Brazilian consumer lending credit analysts need to learn to supervise agents that gather and summarise credit evidence while the credit decision, fraud judgement and over-indebtedness checks stay with them. Training covers depth, verdicts, provenance and reviewing held decisions, shaped by LGPD, the ANPD and the Banco Central do Brasil. A 90-day plan makes it routine.
What does a credit analyst at a Brazilian consumer lender do all day?
The analyst works across personal loans, payroll-deductible loans, credit cards, store credit and vehicle finance. A typical file holds the applicant's bureau data, positive credit history, income evidence, open finance data shared with consent, and the lender's own history with the customer. The analyst reconciles all of it, checks policy and either approves, declines or refers.
Two pressures shape the job. Fraud is persistent, especially in payroll-deductible loans aimed at retirees and pensioners, where impersonation and pressure from third parties are common. And over-indebtedness rules require lenders to offer credit responsibly and to protect a minimum income for the borrower. Analysts are the people who see when a file does not add up.
How will agents change the assessment?
Certification on the lender's own files, including past payroll-deductible fraud cases, sets the agent's depth, up to its role's ceiling. A capability that has never been tested on fraud cases should not go beyond reading them.
| Use | Depth | What the analyst does |
|---|---|---|
| Assembling the file from bureau, positive history and open finance data | 1: reads and lists | Checks nothing is missing or stale |
| Income and commitment summary | 2: drafts | Tests each figure against its source |
| Payroll margin check for payroll-deductible loans | 2: drafts the calculation | Confirms the margin and the employer data |
| Fraud and impersonation signals | 1: flags only | Decides whether to stop, call or escalate |
| Decision letters and counter-offers | 2: drafts from the analyst's decision | Approves the wording before it goes |
Disbursement by Pix is a depth 4 capability, and in most consumer lenders it should be held for a named person after the analyst's approval, never released on the agent's own reading of the file.
How should analysts review an agent's work?
A review is more than agreeing with the conclusion. The approach in how credit analysts review an agent's decisions comes down to four checks:
- Read the verdict. Allowed, held, refused or withheld, and which of five bounds decided it: role scope, certificate, sources, the lender's limits or the platform guardrails.
- Check provenance. Bureau data and the lender's own ledger are strong sources; an income declaration is declared; a payslip arriving by email is external. An external document cannot by itself justify an approval.
- Test one figure end to end. Pick the income or the margin and trace it to the source line.
- Record a reason. Every release or rejection carries a sentence a colleague could understand.
When the analyst finds an error, the fix should become a banked correction in the agent's sources or brief. Teaching credit analysts to bank corrections sets out how.
What stays with the analyst?
The credit decision itself, any judgement about over-indebtedness, fraud calls, exceptions to policy, and conversations with applicants who seem to be under pressure from someone else. These need context and accountability an agent does not carry. The question of whether analysts are being replaced is answered in will agents replace credit analysts: the volume moves, the judgement does not.
Analysts also need to understand the agent's lifecycle. Its certificate lapses after 90 days, and a lapse or a run that clears no depth drops it to depth 1 the same day, at which point the file assembly agent reads but no longer drafts. Record who may approve what in an approver authority matrix for the lending team, and who is trained for it in an AI agent competency matrix for the credit team.
Which Brazilian rules shape oversight and literacy?
The Banco Central do Brasil supervises banks and other financial institutions, including how they manage operational risk and treat customers. Its expectations on responsible credit and conduct apply to decisions prepared by agents exactly as they apply to decisions prepared by people. Pix, the central bank's instant payment system, settles in seconds, which is why disbursement must be held for a person.
LGPD governs personal data, enforced by the ANPD. It requires a lawful basis and a specific purpose for processing, limits data to what is necessary, and gives people the right to request a review of decisions taken solely on automated processing that affect their interests, along with information about the criteria used. Analysts must be able to explain a decline in terms a customer can understand, and the record must show a person decided.
Consumer protection and over-indebtedness rules require clear information before contracting, responsible credit offers and protection of a minimum income. If the lender also serves customers in the EU, GDPR may apply.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
How to run the first 90 days
Month one: shared vocabulary. Every analyst and credit manager learns depth, the four verdicts, provenance classes and the record, using the lender's own anonymised files. Credit managers draft the approver matrix.
Month two: supervised review. Analysts review agent-drafted summaries on live files and calibrate weekly on a shared set, including at least one payroll-deductible case with fraud signals. Disagreements are written down and resolved.
Month three: rights tied to competence. Approval rights follow a current certificate. Pryme Intelligence Academy's Operator track, with its 40-question exam and two-week practicum on real decisions, gives an external standard. Rehearse a certificate lapse before month end. The wider map is in AI agent skills by role.
Questions people ask
Can an agent decline an application on its own?
It should not. Declines carry the right to review under LGPD and the reason must be explainable, so the agent drafts and a trained analyst decides.
How should analysts treat open finance data?
As data from a named financial institution, shared with the customer's consent, which is stronger than a self-declared income. It still needs checking for completeness and the period it covers.
What is the biggest fraud risk agents can help with?
Impersonation in payroll-deductible lending. The agent can flag mismatched contact details, new devices and unusual timing, but the decision to stop or proceed belongs to a person.