MARKING WORKLOAD
Reducing Marking Workload Without Reducing Feedback Quality
Marking is one of the biggest pressures on UK lecturers. AI assessment tools can cut the time spent marking while keeping, or improving, feedback quality. Here is how.
By Eduface · June 2026 · 9 min read
Every lecturer knows the feeling: a new batch of submissions arrives, the deadline to return marks is looming, and you still have lectures to prepare, meetings to attend, and your own research to push forward. Something has to give. Usually, it is the depth of the feedback. This trade-off is so familiar it has become accepted as inevitable. It is not.
Can you reduce marking workload without reducing feedback quality?
Yes. AI assessment tools, used with human oversight, separate the cognitive burden of generating feedback from the professional judgement of approving it. Lecturers review and approve AI-drafted grades and comments rather than writing everything from scratch. The result is less time per submission, with no reduction in the specificity or consistency of feedback students receive.
How much time do lecturers actually spend on marking?
The numbers are significant. A 2016 University and College Union (UCU) Workload Survey found that 75% of higher education staff describe their job as stressful, 46% report unrealistic time pressures, and 26% regularly work more than 50 hours per week. Marking sits at the centre of this. For a module with 150 to 200 students, spending even 20 minutes per submission adds up to 50 to 67 hours of marking alone, before moderation, administration, or appeals.
This is not a niche problem affecting a handful of large modules. It is structural. As student numbers have grown without proportionate increases in academic staffing, marking has become one of the most time-intensive and least reducible parts of an academic’s role. The cost of not addressing it shows up as staff turnover, burnout, and recruitment difficulty.
Why does time pressure reduce feedback quality?
When lecturers are under pressure, feedback gets shorter. Comments become more generic. The specific connection between a student’s argument and the marking criteria gets lost. Hattie and Timperley (2007) found high-quality feedback has an effect size of d=0.73 on student outcomes, but quality here means specific, timely, and criterion-referenced. Nicol and Macfarlane-Dick (2006) found feedback must be both timely and specific to support genuine learning.
Time pressure undermines both conditions. A lecturer working through submission 150 of 200 is unlikely to produce the same quality of comment as at submission 20. This is not a failure of professionalism; it is cognitive fatigue. Bloxham (2009) found human inter-rater reliability in higher education varies by 20 to 40%, partly attributable to fatigue and marking order. The 150th essay is judged by a different marker, in practical terms, than the 5th.
What does AI-assisted marking actually look like in practice?
It is worth being precise, because “AI marking” means different things to different people. In Eduface’s workflow, the AI does not make final decisions. It reads each submission against the rubric and generates two things: a draft grade per criterion, and written feedback explaining that grade in relation to the student’s work. This draft sits in the lecturer’s queue, ready to review.
The lecturer opens each submission, reads the student’s work, reads the AI draft alongside it, and decides whether to approve, adjust, or rewrite. In most cases the draft is accurate enough to approve with minor edits. The lecturer keeps full control. No grade or feedback is released until they sign off.
The workflow shift is significant. The lecturer moves from “generate and evaluate” to “evaluate and approve”. That is a fundamentally different cognitive task, and it is faster.
Stage
Traditional marking
With Eduface
Read submission
Lecturer reads cold
Lecturer reads with AI draft alongside
Write feedback
Lecturer writes from scratch
Lecturer reviews and edits AI draft
Apply rubric
Manual, per marker
Consistent AI application, lecturer confirms
Time per submission
Full marking time
Review and approval time, significantly reduced
Consistency across cohort
Variable (fatigue, order effects)
Consistent rubric application
Feedback depth
Varies under time pressure
Per-criterion comments on every submission
Does reducing marking time mean students get worse feedback?
This is the assumption worth challenging directly. Under traditional marking at scale, students often receive brief, generic comments because the lecturer does not have time to write more. A rubric grid with ticks and a paragraph of general observations is common. Students get a grade but not a clear account of how their work performed against each criterion.
