Microlearning: the new way of learning only works if you actually practice too
Over the past few years, a great deal has changed in how organizations learn to organize. The training day has been broken up, the learning platform is full of short modules, and microlearning has found its way into virtually every L&D plan. Completion rates are good. Satisfaction scores are too.
And yet, in the boardroom, the same question is still being asked: why do we see so little of it reflected on the work floor?
That question is justified, and the answer is uncomfortable. The new learning has made learning shorter, not more effective. Whoever replaces a six-hour training day with eighteen twenty-minute videos has adjusted the format while leaving the mechanism intact. In both cases, someone is consuming content. And consumption is not the same as capability.
What the new learning promised
The promise was attractive and sensible in its own right. Learning was supposed to move from the classroom to the workplace. Short, at the moment you need it, adapted to who you are and what you already know. Learning in the flow of work, as it has come to be called in professional literature.
The form of that promise has succeeded well. Modules are shorter, mobile-friendly, and easier to find. Libraries with thousands of topics are available for a few tens of euros per employee per year.
The working of that promise has been less successful. Because most microlearning still consists of someone explaining something, followed by a few multiple-choice questions. That is a miniaturized training day, not a different learning process.
We wrote earlier that the traditional training day has had its day. That still stands. Only: the successor has inherited the same flaw.
Why shorter is not automatically better
Here it helps to look at memory research, because on this point it is remarkably unambiguous.
Jaap Murre and Joeri Dros from the University of Amsterdam replicated Ebbinghaus’s classic forgetting experiment from 1885 in PLOS ONE (2015). Their conclusion: the forgetting curve holds up. The loss is steepest in the first hours and the first day, after which it flattens out. That happens regardless of whether the content is presented in six hours or in six minutes. Format does not change the curve.
What does change the curve is spacing. Cepeda and colleagues brought together 839 measurements from 317 experiments in Psychological Bulletin (2006) and found that spaced learning consistently wins over massed learning. More importantly: the optimal interval between two learning moments increases the longer you want to remember something. Anyone who wants to still be able to do something a year later must spread repetitions over weeks to months, not over a single afternoon.
And then the third insight, which runs most counter to intuition. Roediger and Karpicke showed in Psychological Science (2006) that retrieving information sticks better than restudying it. In a test immediately afterwards, repeated restudying seemed to work better. In a test after two days and after a week, it was the reverse: those who had been tested remembered considerably more. But the participants who had reread were convinced themselves that they would remember it better.
Read that last sentence one more time, because that is where the problem of microlearning lies in a single observation. Consuming feels effective. Retrieving feels difficult. And precisely the difficult part is what works.
A learning platform that measures completion rates is therefore measuring feeling, not effect.
Microlearning and microtraining are not the same thing
It is useful to pull those two concepts apart, because they are constantly being used interchangeably.
| Microlearning | Microtraining | |
| Core activity | Watching, reading, listening | Doing, formulating, reacting |
| Role of the learner | Recipient | Performer |
| What is measured | Completion and satisfaction | Exhibited behavior |
| Difficulty | Low, comfortable | High enough to create friction |
| Works well for | Knowledge, procedures, systems | Behavior, conversations, soft skills |
| Feels afterward | Smooth and useful | Uncomfortable and educational |
Microlearning is fine for what it is. Anyone who needs to know how the new declaration procedure works or what is in the amended collective labor agreement is exceptionally well-served by a seven-minute module. That is knowledge transfer, and short, easily searchable content is precisely the right tool for that.
The misunderstanding arises as soon as microlearning is deployed for behavioral change.
Soft skills do not lend themselves to consumption
You do not learn how to de-escalate by watching de-escalation. You learn it by sitting in a conversation where someone starts talking louder and you have to choose what to say.
That sounds self-evident, and yet the vast majority of soft skills offerings are still explanations. Modules on giving feedback, on listening, on dealing with resistance. All information about behavior, instead of behavior.
Phanish Puranam of INSEAD describes in his article on meta-skills for the World Economic Forum (June 2026) the conditions under which the ability to acquire new skills strengthens: novel and complex tasks, timely feedback, and working in mutual interdependence. None of those three conditions are present in a video.
The OECD concludes in Skills in the AI age (July 2026) that AI unevenly shifts the skills demand across sectors and levels, and that adaptation requires developing skills that align with what the labor market actually demands. In practice, that is mainly about the work that remains when systems take over the routine tasks: conversations, exceptions, assessments.
Precisely that work cannot be explained. It must be practiced.
Practical example: a housing corporation with too many escalations
Take a housing corporation with sixty residential consultants. They conduct difficult conversations on a daily basis: a tenant three months behind on rent, a neighbor dispute that has been running for two years, a home visit where it becomes clear that there is more going on than rent arrears.
The corporation has invested in a learning platform. A learning path “conducting conversations during payment arrears” is available, built out of eight ten-minute modules. The completion rate is 94. The evaluations are positive.
Yet the number of files going to the bailiff is not decreasing, nor is the number of complaints about treatment. In conversations with teams, the training advisor hears the same thing over and over: “I know very well how it should be done. But when someone starts screaming in the doorway, I just do what I always do.”
