What happened
Mercado Libre lends to small merchants on its e-commerce platform through Mercado Crédito. Its internal credit rating uses machine learning on platform data on sales, reputation and industry characteristics. Economists from the Bank for International Settlements, the Financial Stability Board and the Graduate Institute, Geneva, wrote the paper with a Mercado Libre employee. He disclosed his employment and investments in the company, and the data came from Mercado Libre. Using 2017 loan data from Argentina, the authors compared the internal rating with the local credit bureau's ratings. Loss rates ranged from 0.0% to 10.2% across the internal rating's groups and from 0.7% to 2.8% across the bureau's. In default regressions on 7,300 observations, the machine-learning score had an area under the ROC curve of 0.76. Bureau scores reached 0.64 alone and 0.68 with borrower characteristics. About 30% of the portfolio fell in the bureau's high-risk group, which had a 2.8% loss rate. The authors write that it remains to be verified whether the model predicts defaults better over a full business and financial cycle.
Sources
-
Peer-reviewed study, by the makers.
data from Mercado Libre show that credit models using machine learning and data from the e-commerce platform are better at predicting losses than traditional credit bureau ratings
Details
AI system
- Workflow
-
The score drew on platform data on transactions, reputation and industry characteristics. Mercado Libre lent through Mercado Crédito without asking merchants for further documents.
Follow-up
- Conditions for replication
-
Merchants must trade on the platform to generate the data that the score uses. The score's edge over bureau ratings has yet to be shown over a full business and financial cycle.
Entry details
- Sector
-
Small-business lending.
- Purpose
-
Mercado Libre built its internal credit score to assess merchants on its platform for Mercado Crédito loans from their transaction, reputation and industry data.
Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Mercado Libre credit score predicted defaults better than the bureau (AIBD-0865). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0865.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)