Bogota's oversight office deployed VigIA, a random-forest model (AUC 0.943) trained on 87,387 SECOP II contracts, to flag procurement at risk of cost overruns and delays -- winner of a CAF regional AI-for-government contest.
Veeduria Distrital de Bogota, with Universidad del Rosario's TicTank and CAF
Period
2021-2024
Keywords
anti-corruption, public procurement, oversight, machine learning
Context
In 2021 the Development Bank of Latin America (CAF) ran a regional call inviting public institutions across 11 Latin American countries to propose uses of data and AI for better governance, drawing 89 proposals from nearly 70 cities and municipalities. Veeduria Distrital, Bogota's district oversight and accountability office, won with a proposal to build an early-warning system for irregular or inefficient public contracts.
Activities
The resulting tool, VigIA, was built by Universidad del Rosario's TicTank with researchers from Universidad de los Andes and the Inter-American Development Bank, with CAF support. It combines logistic regression and random-forest classifiers with two purpose-built risk indices (IRIC and IRICP) to flag contracts likely to suffer cost overruns or delivery delays, using SECOP II procurement records together with sanction data from Colombia's Superintendencia de Industria y Comercio, and deliberately limits each model to 6-8 explainable variables so investigators can see why a contract was flagged.
Results
On a test set of 87,387 Bogota contracts from 2018-2020, the random-forest model reached an AUC of 0.943 and 87.47% accuracy for predicting cost overruns, and 0.936 AUC for predicting delivery delays. The methodology was peer-reviewed and published in Data & Policy in December 2024.
Conclusions
As of the published research, VigIA is a validated model with a public-facing dashboard inside Veeduria Distrital; no independent source yet documents its rollout to other Colombian cities, despite the CAF call being open to nearly 70 municipalities, or the number of investigations or sanctions it has directly triggered.
Implementation
Indicative cost
Medium (€50k–€500k) — Developed with CAF competition support plus academic partners (Universidad del Rosario, Universidad de los Andes, IDB); no operating budget for the dashboard or ongoing maintenance is disclosed.
Time to results
Medium (1–3 years) — Competition win in 2021, model development and validation through 2024, published December 2024; no confirmed rollout date beyond Bogota.
Staffing & skills
Veeduria Distrital de Bogota (host institution, public dashboard), Universidad del Rosario's TicTank, Universidad de los Andes and Inter-American Development Bank researchers (model development)
Conditions for success
Access to structured national procurement data (SECOP II) plus a linked sanctions database
Explainable models limited to 6-8 variables so investigators can interpret flags
External funding/competition (CAF regional call) providing both resources and a peer network of participating cities
Peer-reviewed publication of the methodology supporting external scrutiny and reuse
Common failure modes
No independent source documents rollout to other Colombian municipalities despite the originating competition being open to nearly 70 of them
Published results are model-validation metrics, not yet documented downstream outcomes such as sanctions issued or costs avoided
Public sources do not document funding for maintenance beyond the initial pilot
Where it fits
Governance type
municipal oversight/accountability office with academic and development-bank partners
Scale
city (Bogota)
Income level
upper-middle-income
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
