Cobranza como primer edge de IA: economía, riesgo y control
In financial services, almost every conversation about artificial intelligence starts in the wrong place: the most ambitious project, the most visible one, or the hardest to measure. Collections is the opposite. It’s a bounded, high-volume flow with clear rules and a result measured in pesos recovered.
That’s why it’s a good candidate to be an institution’s first edge: a critical workflow rebuilt with AI, in parallel to the current operation, until it proves it can replace it. This article explains why AI-powered collections is a good place to start, what the latest data says, what regulatory risk needs to be managed, and how to keep control.
What’s happening with delinquency in Mexico
Banco de México’s (Banxico) Financial Stability Report of June 10, 2026 notes that consumer loan delinquency rose slightly in banks and unregulated intermediaries, mainly in personal loans and auto loans, although it remains at historically low levels even when adjusted for write-offs.
The latest data from the CNBV (Mexico’s National Banking and Securities Commission), published by Expansión in August 2026, shows the other side: in June 2026, the adjusted delinquency rate, which adds past-due loans and loans written off over the last year, reached 11.31% in consumer lending and 13.70% in credit cards, levels not seen since 2021. That same month, bank cards in circulation hit a record high of more than 47 million.
The Mexican Banking Association (ABM) reads those numbers as a marginal adjustment due to greater credit penetration: commercial bank delinquency is at 3.3%, a level it considers under control, and its goal is to raise financing to the private sector from 38.8% to 45% of GDP by 2030.
All three readings are compatible. Against long-term history, delinquency remains low; against the last four years, it’s at its highest point; and credit will keep growing. More credit means more accounts to manage and constant pressure on the cost of collecting.
Where the economic opportunity is
The most recent McKinsey reference on generative AI in collections that we found is from June 2024. In its experience, organizations that deploy advanced generative AI capabilities in customer care and collections can achieve up to 40% lower operating expenses and around 10% higher recovery, with up to 30% higher customer satisfaction. These are results “in its experience” and with a ceiling, not a guarantee.
McKinsey identifies four fronts where that value is generated: reducing the demand for manual intervention (for example, tackling causes such as customers who forgot a payment because they didn’t receive their statement), analyzing all interactions instead of a sample, supporting the human agent in real time, and automating interactions. Among its examples, a consumer lending institution improved its team’s performance by 10% with call analytics, and a bank estimated up to 14% more productivity per agent with a copilot.
The investment context is also current. A Gartner survey of 199 service and support leaders, conducted between April and May 2026, found that AI spending rose 38% while total budgets for the function grew only 2%. Gartner warns that the challenge is making that investment produce measurable value, because enthusiasm alone doesn’t make a technology the right solution. This is a customer service data point in general, not just collections.
In the U.S. collections industry, TransUnion’s most recent report, summarized in February 2026, found that technology budgets are growing faster than other operational areas, that most organizations already adopt or are exploring AI and machine learning, and that payment portals, text messages, email, and chatbots are already common components. And in India, according to Business Standard, several private banks and financial companies already use virtual agents to remind customers of overdue payments and reserve a person’s visit for difficult cases, while remaining cautious about regulation.
These are data from other markets and from consultancies, useful for understanding the direction, not targets for a Mexican portfolio.
Why collections is a good first edge
A first edge needs to meet certain conditions. Collections meets almost all of them:
Bounded scope. McKinsey recommends starting with small, manageable steps, prioritizing high-value use cases in well-defined areas, building a minimum viable product, and scaling afterward. In collections, you can start with one stage, without touching the core.
A priority already defined. Among the function’s four priorities, McKinsey includes proactive outreach from the early stages of delinquency. That’s where you can start.
Measurable result. Every contact ends in a payment, a promise, a dispute, or silence. There’s a before and an after.
Clear rules. The conduct allowed in a collections effort is defined by the authority, as we’ll see in the next section.
Repetitive work. McKinsey notes that AI can automate routine processes, such as taking notes and summarizing interactions, and even some customer interactions, and that many customers prefer negotiating with a machine over explaining their difficulties to a person.
