FEEDBACK · AT SCALE
Scaling Feedback: How to Give Every Student Detailed Comments Without Burning Out Staff
A 300-student module is 150 hours of marking. How to keep feedback detailed and on time at any cohort size, without adding headcount.
By Eduface · July 2026 · 8 min read
The expectation is clear: every student should receive meaningful, specific feedback on their work. The reality is equally clear. With cohort sizes of 200 or 300 students per module, the numbers do not add up. A module with 300 students submitting 2,000-word essays represents 600,000 words to read and comment on. At 30 minutes per essay, that is 150 hours of marking for one module, one submission window. Most lecturers do not have 150 hours to spare.
How do you give every student detailed feedback without more staff?
The answer lies in changing the system, not adding more staff. AI assessment tools generate per-criterion written feedback for every student in the same time it takes to mark one submission manually. The lecturer reviews and approves rather than writes from scratch. The result is detailed, rubric-referenced feedback for an entire cohort, delivered within 24 to 48 hours of submission, without requiring additional headcount.
Why does feedback quality drop as cohort sizes grow?
The relationship between cohort size and feedback quality is well documented. Research by Hattie and Timperley found that feedback is one of the most powerful influences on student learning, with an effect size of d=0.73. But the same research makes clear that only specific, task-focused feedback produces those outcomes. Generic comments do not move the needle.
Nicol and Macfarlane-Dick established that effective feedback must be specific, timely, and forward-looking if it is to support student self-regulation. Time pressure works directly against all three criteria.
A UCU Workload Survey found that 75% of higher education staff describe their job as stressful, and 46% report unrealistic time pressures, with marking consistently identified as a primary driver. When a lecturer faces 300 scripts and a tight return window, the rational response is to shorten feedback, focus on common errors, or rely on stock phrases. Students receive variations of good argument but needs more referencing rather than comments tied to their specific submission.
There is also a consistency problem. Research by Bloxham found that human inter-rater reliability in UK higher education varies by 20 to 40 percentage points, and that this variance increases further under time pressure. Student 1, marked on day one of the window, may receive substantively different quality feedback from student 300, marked on the final evening before the deadline.
What does personalised feedback at scale actually look like?
Personalised feedback at scale does not mean template text with a student’s name inserted. It means the feedback is generated from each student’s individual submission, referencing that student’s arguments, structure, and use of evidence.
Eduface generates per-criterion written comments for every submission. The AI reads the actual text submitted, maps it against the rubric criteria, and produces comments that reflect what that specific student wrote. Two students who receive the same grade on a given criterion will receive different written feedback, because their submissions are different.
From the lecturer’s perspective, the process works as follows: set up the rubric, upload or receive submissions via LMS integration, review the AI-generated grades and feedback sets, and release to students. The lecturer is not removed from the process. Academic judgement is applied at the review and approval stage, where it has the most leverage.
How does AI-assisted feedback hold quality across large cohorts?
Because the AI applies the rubric consistently to every submission, feedback quality does not degrade as cohort size grows. The same criteria are applied with the same rigour to student 1 and student 300. There is no fatigue effect, no variation in attention, and no compression of comments at the end of a long marking session.
95% alignment, 24 to 48 hour turnaround
In UK pilots, Eduface’s grades aligned with lecturer assessments in 95% of cases. Cohort-level feedback that previously took weeks to produce, review, and release can be completed within 24 to 48 hours of submission.
Does automated feedback feel generic to students?
The concern is legitimate and worth addressing directly. Template-based feedback tools do produce generic output. A system that inserts a student’s name into a pre-written paragraph is not personalised feedback: it is mail merge.
Eduface’s approach is different. The feedback is generated from the submission itself, not from a bank of pre-written phrases. The AI reads what the student wrote and produces comments that reflect it. A student who presents a well-structured argument with weak evidencing receives different feedback from a student whose argument is underdeveloped but whose sources are strong, even if both score similarly overall.
This matters for student engagement. Students who receive feedback that clearly addresses their specific work are more likely to read and act on it. Generic feedback reinforces the perception that marking is a procedural exercise rather than a genuine response to their work.
