Practice-Based Training: Why Doing Beats Watching at Work

Nine in ten executives now point to learning and development as the function where AI creates the most value, up from 74% a year earlier. Yet only 32% of workers say their employer has given them proper AI training. Those numbers come from two surveys published within days of each other, and together they explain why practice-based training has become the defining strategy conversation in corporate learning this year. Companies are buying more learning content than ever, while the people that content is meant for keep saying it does not prepare them to actually do the work.

This post walks through the spending data behind the shift, the research on why doing beats watching, and five concrete moves for turning a passive course library into real practice.

Why practice-based training is replacing content libraries

L&D teams are moving budget away from producing course content and toward structured practice because consumption alone does not change behavior on the job. That shift is now visible in the money. Forbes reports that US corporate training spend grew from $98 billion in 2024 to $102.8 billion in 2025, with spending on external learning products and services jumping 29%, from $12.4 billion to $16 billion.

More telling than the totals is what the leaders spending that money now say out loud. The head of learning at one global consumer goods company put it plainly in that Forbes piece: practice, and not content consumption, is the real measure of learning.

The old model this replaces was slow and expensive in exactly the wrong places. One legacy management program described in the article took 16 months, roughly 20 instructional designers, and about $1 million to build. All of that effort went into producing material for people to sit through, before a single manager had rehearsed a single difficult conversation.

What learning science says about learning by doing

The research here is unusually one-sided. A meta-analysis of 225 studies published in PNAS found that students in active learning classrooms outperformed lecture students by 0.47 standard deviations on exams, roughly half a letter grade.

The failure numbers are starker. Under traditional lecturing, 33.8% of students failed, versus 21.8% in active learning formats. Students who were lectured to were 1.5 times more likely to fail than students who spent class time doing things.

That study covered university STEM courses, but the mechanism travels. A compliance module an employee watches end to end is a lecture. So is a 40-minute onboarding video. The learner's brain is doing the same thing in both cases: passively following along, generating nothing, retrieving nothing, and getting no feedback. The formats that beat lecture all share one trait, and it is not production value. They make the learner respond, then respond to the learner.

How AI makes practice cheap enough to scale

Practice lost to content for decades because practice was the expensive part. A video scales to 10,000 employees for free. A role-play needs a skilled partner, a schedule, and a private room, for every single rep. AI collapsed that cost, and the early numbers from large enterprises suggest the effect is real.

The consumer goods company above rebuilt its first-time manager program around AI practice partners. The redesigned program runs nine weeks, serves about 1,500 new managers, and among its manufacturing population, 100% of learners participate in the AI-enabled practice sessions. One learning executive described it as like having your own AI tutor in your pocket.

A global consulting firm measured the same idea on its new hires and reported roughly a 20% improvement in communication skills after adding AI-based practice, alongside 10 to 20% efficiency gains in how programs get built and delivered.

And a health technology company found that AI role-plays gave early-career hires something human role-plays rarely do: psychological safety. Interns could rehearse a hard conversation badly, five times in a row, with no audience and no judgment, until they were confident enough to do it for real.

These are self-reported internal results, not audited studies, and they deserve that caveat. But they all point the same direction as the academic evidence: reps with feedback move skills, exposure to content does not.

Why the AI skills gap is really a practice gap

Workers are not short on motivation or time to learn AI. They are short on structured practice. A new Indeed and YouGov survey of more than 1,000 workers and hirers found that 59% of employers call finding AI-native talent essential within the next year, while only 19% of job seekers feel fluent in AI.

The gap is not effort. About 70% of workers have already started experimenting with AI tools on the job, and only 8% say they lack time to learn. What separates the fluent from the stuck is access to real training: 77% of self-described AI natives say they received adequate training, against just 15% of workers not yet using AI heavily.

Unguided experimentation plateaus fast. Employees poke at a tool, get a mediocre result, and quietly go back to the old way of working. What turns experimentation into fluency is the same thing that turns a nervous new manager into a good one: structured reps, on realistic tasks, with feedback that tells you what to change.

How to shift your programs from watching to doing

You do not need to throw away your content library. You need to demote it from the product to the raw material. Five moves that teams making this shift consistently get right:

1. Start from decisions, not topics. List the 5 to 10 judgment calls a person in the role actually faces, then build practice around those. The content that survives is whatever those decisions require.
2. Interrupt content with choices. Do not put one quiz at the end. Insert a question, decision, or short scenario every few minutes so learners generate answers while they learn, which is where the active learning effect lives.
3. Give safe reps with instant feedback. Role-plays, branching scenarios, and simulations work because the learner can fail privately and retry immediately. Psychological safety is a feature you can design for.
4. Measure practice, not watch time. Completion rates tell you people pressed play. Track attempts, decision accuracy, and performance change instead, the way the consulting firm above tracked its 20% communication gain.
5. Use AI as the practice partner. The historic blocker was staffing every rep with a human. Formats like interactive training videos now put an AI tutor inside the lesson itself, pausing to ask questions, adapting to each learner's answers, and coaching in real time, at the cost of a video.

FAQ

What is practice-based training in the workplace?

Practice-based training is a design approach where the core of a program is the learner doing realistic tasks, making decisions, and getting feedback, rather than consuming content about the tasks. Content still exists, but it supports the practice instead of replacing it. Think rehearsing a customer objection with an AI partner instead of watching a video about objection handling.

Is practice-based training more expensive to build than standard e-learning?

It used to be, because every rep needed a human facilitator or role-play partner. AI practice partners have inverted that: one large enterprise cut a program that once needed 16 months and about 20 designers down to a nine-week, practice-centered experience. The expensive part now is deciding which decisions matter, not staffing the reps.

How do you measure whether practice-based training is working?

Track leading indicators inside the training, such as practice attempts, decision accuracy, and improvement across reps, then tie them to one downstream metric the business already cares about, like ramp time or error rates. Firms doing this well report concrete deltas, such as a 20% measured improvement in new-hire communication skills. If you can only measure completions, you are measuring attendance, not learning.

The pattern across the spending data, the classroom research, and the workforce surveys is the same: people learn what they practice, and almost nothing they merely watch. The organizations pulling ahead are not the ones with the biggest content libraries, they are the ones turning every lesson into a conversation. That is the entire case for interactive learning, and the numbers now back it up.