· 7 min read

Adaptive Learning That Educators Control

What adaptive learning does well, where automatic routing falls short, and how to keep decisions about a learner's next step with the educator.

After class, an educator sits at a learner's desk reviewing a tablet, with a stack of learner notebooks and a mug beside them.

A learner misses two questions on a Monday morning, and their practice software sends them back to a fractions unit from two grades earlier. They were tired and rushed the questions, but the software has no way to know that. Their educator finds out a week later, when the learner asks why they're doing "baby work."

The software did what it was built to do. It adapted. What was missing was anyone deciding whether the change made sense for that learner.

Adaptive learning means adjusting what a learner works on next based on how they're doing: more practice when a skill isn't secure, a step back when an earlier idea is missing, or a move ahead when the work shows a skill is solid. Good educators have always adapted this way. Software makes it possible to track the evidence for many learners at once, which helps as long as the educator still makes the call.

What the research says

Well-designed adaptive practice can help. A meta-analysis by James Kulik and J. D. Fletcher found that learners using intelligent tutoring systems, software that adjusts instruction to each learner's responses, outperformed learners in conventional classes in most of the controlled evaluations they reviewed. A review of experimental research on education technology by Escueta and colleagues found that computer-assisted learning shows particular promise in math.

Those studies measure how well software responds to answers. They say much less about the moments when the software's decision and the learner's situation don't line up, which is where educators spend a lot of their time.

Where automatic routing goes wrong

An answer is evidence, but it's thin evidence on its own. The learner in the fractions example missed two questions for a reason that had nothing to do with fractions. Another learner could miss the same questions because the word problems were hard to read, or because they were absent for one lesson. Each reason calls for a different response, and only one of them might justify going back two grades.

When software makes that call alone, two more problems follow. The educator can't explain the change, because they didn't make it and may not know it happened. The learner, meanwhile, gets a decision with no reason attached, and a learner moved backward without explanation tends to hear it as a verdict on their ability. A remedial track that nobody planned an exit from can also be hard to leave, which is why differentiated instruction for one learner starts by deciding how the support will end.

Keep the data, move the decision

The fix isn't to ignore what the software collects. It's to put that evidence in front of the person who knows the learner.

Kenneth Holstein, Bruce McLaren, and Vincent Aleven tested this in classrooms using math tutoring software. When educators had real-time information about each learner's progress and struggles, they could see who needed help and step in, and learners in those classrooms learned more than learners using the same software without that information reaching their educator.

Software is good at spotting a pattern across thirty learners. Deciding what the pattern means for one learner, on a particular day, still takes someone who knows them.

What educator-controlled adaptation looks like

In practice it's a short loop, and each step depends on the one before it.

  1. Notice a pattern, such as a learner falling below the mastery target on two pads in a row or making the same error across several problems.
  2. Read the actual responses before deciding what the pattern means. Careful reasoning that went wrong at one step calls for something different than answers that were clearly guesses.
  3. Make one change that fits what you found: a short support lesson before the next step, more time, or a lesson moved later. Changing one thing lets you tell whether it worked.
  4. Tell the learner what changed and why, including where the support leads.
  5. Decide what evidence brings the learner back to the main path, and watch for it.

For one learner, the whole loop takes a few minutes. Over a term, it's what separates software that moves learners around from a plan a learner can understand and trust.

Questions to ask before choosing an adaptive tool

If you're comparing adaptive tools, these questions show whether a tool supports your judgment or replaces it:

  • Can you see why a change is recommended, including the learner's actual responses?
  • Can you approve, edit, or undo a change before the learner sees it?
  • Does the tool work alongside your curriculum, or does it assume it will replace it?
  • Can learners see what changed in their work, and why?
  • Who can see learner work, and is it kept private?

How PushPad handles adaptive learning

PushPad provides supplemental practice for the instruction you already deliver. You assign a learning path of lesson pads, each focused on one concept, and PushPad records each learner's scores and responses so you can see the evidence behind every decision.

When a learner needs a different route, you make the change. You can add a focused pad before the step where they're stuck, move a lesson, or extend or remove their deadline, and nothing changes on their path until you confirm it. Learners see their updated path. When an assignment requires mastery, the next pad unlocks only after a learner reaches the target, so nobody moves ahead on a score that says they aren't ready. Your learning paths and pads stay private to you and the learners you invite.

Common questions

What is the difference between adaptive learning and personalized learning?

Personalized learning is the broader idea of tailoring goals, pace, and approach to each learner, often including learner choice. Adaptive learning is one way to do it: adjusting the next task based on how the learner is performing. A program can adapt practice without personalizing anything else about the learning.

Is adaptive learning only for learners who are behind?

No. The same loop works in both directions. A learner who reaches the mastery target quickly can move ahead or take on a more challenging pad, while a learner who needs support gets it before the next step. Adapting means matching the next task to the evidence, whichever way it points.

Which subjects suit adaptive practice best?

Skills that build in a clear order and have checkable answers, such as math procedures, vocabulary, and grammar, are the easiest for software to track, and math has the strongest research base. Open-ended writing and discussion are harder for software to judge, so an educator's review of the actual work matters even more there.


Research behind this article

The fractions example illustrates a common situation. It is not drawn from a specific learner.

Put this into practice with lesson pads you control.

Get started
Back to blog