World Bank's AI Trial Reveals Gaps in Poverty Targeting in Bangladesh

In a recent evaluation conducted by the World Bank, a trial focusing on AI-based poverty targeting in Bangladesh has illuminated significant challenges in leveraging modern technology for socio-economic initiatives. The trial, which aimed to identify the poorest households for a cash-transfer program, compared AI methods utilizing mobile phone data against traditional door-to-door surveys. The findings revealed a concerning performance of the AI system, identifying only 32% of the intended poor households relative to the 52% identified through conventional surveying techniques. This emerging scenario not only raises questions about the performance of AI in sensitive applications but also presents compelling procurement prospects for contractors adept in AI governance and digital service provision.

The trial included researchers conducting in-person surveys of 5,000 households to establish poverty levels, subsequently developing two distinct models to evaluate 106,000 potential beneficiaries. One model employed traditional proxy means testing, focusing on asset ownership, while the alternative model relied on extensive data drawn from mobile records—over 1,500 data points. Although AI has demonstrated promising results in sectors such as healthcare and resource management, its performance in this context emphasizes the necessity for robust validation against localized datasets before its widespread application.

The World Bank report underscored not only the inadequacy of the AI model for poverty targeting but also hinted at an overarching responsibility for rigorous testing of AI methodologies relative to the socio-economic context they are intended to serve. The evaluation noted that 6% of households participated could not be matched with mobile phone records, suggesting significant gaps in data access and integration. Moreover, only 45% of recipients believed that the phone-based targeting approach was fair—a critical insight into public perception and acceptance of AI systems in governance.

Despite these challenges in poverty targeting, the report highlighted more successful applications of AI in Bangladesh. For instance, AI-assisted screening for diabetic retinopathy resulted in a 39.5% increase in patients screened at local clinics, demonstrating that when properly implemented, AI can significantly enhance healthcare delivery outcomes. This divergence of results between different applications of AI points to an essential takeaway for stakeholders: while the technology holds considerable potential, its efficacy is highly contingent on contextual maturity and infrastructure.

As the Government of Bangladesh advances its digital transformation agenda, particularly through initiatives spearheaded by the ICT Division and the Aspire to Innovate programme, procurement professionals should consider the existing gaps and limitations highlighted by the World Bank evaluation. There is a clear and present opportunity for contractors specializing in AI policy design, governance frameworks, and holistic digital public service solutions. As the draft National AI Policy for 2026-2030 unfolds, aimed at establishing a National Data Governance Authority and other pivotal regulatory bodies, the demand for expertise in these areas is expected to surge.

This trial exemplifies how critical it is to merge technological innovation with appropriate governance structures to ensure the responsible implementation of AI in public sector applications. As Bangladesh continues to develop its AI capabilities, contractors with the right qualifications stand to benefit from increased investment and participation in shaping the future of digital governance.

• The Government of Bangladesh is actively seeking to enhance its AI governance framework as part of digital transformation.
• The trial identified that only 32% of the poorest households were recognized by AI compared to 52% by traditional methods.
• Only 45% of recipients found the AI-targeting approach to be fair, shedding light on public sentiment towards AI.
• As AI evolves, procurement opportunities arise for companies specializing in developing governance frameworks and digital solutions.
• The World Bank suggests AI technology must be thoroughly vetted against local data.
• AI applications have significantly improved healthcare screening in Bangladesh, showing AI's potential in diverse sectors.
• Challenges seen in the AI approach may spur demand for integrated solutions that combine AI and traditional methodologies.