Using technology and complexity in evidence generation

SBCC practitioners have long thought about how their communications and other tools can drive broad social change in the context of complex social systems.  Existing frameworks the Socio-Ecological Model help structure that understanding, and guide us to more effective approaches.

In this talk, we'll briefly introduce three newer approaches that practitioners are using to understand the complex interactions between communities, the institutions they participate with (or oppose), and the SBCC initiatives.

We'll start with behavioral system maps: a qualitative tool to visualize complex systems in the context of behavior change efforts. The approach is deeply participatory and best done hand in hand with community members. 
Second, we'll look at individual level modeling (micro-simulation, agent-based modeling , etc.): and how it can provide much more nuanced insight, but at the cost of community participation and with a high technical burden.
We'll conclude with a more detailed look at a recent technique to combine behavioral insights with simulation model and AI: which allows for non-technical participants to co-create, interact with, and gain insights into the dynamic of their community or other systems of interest. 
By using interpretable and transparent AI modeling, this approach clarifies how behavior change interventions ripple across sectors, how outcomes shift over time, and where unintended consequences may arise. This approach redistributes power by embedding local knowledge at the center of computational models, providing direct access to non-technical stakeholders, and grounding insights in equity, accountability, and our deep interconnections as people and as communities.

Social and Behavior Change Communication (SBCC) depends on timely, high-quality behavioral data, yet traditional data collection methods are often slow, costly, and difficult to scale. AI-powered voice interviewers offer a promising pathway for rapid, large-scale behavioral insight generation but their performance, reach, and equity implications remain underexplored.
This presentation introduces a two-part study in Kenya designed to rigorously test and compare their efficacy.
Part 1 focuses on the uptake and performance of an AI voice interviewer across multiple recruitment channels, namely random digit dialing, a local research panel, SMS, WhatsApp, and digital advertisements. The AI voice interviewer will engage caregivers of children under five in 10-minute AI-led interviews about beliefs and decisions around childhood vaccination (target n=1,000). We will evaluate response completeness, adherence to protocol, and participant experience.
Part 2 directly compares data collected by AI voice agents (n=400) and human enumerators (n=400) from the same population. We investigate key methodological questions: 
How do measures of data quality, response honesty, and participant comfort differ between methods? 
Does the "interviewer" (human vs. AI) influence the disclosure of sensitive information? 
By moving beyond a simple "AI vs. human" debate, we aim to provide the SBCC field with nuanced evidence on the trade-offs, opportunities, and guardrails needed for these emerging technologies. Our goal is to foster a critical discussion on how innovations in knowledge collection can be responsibly integrated into SBCC research and practice.

Reliable data are central to effective social and behaviour change communication (SBCC) and family planning (FP) program management. Yet, in Nigeria, routine health information systems often face inconsistencies and weak validation mechanisms that limit the use of evidence in decision-making. The Challenge Initiative (TCI) developed the Family Planning e-Data Quality Assessment (e-DQA) Tool to improve data integrity and strengthen accountability at the health facility level. Built on the Kobo Collect mobile platform, the e-DQA Tool enables real-time data verification, automated quality checks, and on-the-job coaching for FP providers and M&E officers.
Implemented across 427 TCI-supported health facilities in 10 Nigerian states, the tool transformed traditional data supervision into an interactive, learning-based process. Findings from two assessment rounds (2023–2024) revealed measurable improvements in data availability, validity, and consistency across most states. The participatory digital coaching process improved documentation accuracy fostered local ownership, and empowered health workers to identify and address reporting gaps autonomously.
By reframing data validation as a communication and learning activity rather than a compliance task, TCI's e-DQA model created a behavioral shift in how health data were perceived and used. The tool proved to be a creative solution for enhancing understanding of data gaps within the facility. This experience demonstrates that digital innovations, when embedded in coaching and feedback systems, can build sustainable cultures of data-driven decision-making. The interactive nature of the tool fostered collaboration, allowing for shared learning experiences to make informed decisions based on accurate data.

Mozambique faces persistent public health challenges, with a child mortality rate of 60 per 1,000 live births in 2023 well above the 25 per 1,000 target for the Sustainable Development Goals, which is mostly from preventable diseases. Malaria, pneumonia, and diarrhea remain leading causes of death for children under five, largely due to inadequate access to and gaps in water, sanitation, and hygiene (WASH) practices. While infrastructure and resource gaps are significant, behavioral barriers such as habitual practices also limit WASH adoption. To address these, the Ministry of Health and UNICEF launched the "Model Families" initiative, recognizing families who adopt key health and sanitation behaviors. In 2023, UNICEF Mozambique partnered with Nudge Lebanon to accelerate Model Family certification in Sofala province using behavioral insights, focusing on handwashing stations, correct handwashing, latrine upgrades, and hygienic use.
The study used a difference in difference experimental design with treatment and control groups, collecting baseline and endline data from about 500 households, and integrating qualitative insights to deepen understanding of behavioural enablers and barriers. Results showed significant improvements in handwashing infrastructure and behaviors, with fixed tip-tap facilities rising from 13.8% to 71.2% and soap/ash availability from 9.3% to 47.8%. Correct use of latrine pit lids also improved, though overall infrastructure did not change significantly.
For sustainability, future programs should integrate low-cost materials with behaviorally informed communication, strengthen community ownership, and institutionalize successful approaches into national WASH policies, standardizing monitoring and leveraging pilot findings for scale-up.