Questions and answers
AI agents at a Mexican lender: the questions worth asking
If you run or evaluate a Mexican lender (a SOFOM, SOFIPO, IFPE or leasing company) and you’re looking at AI agents, these are the questions operations, risk and compliance leaders actually ask. Each one has a short answer and a page that goes deeper. Anything regulatory links to its official source.
Updated October 4, 2026
What can an agent actually do in each process?
- Can an AI agent run collections and stay within CONDUSEF rules? Yes, if it is built around the conduct rules: identify itself, contact only between 7 a.m. and 10 p.m., never pressure third parties, and log every attempt.
- How much of credit origination and underwriting can an agent handle? Almost everything up to the decision. Approval stays with a person or the credit committee.
- KYC and AML with agents: what can the agent do, and what can it not? It builds and checks files, screens lists and prepares alerts. Deciding and filing reports stays with the compliance officer.
- Can an agent handle customer service without clashing with the UNE? Yes. It resolves routine questions and routes formal complaints to the UNE with a case number and the full record.
How is it governed and controlled?
- How do you govern an AI agent inside a supervised institution? A named owner, rules enforced by the system, a full log, a kill switch, human takeover and regular reviews.
- What are the risks of an AI agent, and how are they controlled? Six main risks, each with a concrete control that lives in the system.
- What happens to my customers’ personal data? It stays theirs. Privacy notices, ARCO rights and the right to object to certain automated decisions all apply.
How do you roll it out?
- Can agents plug into my lending core without a migration? Yes: through APIs, MCP or whatever door each system has, with tightly scoped permissions.
- Where should a lender start? With a process that has volume, clear rules, reachable systems and a measurable outcome.
- How is an agent implemented, step by step? Six phases, from picking the process to running it with continuous review.
- Synthetic agent, chatbot or call center: which one fits? Agents for volume, people for judgment. A chatbot is enough when all you need is to point people in the right direction.
- What does each term mean? A plain-English glossary of the technology and the Mexican regulation.
Who are these answers for?
For the people who have to make the call, or answer for it: CEOs, COOs and the heads of credit, collections, risk and compliance at regulated and unregulated lenders in Mexico, plus the investors and partners who back them. A quick map of the acronyms: a SOFOM is a multi-purpose financial company (Mexico’s most common non-bank lender), a SOFIPO is a regulated savings and loan company, an IFPE is an e-money institution under the Fintech Law, the CNBV is the banking and securities regulator, and CONDUSEF is the financial consumer protection agency. Everything is written in business language, with no promises and no numbers we cannot back up. When we mention a rule, we link to the official source so your legal team can check it. None of this is legal advice.
Who writes them?
InnovaBlack. We build and run agentic ecosystems, with financial institutions as our core focus. We use what we sell: our own sales operation runs on agents that introduce themselves as AI agents, with people watching, stepping in and hitting the brakes when needed. Our ecosystems are built on Claude, Anthropic’s model.