What does the NSS say about feedback quality at scale?
The NSS Assessment and Feedback section tests three dimensions that are directly affected by scale: timeliness, detail, and usefulness. All three suffer when lecturers are marking large cohorts under time pressure.
Institutions that struggle with NSS feedback scores often identify the same proximate causes: feedback arrives late, comments are too brief, and students cannot see how the feedback connects to the assessment criteria. These are not academic quality failures in origin. They are resourcing and systems failures that surface as academic quality failures in the data. Improving NSS feedback scores at scale requires changing the production process, not asking lecturers to work faster or harder.
Feedback quality at scale: a comparison
Cohort size
Traditional marking
AI-assisted (Eduface)
30 students
Detailed feedback feasible
Detailed feedback feasible
100 students
Time pressure begins to affect quality
Full per-criterion feedback maintained
200 students
Generic feedback common; turnaround extends
Per-criterion feedback; under 48-hour turnaround
300+ students
Feedback often shortened or delayed
Consistent quality regardless of cohort size
Frequently asked questions
Can AI generate genuinely personalised feedback for each student?
Yes, when the system is built to do so. Eduface generates feedback from each individual submission rather than selecting from pre-written templates. The AI reads what the student wrote, maps it against the assessment criteria, and produces criterion-specific comments reflecting that student’s actual arguments, structure, and evidence. Two students submitting on the same topic will receive substantively different feedback if their work is different.
How quickly can Eduface deliver feedback to a cohort of 300 students?
In UK pilots, cohort-level feedback that previously took weeks to produce and release was completed within 24 to 48 hours of submission. The AI generates grades and feedback sets for the full cohort simultaneously. The lecturer’s time is spent reviewing and approving the outputs rather than writing from scratch, which compresses the process significantly without removing human oversight.
Does the lecturer still review feedback before it reaches students?
Yes. Every grade and feedback set is reviewed and approved by the lecturer before it is released to students. Eduface operates a human-in-the-loop model: the AI produces the initial output, the lecturer reviews it, and only approved feedback reaches students. This satisfies the requirements of the EU AI Act for human oversight in high-risk AI applications, including educational assessment.
How does scaling feedback with AI affect NSS scores?
NSS Assessment and Feedback questions reward timeliness, detail, and usefulness. All three are constrained when lecturers are marking large cohorts manually under time pressure. AI-assisted assessment addresses each constraint directly: feedback is released faster, comments are criterion-specific rather than generic, and the connection between feedback and rubric criteria is explicit in every response.
Is AI-generated feedback appropriate for summative as well as formative assessment?
Yes. Eduface is used for summative assessment in UK and Dutch higher education institutions, including Bath Spa University, Tilburg University, and UMCG. The system generates per-criterion written feedback aligned to the assessment rubric, and every grade is approved by a lecturer before release. The human-in-the-loop model ensures that final marks carry the same academic authority as traditionally marked assessments.
Scaling feedback is a systems problem, not a staffing problem
Institutions that treat feedback quality as a resourcing question will keep arriving at the same answer: there are not enough hours. The economics do not improve as cohort sizes grow. AI assessment tools change the underlying equation by making per-student, per-criterion feedback viable at any cohort size, without additional headcount.
Eduface integrates with Canvas, Brightspace, Moodle, and Blackboard via LTI 1.3, and is approved through Jisc/CHEST (UK) and HEAnet (Ireland).
References
Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81-112.
Nicol, D. J., & Macfarlane-Dick, D. (2006). Formative assessment and self-regulated learning: a model and seven principles of good feedback practice. Studies in Higher Education, 31(2), 199-218.
University and College Union. (2016). Workload is an education issue: UCU workload survey report 2016. UCU.
Bloxham, S. (2009). Marking and moderation in the UK: false assumptions and wasted resources. Assessment & Evaluation in Higher Education, 34(2), 209-220.
Feedback for every student, at any scale
Book a demo or set up a free lecturer account, and see per-criterion feedback on your own cohort.