AI agent training for credit analysts in retail banking in Spain

Spanish retail bank credit analysts need training for two agent uses above all: preparing mortgage files and generating pre-approved loan offers. Both are creditworthiness assessments, high-risk under the EU AI Act. Analysts must read each agent's depth and verdicts, verify sources and own the decision, under the Banco de España, the AEPD and GDPR.

What does a credit analyst at a Spanish retail bank work on?

Mortgages are the core: purchases of a main home, second homes, subrogations from other banks and novations of existing loans, mostly through the branch network and a growing share online. Analysts also handle personal loans, many of them pre-approved for existing customers on the strength of their account history, and credit cards.

A mortgage file includes payroll and tax returns, the applicant's account history, the Banco de España's central credit register, the appraisal from an approved valuer and the debt service ratio. Spanish mortgage law requires extensive pre-contract information and a notary's explanation to the borrower before signing, so the file must be clean and complete well before completion day.

What will agents prepare, and at what depth?

The agent is certified on the bank's own past mortgage and pre-approval files, and its depth never goes beyond the role's ceiling.

1. Mortgage file assembly, depth 1. The agent reads payroll, tax returns, the register and the appraisal, and lists gaps.
2. Debt service ratio and stress test, depth 2. The agent drafts the ratio at the bank's stressed rate, showing each input and source.
3. Pre-contract documentation, depth 2. The agent drafts the standard information sheet and warnings from the approved offer.
4. Pre-approved loan candidates, depth 2. The agent proposes which customers qualify and at what limit, under the bank's rules. A person approves the campaign list.
5. Drawdown at signing, depth 4, held. Funds move only on release by named people.

Why do pre-approved offers need special care?

Pre-approved loans are where an agent's output reaches thousands of customers without an analyst seeing each one. If the agent misreads account history, it can offer credit to people who should not get it, or deny the offer to people who should, at scale and without a visible error.

That makes the review different. Analysts should review the rules and the sample, not just individual cases: which inputs the agent used, how it treated irregular income, whether it excluded customers in arrears, and whether any group is consistently left out. They should test a sample of offers and non-offers against the source data every cycle. A pre-approved offer is still a creditworthiness assessment, and the oversight duties of a high-risk system apply.

What stays with the analyst?

The decision on each mortgage, its amount and conditions, exceptions to the debt service ratio or loan-to-value policy, judgement on an appraisal that looks out of line, and refinancing for borrowers in difficulty all stay with people. So does sign-off on each pre-approval campaign. What credit analysts should never let an agent do explains the reasons.

The review itself follows how credit analysts review an agent's decisions: read the verdict and the bound that decided it (role scope, certificate, sources, the bank's limits or the platform guardrails), check provenance, trace one figure to its source and record a reason. Provenance matters in a concrete way here: the central credit register and the bank's own ledger are strong sources; a tax return uploaded by the customer is declared; an appraisal forwarded by an intermediary is external until confirmed with the valuer.

Analysts should also know that an agent's certificate lapses after 90 days, and a run that clears no depth drops it to depth 1 the same day. Track who is trained for which capability in an AI agent competency matrix for the credit team. For how other roles fit, see AI agent skills by role.

Which Spanish and EU rules shape oversight and literacy?

Under the EU AI Act, creditworthiness assessment and credit scoring of natural persons are listed in Annex III as high-risk. The bank as deployer must meet Article 26, including human oversight by people with the competence and authority to exercise it, and Article 14 sets what that oversight must enable, such as understanding the output and overriding it. Article 27 requires a fundamental rights impact assessment for certain deployers. Article 4 requires measures to ensure sufficient AI literacy among staff.

GDPR, enforced by the AEPD, restricts decisions based solely on automated processing with significant effects under Article 22, and requires a data protection impact assessment under Article 35 where risk is high. Pre-approval campaigns sit close to that line. The Banco de España and, for significant banks, ECB banking supervision oversee credit risk, conduct and transparency. DORA applies from 17 January 2025 to the ICT risk behind these systems.

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

A 90-day plan, with a rota

Build a reviewer rota so a certified analyst is always available for held drawdowns on signing days. Record each person's progress in an AI agent training plan template for credit analysts.

Where the Academy fits

Pryme Intelligence Academy lets a bank make credit authority depend on a current certificate, depth by depth. Analysts take Foundations, then the Operator track, whose written exam needs 32 of 40 correct and whose practicum is two weeks of real review decisions. Builders who design the pre-approval agent can take the Builder track, which ends in a capstone agent certified in Attest.