AI-powered collections: what an agent does and doesn’t do
An AI collections agent doesn’t replace the person who negotiates a difficult case. It absorbs the volume work so that human judgment arrives where it matters most. The capabilities of Mobiik’s Agentic Contact Center apply directly to this flow:
Automates high-frequency interactions: reminders and repetitive cases resolved with rules, context, and integration with systems.
Assists the human agent in real time: suggests responses, next steps, and customer context during the conversation.
Continuously evaluates conversations: measures quality, compliance, sentiment, and escalations.
Escalates with full context: when a case requires a person, it arrives with the history and classification already organized.
For this to work, the agent needs information that almost always lives scattered. That’s where the Operational Brain comes in, connecting operational data into a unified context.
The risk: collections regulation in Mexico
Collections is one of the most conduct-regulated financial activities. The Law for the Protection and Defense of Financial Services Users, whose current text includes a reform published in the Official Gazette of the Federation (DOF) on November 14, 2025, defines CONDUSEF’s powers to supervise financial institutions. And CONDUSEF’s provisions for collection agencies, which the commission itself continues to communicate in 2026, as the press recalled in April, establish, among other things, that:
Communications and visits may only take place from 7:00 a.m. to 10:00 p.m.
It is prohibited to threaten, insult, or intimidate the debtor, their relatives, their coworkers, or any person unrelated to the debt.
The agency must identify itself, and cannot use names resembling those of public institutions.
Financial entities must supervise their collection agencies and periodically report their conduct to CONDUSEF, and users can file complaints through the Collection Agencies Registry (REDECO).
McKinsey adds a design criterion: in collections, you need to manage the intensity and tone of messages and use analytics-based guardrails to avoid bias. The consequence is direct: an AI agent handling collections should be built to comply with these rules by design, not by good intention. A conduct error, at scale, stops being an isolated case, and legal and compliance review is part of the project, not a later formality.
Control: how governance is maintained
Gartner predicts that more than 40% of agentic AI projects will be canceled before the end of 2027, among other reasons due to inadequate risk controls. Forrester’s AEGIS framework starts from a similar idea: agents can’t be governed with the same controls as a traditional application. In collections, this translates into five design decisions:
Hard rules, not suggestions. Hours, message types, and contact limits coded as constraints the agent cannot bypass.
Defined escalation. Disputes, financial hardship cases, and any sign of conflict go to a person. McKinsey describes the same principle: when the technology doesn’t know what to answer, it escalates to a human agent.
Complete traceability. Every interaction recorded, with its outcome, so you can reconstruct what happened and why. Mobiik’s offering for financial services includes traceability and role-based permissions.
Continuous evaluation. Reviewing all conversations, not a sample, against conduct rules.
Deployment within the client’s environment, with data under their control.
How to measure a first collections edge
A first edge must be able to prove value with numbers. The indicators McKinsey uses to evaluate these initiatives serve as a base: recovery, operating expense, customer satisfaction, productivity, and handling time. It’s worth adding:
Payment promises generated and kept.
Recovery by delinquency stage, to see where the agent helps and where it doesn’t.
Cost per peso recovered, with a clear definition of what it includes.
Complaints and escalations, as a gauge of conduct quality.
The order matters: set the baseline with the current operation, run the agent in a bounded stage, and compare. Only when the numbers support the decision do you extend it to the next stage. That’s how replacement is earned, one stretch at a time.
Frequently asked questions
What is AI-powered collections?
It’s the use of artificial intelligence agents to manage collections tasks, such as reminders, clarifications, tracking payment promises, and escalating cases, with human oversight on decisions that require judgment.
What rules must collections comply with in Mexico?
CONDUSEF’s provisions define hours, treatment of the debtor, identification of the agency, and supervision by the financial entity. When integrating AI into any process, it’s wise to define from the start what the agent decides, what a person decides, and how its performance will be measured.
Where do you start?
With a bounded, measurable stage, usually early delinquency, with indicators agreed upon in advance and without touching the operational core.
The question worth asking
It’s not whether your institution should use AI in collections. It’s whether today you can say how much it costs you to recover each peso, at which stage of delinquency you lose the most accounts, and whether every contact your operation makes complies with the conduct rules.
At Mobiik, we design and operate AI agents with governance and traceability for financial services, integrated with your systems and deployed in your own technology environment. If you want to identify what your first edge would be, let’s talk.
