Formative Assessment at Scale: Personalised Feedback for Every Student

Formative assessment AI uses machine grading to handle frequent, low-stakes checks for understanding so teachers can return specific, personalised feedback to every student quickly enough to adjust the next lesson. Unlike summative tests that measure learning at the end, formative assessment happens during learning, and AI removes the marking bottleneck that has always made it hard to do well at scale.

What is formative assessment AI?

Formative assessment is the ongoing, low-stakes feedback loop inside a unit: exit tickets, short quizzes, problem sets, and quick writing tasks that tell you what students actually understand right now. The catch has always been time. Marking 30 essays or 200 problem sets weekly, then writing individual feedback, is unsustainable for most teachers.

AI changes the economics. Instead of grading every response by hand, the teacher designs the task, the model grades and tags responses by concept, and the teacher reviews and approves. The human stays in charge of judgement; the machine absorbs the repetitive labour.

How AI makes personalised feedback scalable

The shift is from grading to reviewing. According to IntelGrader, a workload that once meant 40-plus hours of weekly marking for 200 students can compress to roughly 45 minutes of tutor time: about 15 minutes to design the task, AI grading and concept-tagging in the middle, and 30 minutes of human review before feedback reaches students within 24 hours.

Good systems give each student more than a score. Useful feedback usually includes four parts:

• Concept-tagged feedback that names the specific misconception, not just "wrong."
• Targeted practice aimed at the gap the student showed.
• A reflection prompt so the student does the metacognitive work themselves.
• Trend data so both teacher and student can see progress over time.

That last element matters most for instruction. When responses are tagged by concept, you can see at a glance that, say, 18 of 30 students missed the same step, and reteach it tomorrow instead of discovering it on the exam.

Practical ways to start

You don't need to transform your whole assessment system. Start with one subject or one class and a single recurring task type, such as a weekly quiz or a short written response. Run it for about four weeks and track two things: how much time you actually save, and whether students re-engage with the feedback.

Weekly tends to be a sensible cadence. It's frequent enough to catch misconceptions early but not so frequent that it feels like constant testing, which raises anxiety and defeats the "low-stakes" point.

The limits worth naming

The technology is not the hard part; the practice around it is. Three failure modes are common. First, teachers collect rich data and then don't change their lesson plans, so the loop never closes. Second, students treat the checks as high-stakes and the anxiety benefit disappears. Third, generic AI feedback that ignores your curriculum and rubric loses credibility fast, which is why human review of the model's output is non-negotiable, not optional.

AI grading is also imperfect on open-ended and creative work, and it can be confidently wrong. Treat it as a first-pass marker that surfaces patterns, with a teacher signing off before anything reaches a student. Tools like IntelGrader are built around that review step rather than full automation, and that design choice is the point: the goal is to give teachers their time back, not to take their judgement out of the loop.

Used this way, formative assessment AI doesn't replace good teaching. It makes the kind of responsive, individualised feedback that good teachers already want to give finally possible for every student, not just the few you can reach by hand.

Disclosure: IntelGrader is built by the team behind AI in Education.