Transparency in the Use of AI

1. What We Use AI For

• Structuring informal information (documents, images, operational records) into verifiable fields.
• Detecting inconsistencies between sources and flagging points that require human review.
• Estimating income capacity and stability with ranges and confidence levels, not absolute figures.
• Generating readable explanations that accompany each signal with its source evidence.

2. What It Does Not Do

• It does not approve or reject credit autonomously.
• It does not replace the institution's risk policies, limits or committees.
• It does not issue signals without traceable evidence to support them.
• It does not use prohibited attributes or their direct proxies as decision variables.

3. Explainability

Every result includes the evidence used, its verification status, the relative weight of the factors and the reasons for any alerts. The institution can reconstruct why a signal took a given value at a specific point in time.

4. Human Oversight

The design assumes an analyst in the loop. Cases with low confidence, contradictions or insufficient evidence are routed to manual review before being added to the file.

5. Bias and Quality

• Periodic performance evaluation by segment, geography and product type.
• Monitoring of data drift and degradation of extraction quality.
• Logging of model versions and rules applied in each evaluation.

6. Responsibility for the Decision

The credit decision, its communication to the applicant, and regulatory compliance rest with the financial institution. See also Security and Data and the Terms of Service.