Role: Research Lead
Methods: Field Research, Interviews, Focus Groups, Participatory Design, Thematic Analysis
Timeline: 2 weeks
Collaborators: Child Development Researchers, Home Visitors, Community Stakeholders, Non-profit organization for early childhood development

I explored how conversational AI could support social-emotional assessments for young children in low-income South African communities. Through field research and stakeholder engagement with home visitors, we identified culturally grounded design considerations for accessible, trustworthy, and low-burden assessment tools.

The Challenge:

Early-childhood development programs in underserved South African communities often rely on home visitors to assess children’s social-emotional well-being. Existing assessment approaches can be:

• time-intensive
• difficult to standardize
• linguistically challenging
• inaccessible in low-resource environments

The team wanted to explore whether a conversational AI tool integrated through WhatsApp could support more accessible and scalable assessments without disrupting community trust or existing care practices. A major challenge in product development is that many systems are designed without accounting for cultural nuance, low-resource environments, or existing human workflows. This research helped identify how conversational AI systems could better align with real-world community care practices rather than impose externally designed interactions.

Research Process

Phase 1: Understanding the Context

Conducted focus groups to understand how home visitors assess children’s well-being and how technology fits into their existing workflows.

Key goals:

• identify pain points
• understand communication practices
• evaluate digital accessibility constraints
• uncover trust and adoption concerns

Phase 2: Evaluating Opportunities for Conversational AI

Explored how a WhatsApp-based conversational agent could support assessment workflows while minimizing additional burden on caregivers and home visitors.

Participants evaluated a sample Whatsapp chatbot paper prototype for:

• conversational tone
• language clarity
• workflow fit
• accessibility
• trustworthiness
• perceived usefulness

Synthesizing Insights

Key Findings

1. Trust Was More Important Than Automation

Participants were hesitant about fully automated interactions and emphasized the importance of preserving human relationships within care workflows.

2. Conversational Simplicity Improved Accessibility

Users preferred:

• short prompts
• familiar language
• lightweight interactions
• low cognitive burden

3. Existing Communication Habits Matter

Because WhatsApp was already widely used, participants felt more comfortable engaging with conversational tools within familiar platforms rather than standalone applications.

4. AI Systems Must Adapt to Local Contexts

The study revealed that culturally grounded communication styles and local caregiving practices significantly shaped perceptions of usability and trust.

Design Recommendations

Designing conversational systems in low-resource community health settings should:

• support multilingual and culturally responsive communication
• minimize cognitive and technical burden
• preserve human oversight and relationship-building
• integrate into existing communication ecosystems

Impact

Western conversational UX frameworks often prioritize efficiency, standardization, and task completion. This project demonstrated that environmental context, relational trust, and communal care practices were equally critical to system adoption and meaningful engagement.

Reflection

This project challenged many conventional assumptions about conversational UX design.

A major takeaway was that successful AI interactions are not solely determined by usability efficiency or automation sophistication. In this context, participants valued relational qualities — empathy, trust, familiarity, and human connection — as core components of meaningful interaction.

The Ubuntu-centered caregiving practices observed during the research reinforced the importance of designing AI systems that support collective care relationships rather than replace them.

This experience fundamentally shaped how I approach human-centered AI research, particularly in culturally diverse and community-based settings.

Draper, C. E., Cook, C. J., Ankrah, E. A., Beltran, J. A., Cibrian, F. L., Lakes, K. D., … & Hayes, G. R. (2025). Feasibility and acceptability of the Mazi Umntanakho digital tool in South African settings: a qualitative evaluation. Infant and Child Development, e2567.https://onlinelibrary.wiley.com/doi/pdf/10.1002/icd.2567

Draper, C. E., Cook, C. J., Ankrah, E. A., Beltran, J. A., Cibrian, F. L., Johnson, J., … & Hayes, G. R. (2024). Young children’s mental well-being in low-income South African settings: A qualitative study. Journal of Child and Family Studies, 1-17. https://link.springer.com/article/10.1007/s10826-024-02929-5

Beltran, J. A., Mofid, H., Williams, L., Ankrah, E., Johnson, J., Cook, C., … & Hayes, G. R. (2023, October). Mazi Umntanakho “Know Your Child”: An Accessible Social-Emotional Assessment Tool for Children in Low-Income South African Communities. In Adjunct Proceedings of the 2023 ACM International Joint Conference on Pervasive and Ubiquitous Computing & the 2023 ACM International Symposium on Wearable Computing (pp. 30-34). https://scholar.archive.org/work/yorxwzl6azegtd5boiobvzft7a/access/wayback/https://dl.acm.org/doi/pdf/10.1145/3594739.3610685