Digital lending in Mexico: market, fraud and decision automation

A market that went digital fast

The jump was huge: from one top lender with a fully digital experience to seven of ten in just a few years. But the digital channel also lowers friction for fraud. Per industry data, conversion and approval moved from 36% to 29%, and institutions ended up exposed to fraudsters anywhere in the world.

That creates the underlying tension. More applications come in, but with more noise, more fraud attempts, and less room to review case by case by hand.

Digital fraud grew and changed shape

Between 2021 and 2025, fraud attempts on digital channels multiplied 3.5x. Generative AI put deepfakes within anyone's reach, hard to catch by eye, and the sector started leaning on behavioral biometrics and advanced verification.

It is worth being clear about what uFlow does and does not do. The engine requests anti-fraud and identity signals from specialized providers, combines them, and decides with them inside the same flow. Biometrics and deepfake detection are handled by those providers; uFlow integrates them and turns them into part of the credit policy.

Alternative data: from declared to verifiable

Relying on what the applicant declares is risky: a PDF payslip can be altered in two minutes. So assessment leans more and more on data verified at the source. Real-time employment and income, phone line reputation, utility payments, transactional footprint. In Mexico that means crossing the bureau with formal-employment sources like IMSS or ISSSTE and with local telcos, Telcel and Movistar among them.

Círculo de Crédito is already integrated with uFlow through the Partner Link API Integration certificate, and Buró de Crédito is in as well. The harder the signal, the less the decision misses. No data eliminates bias, but when it stays governed and traceable, you can reduce it and audit it.

Deciding in milliseconds

In digital lending, the best customers are won in fractions of a second. An engine evaluates thousands of data points in milliseconds and can run 10, even 40, calls to different APIs in a single second, deciding at each step what to ask for next. Whoever answers first keeps the customer.

Speed also improves the experience. Instead of a form that asks everyone for 100 fields, the engine opens paths by profile and asks only for what it needs. There is no point asking for 100 fields if 90 will come back empty. uFlow runs and versions those branches and models; it does not train them, it puts them to decide.

Testing without risking the portfolio

Before rolling a policy out to everyone, it is worth testing it small. The engine isolates the change to 1 or 2% of operations. If the new rule goes wrong, the worst case is one extra point of delinquency on that slice. And because it processes fast, it can isolate those odd transactions on the spot and send them to manual review.

What to watch if you lend in Mexico

A couple of things to keep on the radar toward 2026. McKinsey projects that automating with AI can cut manual tasks by about 50% and costs by around 20%, so not automating gets more expensive every year. Embedded finance, with marketplaces, wallets and delivery apps, is pushing hard as an originator. And for people with no history, the "low and grow" approach, approving small and scaling up with behavior, is how you include without blowing up risk.

Underneath all of it sits the same skill that separates the winners: adopting tools fast, connecting data without friction, and experimenting live on a small slice of the portfolio.