AI agent training for underwriters in consumer lending in Brazil

Brazilian consumer lending underwriters should be trained to supervise agents that assemble credit files from open finance and credit registry data, while disbursement by Pix and every decline stay behind held approval. Training covers depth, provenance, the LGPD's right to review automated decisions, Banco Central do Brasil expectations, and protection of pensioners in payroll-deductible lending.

What does a consumer lending underwriter in Brazil handle?

Brazil's consumer credit market has distinctive products. Payroll-deductible loans, repaid directly from salaries or pensions, carry lower rates and attract heavy sales to retirees, along with complaints about loans contracted without real consent. Personal loans, credit cards and instalment purchases make up the rest, much of it originated through apps.

Underwriters read data from the positive credit registry, bank transaction data shared through the open finance regime with the customer's consent, and documents uploaded in the app. They apply policy, check affordability and over-indebtedness signals, approve or refer, and review fraud flags. Disbursement goes out by Pix in seconds.

Where do agents help, and at what depth?

1. Credit file assembly, depth 1 to 2. The agent pulls registry data and open finance transactions, categorises income and commitments, and drafts the affordability summary. Data received through open finance is signed by the sending institution; a screenshot of a payslip is declared.
2. Payroll loan consent checks, depth 2. For payroll-deductible loans, the agent checks that the consent record, identity verification and margin calculation are complete and consistent. Applications from pensioners with signs of third-party pressure, such as a phone number or device shared across many applications, are held.
3. Fraud signal triage, depth 2. The agent compares device, document and behaviour data against known patterns and drafts a risk note.
4. Policy approval for small loans, depth 3. Where the lender allows it, a certified agent approves small, clean applications within every policy rule.
5. Disbursement by Pix, depth 4. Held for a named approver above the lender's own limit. Once a Pix transfer settles, the money is gone.

What must underwriters learn, and what stays with them?

Reviewing an agent's decision is the core skill, and how underwriters review an agent's decisions sets out the routine. In Brazil, the first check is always provenance: which figures came from open finance or the registry, and which rest only on what the applicant uploaded. The second is the bound: was the application allowed because it met policy, or held because the agent's certificate does not cover that product?

Underwriters also need to understand the review right. Under the LGPD, a borrower can ask for a review of a decision made solely by automated processing. If the agent made the decision, a person must be ready to review it properly, with the record of what the agent saw.

Approvals need structure. An approver authority matrix for the lending team sets who may release which held actions, up to what amount, and a reviewer rota for the lending team keeps held disbursements from waiting overnight.

Underwriters keep:

• declines and policy exceptions;
• payroll loans to pensioners where consent is in doubt;
• reviews requested under the LGPD;
• fraud conclusions.

Which Brazilian rules shape oversight and competence?

Banco Central do Brasil regulates and supervises banks, credit fintechs and payment institutions. Its rules on credit risk management, operational risk, customer treatment, open finance and Pix all reach agents that assemble or decide credit. Lenders should be able to show the central bank which agents act in credit, at what depth, and who oversees them.

The LGPD governs personal data and is enforced by the ANPD. It requires a legal basis for processing, respect for purpose and necessity, and transparency. It gives individuals the right to request review of decisions taken solely on automated processing that affect their interests, including credit decisions, and to receive information about the criteria used. Credit protection is a recognised legal basis, but it does not remove the transparency duties.

Brazil also has rules on preventing and treating consumer over-indebtedness, which expect lenders to assess repayment capacity and avoid aggressive offers, particularly to elderly and vulnerable borrowers. Load those requirements into the agent's sources, and keep judgement on vulnerable cases with people. Pix's own rules on fraud reporting and refund mechanisms help after a mistaken payment, but they are not a substitute for the hold before it.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

A 90-day training plan

First 30 days. Foundations for underwriters, credit managers and fraud analysts. Inventory agent capabilities by product, with depth and certificate lapse dates. Replay a month of payroll loan applications at depth 1 and list every case where consent evidence was incomplete. Pryme Intelligence Academy's Foundations track is free and set in lending, which gives the team a shared baseline.

Days 31 to 60. The agent drafts files at depth 2 on live applications. Underwriters correct with reasons. Write the lender's own limits for Pix disbursement and the approver matrix. Approvers sit the operator exam.

Days 61 to 90. Two-week practicum of logged review decisions, then calibration on twenty held payroll loans and ten LGPD review requests. Decide from the evidence whether small-loan approval should move to depth 3. The full role map is in AI agent skills by role, and which roles in consumer lending need training covers the wider team.