BETTO — Colombia’s AI system for targeting early childhood care to vulnerable children

BETTO (Bienestar, Eficiencia, Transparencia, Tecnología y Oportunidad) is an AI-powered targeting and procurement platform developed by Colombia's Instituto Colombiano de Bienestar Familiar (ICBF), with technical and financial support from the World Bank, the Inter-American Development Bank (IDB), UNICEF, Microsoft Colombia and Colombia Compra Eficiente. It launched in 2020 and operates nationally across all 32 Colombian departments.

BETTO addresses two linked challenges: identifying which children aged 0-5 lack access to ICBF early-childhood services, and selecting the best-qualified operators to deliver those services.

The system combines three AI methodologies: geo-referenced optimisation to target potential beneficiaries using 11 poverty and vulnerability variables, knowledge-structure reasoning to evaluate operator qualifications, and predictive modelling to forecast operator performance. Operators are ranked on 28 quality indicators covering territorial experience, operational and financial capacity, historical performance and sanctions history. A companion tool, ADA (Asistente de Adopciones), applies AI to streamline adoption administration.

By 30 June 2021, BETTO had reached 217,000 previously unserved vulnerable children, including 69,600 in rural areas and 14,600 from ethnic minority communities, across 1,138 contracts valued at COP 2.2 trillion. It processed 12,738 operator bids automatically, optimised 69,000 existing service units and created 32 new units in high-vulnerability areas, for a national total of 743,741 early-childhood care slots. ADA cut adoption-processing time from 24 months to 9 months and has facilitated over 3,480 adoptions.

Multi-stakeholder co-design involving government, multilateral institutions and the private sector appears to have accelerated quality and trust in the system. IDB's fAIrLAC network flags BETTO as a replicable Latin American model, though no confirmed cross-country deployment is documented, and no post-deployment algorithmic audit or formal redress mechanism is publicly documented.