With AI-assisted marking, every student receives per-criterion written feedback as standard. Not a template, but generated commentary that references their specific submission. Even if the lecturer only approves rather than expands it, the student receives more granular commentary than under pure human marking of a large cohort under time pressure.
Consistency also improves. Because the AI applies the rubric uniformly across all 200 submissions, order effects and marker fatigue are eliminated as variables. The student who submitted on deadline day is assessed by the same standard as the one who submitted two weeks early.
How does Eduface maintain quality while cutting marking time?
Accuracy
In UK pilots, Eduface’s AI-generated grades align with lecturer assessments 95% of the time. The vast majority of drafts require only light review, making the lecturer’s queue genuinely faster to work through.
Human-in-the-loop by design
Every grade is held as a draft until a lecturer approves it. No feedback is released without sign-off. This is not just quality assurance; it is a governance requirement under the EU AI Act for high-stakes automated decision-making, and it is built into Eduface’s architecture.
Data sovereignty
Eduface processes all submissions on its own infrastructure in the Netherlands. No data is sent to external AI providers. This matters for GDPR compliance and for institutions that need assurance about where student work is processed.
Eduface is approved on the Jisc/CHEST framework in the UK and the HEAnet framework in Ireland, so procurement runs through established channels. Pilot partners include Bath Spa University, De Haagse Hogeschool, Hogeschool Rotterdam, Tilburg University, and UMCG.
Frequently asked questions
How much time can AI marking tools save lecturers?
It depends on cohort size and assignment complexity, but the principle is consistent: reviewing a draft is faster than writing from scratch. For a cohort of 200, a lecturer might move from 20 to 25 minutes per submission to 8 to 12 minutes, a saving of 20 to 30 hours on a single module. Exact figures vary by discipline and assignment type.
Does AI-assisted marking reduce the quality of feedback students receive?
The evidence points the other way. Under time pressure, human feedback tends to get shorter and more generic. AI-assisted marking generates per-criterion comments for every submission as standard. Even where the lecturer approves without expanding the draft, students typically receive more detailed feedback than under time-pressured human marking alone.
What does a lecturer actually do when using Eduface?
The lecturer sets up the rubric at the start of the module. When submissions come in, Eduface generates a draft grade and written feedback per criterion for each student. The lecturer works through a review queue: reading the submission, reviewing the AI draft, and approving or editing. No feedback reaches the student until the lecturer signs off.
Can AI marking tools be used for large cohorts?
Yes, and large cohorts are where the benefit is most significant. For a cohort of 30 the workload is manageable either way. For 300, writing detailed feedback on every submission is not realistic in most institutional contexts. AI-assisted marking lets every student in a large cohort receive specific, criterion-referenced feedback, and consistency improves at scale.
Is AI-assisted marking compliant with academic quality standards?
Eduface is built on a human-in-the-loop model: no grade or feedback is finalised without lecturer approval. This aligns with the EU AI Act requirements for high-stakes automated decision-making and with UK quality assurance expectations around academic judgement. The platform is approved on Jisc/CHEST (UK) and HEAnet (Ireland).
Summary
The assumption that reducing marking time means reducing feedback quality does not hold when the workflow changes. AI-assisted marking shifts the lecturer’s role from generating feedback under pressure to reviewing and approving accurate drafts. Students receive more consistent, more detailed feedback. Lecturers spend less time on each submission without giving up oversight. The trade-off dissolves.
References
1. University and College Union (2016). Workload Survey 2016.
2. Hattie, J. and Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81 to 112.
3. Nicol, D. and Macfarlane-Dick, D. (2006). Formative assessment and self-regulated learning. Studies in Higher Education, 31(2), 199 to 218.
4. Bloxham, S. (2009). Marking and moderation in the UK: false assumptions and wasted resources. Assessment and Evaluation in Higher Education, 34(2), 209 to 220.
See how this works in practice
Book a demo, or create a free lecturer account and assess your first cohort.