AI in the classroom: it can predict dropouts, but the risk is drowning teachers in data

By Zendoric · October 1, 2026.

The 2025-2026 school year marked, according to Laura Elizondo, co-founder and CEO of the educational platform Lexium, the real turning point for artificial intelligence (AI) in the classroom: a shift that, in her words, happened "far too fast." The claim comes in an interview with Publimetro México and describes a pattern that is not unique to one country: schools that in barely a year went from using AI as a curiosity to making it daily infrastructure for academic management.

According to Elizondo, the effect is not uniform across educational stages. In primary and secondary school, she says, AI works as limited support for the teacher and as a way to enliven class. In high school and university the picture is more fraught, because there academic use coexists with decisions about retention, tutoring and guidance that carry higher stakes.

The central diagnosis from Lexium's CEO is counterintuitive: the problem is not a lack of data, but a poorly digested excess of it. She calls it "paralysis by analysis" and compares it to a luxury car so overloaded with features that the driver ends up using 10% of what it offers. Her proposal — the product her company sells, it is worth underlining — are dashboards (Goldmind) that reduce that flood of data to actionable indicators on recruitment, admissions, retention and persistence, supported by a module called Intelection which, as the company describes it, operates within tight bounds to prevent the system from roaming freely and drawing conclusions on its own.

Elizondo's strongest argument is dropout as the "tip of the iceberg": leaving school does not begin the day the student walks away, but months or years earlier, in signals such as poor study habits or reading comprehension difficulties. Detecting those signals in time, she argues, makes it possible to trigger tutoring before the problem becomes irreversible, both in higher education and — preventively — in primary and secondary school.

In general, the challenge she describes is not exclusive to the education sector: any organization adopting AI at speed faces the same overload problem before it finds the right filter. What is distinctive here is the terrain: classrooms of 30 to 40 students with very different socioeconomic and cultural profiles, and families that now ask not just for academic standards but for comprehensive support. That heterogeneity is what makes it tempting to delegate to a system the task of deciding who needs help and what kind.

This testimony should be read for what it is: the voice of a company describing the problem its own product claims to solve, with no independent studies or efficacy figures to back it up in the article. That does not invalidate the diagnosis — data overload in institutions digitizing at forced march is a real phenomenon, documented in other sectors — but it does require treating the claims about Lexium as what they are: marketing with a reasonable basis, not verified evidence.

The most valuable part of the testimony, in fact, is the warning Elizondo levels against her own business: that predictive analytics should serve to illuminate patterns, never to replace the judgment of the teacher or the administrator. It is a distinction that connects with something we have already noted when analyzing AI's impact on education by sector: the teacher who gains from this wave is not the one who transmits content, but the one who orchestrates the technology and keeps the final say over decisions affecting a specific person.

There too lies the short-term risk that should not be minimized: a system that classifies a teenager as "at risk of dropping out" based on their study habits can get it wrong, stigmatize or pigeonhole them, and the temptation to automate that label to save administrative time is real, especially in systems with few human guidance resources. Elizondo's own warning — "you, principal, you, teacher, make the decisions" — is right precisely because the commercial and organizational incentive pushes in the opposite direction.

Over the long term, however, the underlying promise deserves to be taken seriously: predictive analytics applied to education is one of the most direct routes to the abundance we defend at this outlet. Detecting in time that a student from a family without the means for private tutors is losing ground in reading comprehension, and being able to intervene before they drop out, is democratizing something that for generations only wealthier families could buy: personalized, early attention. If the technology delivers on that promise without turning the student into a number managed by an algorithm, the hard transition of these years of accelerated adoption — data paralysis included — will have been worth it.