Name in original language
Cali ResuelVE Biblioteca de Conocimiento
Initiative overview
Santiago de Cali, Colombia’s third largest city with approximately 2.5 million inhabitants, faces persistent challenges in reducing delays and inconsistencies in citizen attention. The City’s One-Stop Window operates across in-person, phone, and virtual channels, managing around 1,000 daily requests related to 218 procedures and administrative processes. Fragmented information sources and lack of uniform guidance have historically generated delays, confusion, and reduced trust in public institutions.
To address this issue, the City—through the Department of Administrative Development and Institutional Innovation (DADII), with support from the IDB’s Knowledge for Results (K4R) initiative—implemented the Cali ResuelVE Knowledge Library. The initiative consists of a virtual assistant powered by Anthropic’s Claude Sonnet language model. In its first phase, the tool supports public servants working across all citizen attention channels, providing immediate access to consistent and updated official information on procedures and administrative requirements.
The system uses artificial intelligence to understand questions posed in natural language and generate relevant, context-aware responses in real time. Its outputs are grounded exclusively in verified institutional data, ensuring accuracy and reliability. By design, it is not fully generative, preserving human oversight and maintaining the central role of frontline officials in citizen interaction. This creates an augmented experience that combines technological efficiency with human judgement and empathy.
The initiative aims to reduce response times, improve consistency of information, and strengthen institutional knowledge management. By enhancing the capacity of public officials, it contributes to more transparent and accessible government processes. Looking ahead, the model has potential to expand within the municipal administration and evolve into a broader digital tool accessible directly to citizens, offering a scalable and trustworthy example of AI adoption in subnational government.
Results, outcomes and impacts
During a pilot conducted while improving the library’s information base, processing times decreased by 25%, error rates fell by 18%, and internal user satisfaction rose from 72% to 89%. Results were measured through pre–post analysis based on internal surveys. To rigorously assess impact, we will compare two entities with identical payment plan procedures: one using the chatbot and one following standard processes, establishing a control group to isolate efficiency gains. Future impacts—efficiency, service reliability, and scalability—will be measured using surveys and administrative data.