That is not a knowledge problem, and therefore more microlearning does not solve it. It is an execution problem under pressure.
The approach that does work looks different. The team identifies three conversations that go wrong most often: the first phone call about arrears, the home visit after two fruitless letters, and the conversation where someone fails to keep a payment arrangement. For each conversation, four behaviors are established, so concrete that a colleague can see whether you are doing them. For example: state the amount and the deadline in the first minute, then ask what is going on before offering an arrangement.
Subsequently, every residential consultant practices those conversations. Twice a week an AI roleplay or microtraining of ten minutes, for eight weeks, with a tenant who one time shows shame, the second time becomes angry, and the third time promises what they cannot deliver.
What happens then is not spectacular. The conversation becomes routine. And routine is precisely what you need the moment someone starts screaming in the doorway.
What a microtraining looks like that actually changes behavior
Four criteria make the difference between microlearning and an exercise that changes behavior.
It begins with doing, not with watching. The first action of the learner is an utterance, not a click. Explanation can follow afterward, as an answer to something you tried yourself.
It is short enough to maintain and heavy enough to cost something. Eight to twelve minutes works. Shorter becomes superficial, longer becomes an event for which you have to clear your schedule.
It repeats at intervals. Based on the research by Cepeda and colleagues, spreading over weeks is more effective than everything in a single block. Twice a week for eight weeks resonates better than sixteen exercises in two days.
It ends with something you can use tomorrow. No summary, but one concrete formulation. “I see that the amount has risen to 1,240 euros. Before I propose anything, I would like to know how this came about.” That is a sentence someone takes with them.
Within PractAIce, microtraining is set up around that last point. If a practice conversation went wrong at a single moment, you can return to precisely that moment instead of redoing the entire conversation. The AI coach shows beforehand what an alternative formulation might look like, so that you can apply it and immediately notice what changes in the other person’s reaction. Two to four minutes at a time.
That is retrieval instead of rereading, precisely as Roediger and Karpicke describe it, but with behavior instead of facts.
Rhythm is more important than content
Most organizations spend their attention on the question of which content they purchase. That is the least determining choice.
What is determining, however, is whether a rhythm is created. Three interventions help with that:
- Link the exercise to an existing moment. The last ten minutes of the team meeting, the day before a home visit, Monday morning. Loose activities disappear, linked activities stay.
- Work with one skill per period. Eight weeks on one conversation yields more than eight topics in eight weeks. That feels meager and it works better.
- Make it visible without making it mandatory. A team overview of who has practiced works better than a deadline. Mandatory completion yields completion, not practice.
More about aligning with individual differences is in Hyper-personalized learning.
What this means for the role of L&D
If content is not the determining factor, the work of the training department shifts. From compiling a catalog to designing practice moments.
That is substantively heavier work, and it is closer to the business. You need to know which conversations in which roles are decisive for the result. You must be able to dissect those conversations down to behavior. And you must agree with managers when practice will take place.
McKinsey points out in The rise of the human-AI workforce (April 2026) that AI makes it possible to organize learning closer to the work: tailored to how someone performs, and offered at a moment when there is space in the schedule. That is the direction it is heading. But the technology only provides the moment. What happens in that moment remains a design question.
Whoever makes that translation structural also builds something that is usable outside the learning platform: an image of what people demonstrably can do. We wrote about that in Skill passport: why your job profile ages faster than your people.
What you should measure
The completion rate is the most collected and least informative figure in L&D. Three alternatives:
- Practice frequency per person per month. Not how many modules have been completed, but how often someone has actually done something.
- Development on named behaviors. Does the behavior you predefined increase over eight weeks?
- An operational indicator linked to it. Number of escalations, number of complaints about treatment, lead time of a file. Something from business operations, not from the learning platform.
Frequently asked questions about microlearning
What is microlearning?
Microlearning is learning in short units of a few minutes, usually via video, text, or a short quiz, meant to be consulted at the moment of need. It is primarily suitable for knowledge, procedures, and systems.
What is the difference between microlearning and microtraining?
In microlearning, the learner consumes content. In microtraining, the learner executes behavior and receives feedback on it. The first is suitable for knowing, the second for capability.
Does microlearning work for soft skills?
For explaining a model or a conversation structure, yes. For actually mastering a conversation, no, because skill arises through execution with feedback and not through watching.
How long should a microtraining last?
Eight to twelve minutes for a complete exercise, two to four minutes for a targeted repetition on a single moment from a previous conversation. Longer makes it an activity for which someone has to clear time, and then it doesn’t happen.
How often do you need to repeat to make something stick?
Research into spaced practice shows that the optimal interval increases the longer you want to remember something. For behavior that needs to last a year, twice a week for six to eight weeks works better than the same number of exercises in a single week.
If you suspect that your organization completes a lot and changes little, that can be tested in a single afternoon. Take the conversation that goes wrong most often in your sector, dissect it down to four behaviors, and let one team practice that for eight weeks. On the solutions page you can see what those practice moments look like, from the scenario that an employee builds themselves to the microtraining with which you return to the moment where it went wrong. If you would rather experience for yourself how it feels first, then we schedule half an hour in which you conduct the conversation and we mostly keep our mouths shut.