AI-Powered Vulnerability Screening for Portfolio Insight in Colombia
In partnership with IDB Lab, Fortell Impact piloted a new AI-enabled approach to assess household vulnerability in Colombia using avatar-driven interviews and automated analysis.
Overview
The project tested whether voice-based AI could support a more nuanced and operationally useful understanding of vulnerability beyond income alone, by integrating multiple social, structural, and household-level factors into a composite vulnerability framework.
Rather than treating vulnerability as a binary state, the pilot aimed to distinguish between households that are already in poverty, those that are vulnerable to falling into poverty, and those that are relatively resilient—with a view to improving programme targeting, intervention design, and future portfolio-level screening.
Why this mattered for a development finance context
For development finance institutions, impact funds, and social investment actors, vulnerability screening is a critical input into:
- programme targeting
- beneficiary segmentation
- risk assessment
- impact measurement
- allocation of support and social protection
Traditional income-based assessments often fail to capture the full picture of household resilience. They tend to overlook structural and social pressures—such as caregiving burdens, disability, informality, food insecurity, and exclusion from social protection—that can significantly increase the risk of future poverty even when current income appears stable.
IDB Lab and Fortell Impact set out to test whether a voice-based AI approach could deliver a more inclusive and multidimensional vulnerability screening method that remains practical for reuse across programmes, geographies, and investment contexts.
Background: Beyond income-only vulnerability metrics
Income remains the dominant lens for poverty and vulnerability assessment, yet it provides only a partial view of household resilience.
Building on academic literature and established tools such as the Household Vulnerability Assessment Tool (HVAT) and the Socioeconomic Vulnerability Index (SeVi), the project team designed a composite vulnerability framework that integrates income indicators with additional dimensions related to:
- continuity of access to basic services
- caregiving burdens for children, older people, and people with disabilities
- employment type and informality
- food security
- access to workplace benefits and social protection
- perceived security and stability
This approach makes it possible to distinguish not only between households that are currently poor, but also those that are at high risk of falling into poverty and those that are relatively resilient but still exposed to potential shocks.
Economic Resilience Index (ERI)
To operationalise this multidimensional approach, IDB Lab and Fortell Impact developed the Economic Resilience Index (ERI)—a composite vulnerability screening tool that categorises households into three levels based on a weighted score:
The ERI integrates income indicators with additional dimensions capturing:
- continuity of services
- care burden for children, older people, and people with disabilities
- perceived security and stability
Rather than relying on static poverty thresholds, the ERI provides a more dynamic classification of vulnerability that can be used to segment beneficiary populations, identify households at elevated risk, inform programme targeting, and support more differentiated intervention strategies.
Methodology
The pilot used Fortell Impact's voice-based AI platform to conduct household interviews through a regionalised digital avatar. Respondents engaged in structured, voice-led interviews covering:
- income and employment status
- household composition and dependents
- food security
- caregiving responsibilities and disability
- access to workplace benefits and social protection
- perceived financial stability and vulnerability
Key findings
The pilot surfaced several vulnerability dynamics that are often underrepresented in income-only assessments.
Care burden and household pressures
Over a quarter of respondents (26 percent) reported caring for an older person or someone with disabilities, adding to household pressures frequently overlooked in traditional vulnerability screening.
Income levels and food insecurity
Most respondents earned less than 1 million Colombian pesos per month, or between 1 million and 2 million pesos, placing the majority below or near the national minimum wage. Fifty-five percent of respondents reported that in the past seven days their household had to eat less preferred or cheaper foods due to lack of money or food, and 8 percent reported that this happened every day.
Employment informality and benefits
Although 80 percent of respondents had worked recently, most did so in informal employment, and only 22 percent had access to workplace benefits.
Ethnicity and structural vulnerability
Thirty-six percent of respondents identified as Indigenous, and 23 percent identified as Afro-Colombian, highlighting the intersection between ethnic identity and vulnerability exposure.
Overall vulnerability classification
When applying the ERI framework, the population received an overall score of 51, placing it in the category of vulnerable to income poverty. This indicated that while many households were not yet classified as income poor, a substantial share faced structural and social risks that could push them into poverty under adverse conditions.
What this demonstrates for financial actors
This pilot illustrates how voice-based AI can support a more operationally useful vulnerability screening model for development finance and impact investment contexts. Specifically, it demonstrates the feasibility of:
- moving beyond income-only poverty metrics
- integrating social and structural risk factors into vulnerability screening
- reaching populations that are difficult to survey using traditional methods
- transforming voice data into structured vulnerability indicators
- segmenting beneficiary populations by resilience level rather than income alone
For DFIs, impact funds, and social lenders, this type of vulnerability screening can inform beneficiary targeting and prioritisation, programme eligibility criteria, differentiation of support packages, monitoring of vulnerability exposure, and more nuanced impact measurement frameworks.
Why Fortell Impact's approach was different
Unlike traditional household surveys, this pilot used digital avatars to conduct voice-based interviews directly with respondents.
The conversational format reduced barriers related to literacy and technology access, while enabling respondents to engage in their own language without requiring smartphones or stable internet connectivity.
At the analytical level, Fortell Impact's automated transcription, translation, and coding pipeline made it possible to integrate large volumes of voice data into a structured vulnerability framework, while maintaining reproducibility and methodological consistency.
This combination of accessibility, automation, and analytical structure enabled the delivery of a multidimensional vulnerability screening model that is both inclusive and operationally scalable.
Implications for vulnerability screening and portfolio insight
The results confirm that income alone significantly underestimates household vulnerability and that structural and social factors play a critical role in shaping resilience.
The ERI framework demonstrates how vulnerability screening can move beyond static poverty thresholds toward a more dynamic and multidimensional model that is better suited for:
- programme targeting
- social protection planning
- beneficiary segmentation
- impact measurement
- portfolio-level monitoring
By capturing lived experience directly from households, the approach offers a more inclusive and human-centred foundation for vulnerability assessment in development finance contexts.
Looking ahead
Building on this pilot, IDB Lab and Fortell Impact plan to extend this method to additional contexts and explore its use as a repeatable vulnerability screening tool for programme targeting and portfolio-level monitoring.
The longer-term vision is to integrate voice-based vulnerability screening into regular monitoring workflows, enabling institutions to track changes in household resilience over time and generate structured vulnerability insights to support decision-making.
"One in four respondents cared for an older person or someone with disabilities, adding to household pressures often overlooked in traditional assessments."
"We believe AI has the potential to reveal a more complete and human picture of what vulnerability means and, in the future, to become a powerful tool to anticipate vulnerability even before it fully arrives."
Decision Impact
Results informed IDB Lab's approach to portfolio-level vulnerability screening and targeting.
Have a similar challenge?
We'd be happy to discuss how we might approach your specific context and requirements.