MIRERC096/2025: Validating AI Voice Agent Methodologies for Behavioral Data Collection in Kenya: A Comparative Pilot on Vaccine Confidence Using Multi-Channel Recruitment

Abstract

This project aims to validate the feasibility, quality, and cost-effectiveness of AI voice agents as an
innovative method for behavioral data collection in Kenya. Traditional data-collection approaches in
low- and middle-income countries (LMICs) remain costly, slow, and limited in their ability to capture
nuanced behavioral drivers. These limitations restrict capacity to generate timely and representative
insights to inform public health programs, including vaccination campaigns. Recent advances in
conversational voice AI offer a promising alternative, but real-world evidence from LMIC settings is
lacking.
To address this gap, the project will conduct a comparative pilot study on vaccine uptake and
confidence, evaluating how AI voice–agents perform across different languages (English and Swahili)
and multiple recruitment channels: random digit dialing, local panels (Busara), social media
advertisements, and SMS/WhatsApp outreach. The pilot will also include a direct comparison with

trained human interviewers to assess relative data quality and participant experience.
The project is structured into three phases. Phase 1 focuses on system setup, multilingual AI voice
agent development, technical testing, and integration with Busara’s incentive platform. Phase 2
comprises two studies: (1) administering 1,000 AI-led interviews to evaluate recruitment-channel
performance, demographic reach, cost, and behavioral insights; and (2) a randomized comparison of
400 AI-led versus 400 human-led interviews, conducted via the Busara panel, to assess protocol
adherence, response quality, depth, participant trust, empathy, and overall satisfaction. Data will be
analyzed using quantitative, qualitative, and AI-assisted techniques, including external blind review of
a subset of study 2 transcripts by Harvard University. Phase 3 will disseminate findings through
peer-reviewed conference publications and presentations, and open-source tools.
The project will generate practical evidence on the operational efficiency, cost savings, and
representativeness of voice AI-based interviewing and uncover behavioral drivers of vaccine
confidence in Kenya among caregivers, as well as measles and pentavalent immunization uptake
among children under 5 years old. Outputs will include recruitment-channel performance reports,
behavioral insights dashboards, and a head-to-head AI-versus-human quality assessment. All project
costs will be covered by a Gates Foundation innovation grant. By modernizing behavioral data
collection and enabling richer, faster, and more inclusive insights, this project has the potential to
transform how public health data is gathered in Kenya and comparable LMIC settings.