Why Your Company’s AI Training Isn’t Working

Your company probably ran an AI training session last quarter. Maybe it was a half-day workshop. Maybe it was a LinkedIn Learning playlist emailed out with a note saying “complete by Friday.” Maybe it was a vendor demo disguised as education.

Whatever it was, it almost certainly didn’t work.

That’s not a guess. According to Josh Bersin Company’s February 2026 research across 800 organizations, companies spend more than $400 billion annually on corporate training. Despite that investment, 74% of senior leaders say their companies still lack the skills to compete. The money is being spent. The behavior isn’t changing.

The forgetting curve is brutal

The core problem is well-documented but widely ignored: people forget most of what they learn in a training session almost immediately. Research on the Ebbinghaus forgetting curve, applied to corporate settings, shows employees forget roughly 70% of new information within 24 hours and up to 90% within a week without reinforcement.

Harvard Business Review put it more bluntly: only 12% of employees actually apply new skills learned in L&D programs to their jobs. And only 25% of respondents to a McKinsey survey believe their training measurably improved performance.

These numbers explain why the typical corporate AI training approach — bring in a speaker, run a workshop, distribute some course links — produces so little change. A one-day session on prompt engineering is inspiring in the moment and forgotten by Monday.

Why MOOCs and playlists don’t fix it

The usual response is to buy seats on an online learning platform. But the completion data is damning. A cross-platform review of 221 MOOC courses found a median completion rate of 12.6%. MIT and Harvard research on their own MOOCs found completion rates of 3–6% among all registrants. Coursera hovers between 5–15%.

LinkedIn Learning does slightly better — 15–25% — but that’s partly because courses are shorter. Completing a 45-minute video isn’t the same as developing a new capability.

The fundamental issue: passive content consumption doesn’t produce behavioral change. Watching someone explain AI tools is not the same as using them under guidance, getting feedback, and being assessed on your ability to apply them.

What actually works

The research on effective training is consistent and clear. A meta-analysis of 225 studies published in PNAS found that active learning increases exam scores by an average of 6 percentage points and that students in traditional lecture formats are 1.5 times more likely to fail.

Structured programs with assessment, accountability, and spaced repetition produce dramatically better outcomes. Cohort-based courses with live interaction boost completion rates from 10% to 85%. Organizations with structured onboarding using active learning see 62% greater new hire productivity.

The pattern is clear: training works when it has structure, when it requires doing (not just watching), when it includes assessment, and when it spaces learning over time rather than cramming it into a single session.

The AI advantage in training delivery

Here’s where it gets interesting. A 2025 Harvard randomized controlled trial found that AI tutoring produced more than double the learning gains of active learning classrooms, with effect sizes of 0.73–1.3 standard deviations. AI-adaptive platforms improved student test results by 62%.

The reason is straightforward: AI tutors deliver exactly the kind of training that research says works — structured progression, active practice, personalized feedback, spaced repetition — but at scale. A human instructor can’t have individual Socratic dialogues with 500 employees simultaneously. An AI tutor can.

Yet fewer than 5% of companies have deployed AI-native learning technology, according to Josh Bersin’s research. Companies using what he calls “dynamic enablement” (AI-native training) are 6x more likely to exceed financial targets and 28x more likely to unlock employee potential. The early adopters are seeing massive results. Everyone else is still running workshops.

Beyond AI skills: the platform applies to any training

The irony is that the same structured-curriculum-with-AI-tutoring approach that fixes AI training also fixes every other training problem. Employee onboarding, compliance certification, professional development, product knowledge, sales enablement — any domain where you need people to actually learn and apply skills benefits from the same model: structured curriculum, Socratic instruction, exercises, assessment, and accountability.

The AI isn’t just the subject being taught. It’s the teaching method. And that distinction matters for every L&D leader deciding where to invest next.