Humanitarian and development work can only be effective if it’s based on the actual needs and lived experiences of the communities it aims to serve. For this reason, structured evidence-gathering is central to how programmes are designed, prioritised and funded. Organisations therefore invest heavily in understanding local needs before and throughout programme implementation.
But in many of the places where this work is most needed, the voices of local people are often the hardest to hear. In these often-remote, under-resourced areas, the barriers of language, education and culture are compounded by low to non-existent access to the internet and modern information processing technologies — along with challenges like geographic isolation, armed conflict or low availability of funding.
If development programmes hope to reflect these communities’ priorities and lived experiences, the methods used to collect this information have to change.
Over the past two years, through my work at Fortell Impact — a company that leverages AI to boost the effectiveness of government and nonprofit agencies — I have collaborated with humanitarian and development organisations, NGOs, foundations and other impact-focused programmes operating in emerging economies around the world. In this role, I have used AI-enabled tools to support the delivery of needs assessments, community consultations, baseline studies and impact evaluations, facilitating data collection, transcription, translation and analysis in low-access, multilingual and low-connectivity contexts, while complementing the work of local teams on the ground.
This work has shown me how much some communities struggle to take part in efforts intended to understand their living conditions, priorities and constraints — and how AI can be utilised to help enhance these efforts.
Language is Just One of Several Barriers
One of the clearest factors shaping who is able to participate in these processes is, of course, language. In many of the contexts in which I have worked, needs assessments and evaluations prioritise a single, dominant or official language. This means that people who do not speak that language, or who rarely use it in their daily lives, are often excluded from these processes altogether.
Even when people understand the dominant language, it is often not the language they feel most comfortable using to explain their experiences. Describing priorities or concerns in a non-native language can be difficult, particularly when interviews rely on qualitative or open-ended questions. This is especially true in countries with high linguistic diversity, such as Kenya, where more than 30 local languages coexist.
During several studies we conducted with UNICEF in Mozambique to understand the impact of prolonged drought and the needs of people suffering from cholera, some conversations naturally switched between Portuguese and local Bantu languages, while others took place entirely in local languages, as not all participants spoke Portuguese. This added an extra layer of complexity for teams conducting qualitative interviews across different communities in the country. While interviewers were able to adapt to this reality, it often meant deploying interlocutors who could communicate in multiple local languages, or relying on community members who spoke Portuguese to support real-time translation during interviews. These arrangements made it possible to carry out the conversations, but they also introduced additional challenges at later stages of the process, particularly for transcription and translation. In this case, UNICEF relied on Fortell Impact to support the transcription and translation of multilingual interviews, as our AI-powered tools and processes enabled us to manage a level of linguistic complexity that would otherwise have required significant manual effort and specialised capacity.
Another major barrier is limited connectivity or difficult geographical access to the area. In places like Haiti and Somalia, many families and communities supported by ADRA Canada (Adventist Development and Relief Agency), one of our clients, live in small, remote settlements with no internet access and limited infrastructure. Conducting needs assessments and impact measurement in these communities typically requires training and deploying teams of local interviewers responsible for administering surveys and conducting in-person interviews, often across dispersed locations. This involves repeated travel and significant logistical and security planning, all of which increase costs and limit how many people can realistically be included in an assessment.
In this context, ADRA partnered with Fortell Impact to pilot an alternative approach to monitoring and evaluation in its food assistance programmes. Instead of training and sending teams of interviewers to reach hundreds of beneficiaries, we used AI-enabled avatars to conduct interviews offline in common local languages spoken in Somalia and Haiti. Interviews took place on local devices without requiring continuous internet connectivity, with data uploaded only once a connection became available. This reduced the need for in-person enumerators and allowed large numbers of participants to be reached without the time, cost and security risks associated with field deployment. As ADRA noted in its internal review, participants responded positively to the avatars, which required minimal support from staff and offered clear potential for reducing operational costs and risks in settings where in-person data collection is logistically challenging.
However, the challenge of gathering meaningful information in these contexts is not only linguistic or geographic. It also lies in how difficult it can be for people to speak openly about sensitive aspects of their lives. Topics related to health, including HIV, or experiences such as gender-based violence or informal work, often involve fear of judgement, stigma or potential repercussions — particularly when respondents are asked to discuss them with health workers, enumerators or other authority-linked figures.
This dynamic became clear during our collaboration with Jhpiego in Malawi, which aimed to capture client feedback following voluntary medical male circumcision, a highly sensitive health intervention linked to HIV prevention. By using a Chichewa-speaking AI avatar instead of a human interviewer, we enabled private, one-on-one conversations in which clients reported feeling comfortable and able to speak freely. They were more willing to speak to the avatar because it can’t judge them, whether for the clothes they’re wearing, how they speak, or their social status and lived reality. The resulting feedback showed strong consistency and completeness when compared with independent human analysis, suggesting that this approach can reliably capture sensitive experiences while reducing the barriers that often prevent people from speaking openly.
A further constraint frequently faced by humanitarian and development organisations relates to scale and timing. Even when language, access and sensitivity are addressed, it is often not feasible to engage the number of people required to produce timely and representative insights. Reaching large populations through in-person interviews typically requires significant human and financial resources, extended fieldwork timelines and complex coordination. As a result, organisations may limit sample sizes or rely on written or highly structured tools that prioritise speed over depth, often at the expense of qualitative insight.
This challenge emerged in a separate collaboration we conducted with UNICEF in Mozambique, which focused on understanding the challenges faced by children and adolescents across the country. Rather than deploying large teams of interviewers or relying on written questionnaires, we engaged more than 2,300 people in less than a week through AI-enabled avatars. Participants were able to speak openly and describe the challenges they faced on their own terms, without strict time limits or predefined response options. We completed the data processing and analysis within 72 hours, allowing UNICEF to review these insights quickly and use them to inform programme planning.
How AI Tools Can Help development organisations Understand Community Needs
AI is starting to fundamentally change how humanitarian and development organisations plan and scale their work. In the projects we’ve managed, Fortell Impact’s role hasn’t been to replace local teams, but to back them up. We are using these tools to handle the heavy lifting, such as transcription, translation, analysis and validation, so that the actual interpretation stays where it belongs: in the hands of the people who know these communities best.
One of the biggest shifts here is simply meeting people where they are. With the use of voice-based interfaces and conversational tools available on basic mobile phones, we are finally moving past the need for massive interview teams or paper-heavy surveys. This is a game-changer for community participation: It means we can actually hear from the people who are usually left out of the conversation, and ensure that their real-world experiences reach key decision-makers at the organisations that aim to serve them.
Of course, none of this is a magic solution to the deep, structural problems in humanitarian work. Technology cannot fix a broken road, stop a conflict or erase political gatekeeping. AI-powered communications technologies offer a way to widen the door when traditional methods fail, but they are no substitute for long-term, direct engagement with the realities these communities face.
In the end, it is not really about the tool itself. The real question is whether the methods we choose to engage with people in the world’s toughest contexts allow them to be heard in a way that is respectful, meaningful and true to their lives.
Talía Jiménez Romero is Head of Partnerships at Fortell Impact, and works on community voice, impact evaluation and context-adapted programme approaches across Africa, Asia, Latin America, the Middle East and Europe.
Photo credit: ArtemisDiana
Publisher: Source link
