to farmers in addition to reminders increases the likelihood of calling in by 1.3 percentage points. This is a 53 percent increase over the control group mean likelihood, which is 2.5 percent. We also find that 2.2 percent of this group ask a valid question compared with 0.1 percent in the control group. However the absolute numbers here are still low and hence we weren't able to assess the statistical significance.

    In terms of heterogeneity, we also found that the training was more impactful for certain groups, including:
    • Farmers who are active on the outbound service: Those who have an above median outbound listening rate in the 6 months before the original experiment call in at a higher rate (3% of farmers asked a question, as compared to 1.6% of the farmers who had a below median listening rate).

    • Smartphone users: A larger proportion of smartphone users in the Training group called in as compared with non-smartphone users (3% versus 1.8%).