The
Disclosure
Effect

VMMC client satisfaction data is typically collected through brief exit forms administered by clinic staff, often in the facility where the procedure just took place, by representatives of the same organisation that performed it. The social dynamics of this setting — a patient recovering from an intimate medical procedure, surrounded by staff, in a context where future access to care may feel contingent on positive feedback — create powerful pressures against honest reporting.

Fortell Impact built and deployed the first Chichewa-speaking AI avatar for data collection in Malawi — a significant methodological choice. Chichewa is spoken by over 12 million people across Malawi and Zambia, yet is almost entirely absent from standard survey platforms. By meeting respondents in their own language, the avatar removed a layer of friction that conventional data collection rarely acknowledges: the cognitive and emotional cost of answering sensitive questions in a second language. People are more precise, more candid, and more willing to volunteer complexity when they are not simultaneously translating. The avatar was deployed across four data collection batches. It met clients in a private setting on their own device, without clinic staff present. The results show what changes when both the language and the power dynamics of data collection are changed at once.

The majority of respondents described high satisfaction with clinical care, professionalism, and post-operative education. This is consistent with what standard exit forms report. But the avatar captured something forms cannot: the texture of what a positive experience actually felt like, and, crucially, where it fell short.

The inter-rater reliability test conducted on the AI analysis found substantial agreement between AI coding and independent human reviewers (kappa = 0.75), confirming that the qualitative analysis was consistent and reproducible.

The 3-delays framework for maternal mortality identifies three critical points of failure: delay in deciding to seek care, delay in reaching care, and delay in receiving adequate care. It is a well-established analytical model. What it lacks is a mechanism for hearing all three delays described by the people who experience them, in their own words, across a large enough sample to be actionable at the programme level.

This project was commissioned by the Jesuit Justice and Ecology Network – Africa (JENA). Fortell Impact deployed its avatar platform across 31 hospitals in Kenya, Uganda, and Zambia, capturing the perspectives of 280 pregnant and postpartum mothers, 148 doctors and nurses, and 162 community health workers simultaneously. A critical design decision was language: for Kenya, the avatar was deployed in Swahili; for Uganda, in Luganda; and for Zambia, in Bemba and Nyanja. In each case, respondents were reached in their own first language — not English, not a lingua franca, but the language in which they think, worry, and describe their daily lives. The three-voice structure produced something rare: a picture of the same system as experienced simultaneously by those using it, delivering it, and supporting it from the community. The divergence between accounts was itself a finding.

Mothers described delays in their own terms — financial barriers, the need for a husband’s permission to seek care, transport failures, and fear of how they would be received — that do not appear in clinical records or facility utilisation data. Healthcare workers described a staffing and resource context that made quality care structurally difficult, framing it as a system problem rather than individual failure. Community health workers described the gap between what they knew and what families would act on, naming social and financial gatekeepers that clinical staff rarely encounter. Three groups. The same facilities. Three fundamentally different accounts of what was happening inside them.

The multi-language deployment (including local languages not serviced by standard survey platforms) was critical. Language barriers are themselves a form of the disclosure effect: people answer differently, more briefly, and less accurately in a language that is not their own. Deploying the avatar in each respondent's primary language — rather than defaulting to a widely spoken regional language — removed a systematic source of data degradation. The avatar poses questions and records spoken responses; when that exchange happens in someone's own language, response length increases, emotional specificity improves, and the gap between what people mean and what they say narrows.

IDB Lab is the innovation arm of the Inter-American Development Bank, focused on testing new approaches to development impact. Their challenge: how do you measure household economic vulnerability across a portfolio of early-stage investees in multiple Latin American countries, at a cost that does not consume the value you are trying to create? Traditional RCTs had cost IDB Lab between $200,000 and $300,000 per project. That model was not sustainable for portfolio-level measurement.

Fortell Impact co-designed the Economic Resilience Index (ERI) with IDB Lab — a framework measuring household vulnerability across income stability, savings, debt, and social support. The Colombia pilot deployed a female-voiced Spanish-language avatar with 177 respondents across low-income urban households. It was the first AI avatar-based vulnerability assessment of its kind.

The avatar was essential precisely because of what the ERI measures. Questions about debt, household conflict, income instability, and financial dependence are among the most sensitive a researcher can ask. In a human interviewer context, respondents routinely underreport on all of these — presenting a more stable picture than is accurate. The avatar, experienced as private and non-judgmental, consistently produced more candid responses on exactly the topics where social desirability bias is highest. It captured the texture of economic precarity that structured questionnaires are designed to miss.

A separate IDB Lab pilot produced a finding that changed how Fortell Impact designs every subsequent deployment: female avatars consistently produced longer, more candid responses than male avatars on sensitive topics including income, debt, and household tension. Avatar design — voice, perceived gender, language register — shapes disclosure in the same way that interviewer characteristics do in human-led research. How you build the avatar determines what you learn.

UNICEF Mozambique commissioned this national consultation with a specific purpose: to understand the challenges faced by children and young people in Mozambique, and to hear directly from them what UNICEF could do to support them better. It was an exercise in accountability to the people the organisation exists to serve — and the methodology needed to match that intent.

Over 2,300 community members participated — the majority children and young people, with some adults. They did not travel to a facility or sit with an enumerator. They participated from wherever they were: at home, in their community, in private. The avatar posed questions in Mozambican Portuguese — the language in which they live their lives — and recorded their spoken responses. No staff member was present. No form was handed back to an authority. The conditions that typically constrain what people say in consultations like this were removed.

What emerged was not a ranked list of issues but a portrait of compound crisis: poverty, hunger, early marriage, climate shocks, and violence all described as interacting and compounding over time. These were not separate problems requiring separate programmes. They were the same conditions, experienced from multiple directions at once, by the same people.

The finding that most directly challenged how consultations are typically used came from the young people themselves. A significant proportion were not only describing what was wrong. They were asking to be involved in the actual planning — not consulted once decisions were already made, but present when programmes are being designed. That is not a finding that a tick-box instrument produces. It requires a space where people feel free enough to say what they actually want, not just what they think they are being asked.

World Vision’s Empowered Worldview (EWV) programme is a faith-based curriculum designed to shift participants from a mindset of poverty and dependency to one of agency, dignity, and collective action. When Fortell Impact was brought in to evaluate the Burundi midline, the question was not just what the data would show — but whether the data collection methodology could do justice to the context.

Rural Burundi is not a context that standard survey platforms are designed for. Connectivity is limited. Kirundi — the language spoken by participants — is used by around 12 million people but is supported by almost no digital research infrastructure. And the evaluation needed to include children: not as subjects described by adult proxies, but as direct respondents sharing their own experience of the programme in their own words. Each of these requirements would have been a significant constraint for a conventional evaluation. The avatar approach resolved all three simultaneously.

The avatar posed questions and recorded spoken responses in Kirundi. Participants completed the survey on their own devices, in their own time, without a facilitator present. Children had a dedicated instrument adapted for their age — a space to describe what the programme had meant to them directly, rather than being represented by the adults around them. In a sector where children’s voices are routinely absent from programme evaluation — not because they are unimportant, but because gathering them systematically at scale is logistically difficult — this is worth noting.

The data showed substantial change across every dimension the programme sets out to influence. Adults described moving from isolation to mutual aid, from passivity to initiative. The qualitative interviews gave that change texture: widows building livelihoods alone, farmers running savings groups for the first time, people describing a renewed sense of purpose. The programme findings are World Vision’s to publish in full. What this project demonstrated is that this depth of evidence — from adults and children alike, in a minority language, in low-connectivity conditions — is now collectible at scale.