BUSINESS CASE · COSTS

How Much Does AI Grading Actually Save? A Cost Per Assignment Breakdown

A deliberately narrow, mechanical method for turning your own numbers into a figure a finance team will accept.

By Eduface · July 2026 · 8 min read

Our companion piece on building the business case for AI assessment makes the qualitative argument. This one is deliberately narrower and more mechanical: a worked method for turning your own numbers into an actual cost-per-assignment figure, rather than relying on a vendor’s headline percentage that may or may not reflect your situation.

Quick answer

The only cost figure worth acting on is one built from your own inputs: the fully loaded assessor cost per assignment today, the same figure with AI-assisted marking including a realistic human review step, and the difference multiplied by your actual annual assignment volume.

Why a borrowed number is the wrong number

Every vendor in this space, ourselves included, can point to case studies with impressive-sounding time or cost savings. The honest problem with using someone else’s figure directly is that marking time varies enormously by discipline, assignment length, and how experienced your specific assessors are, which means a percentage that is accurate for a first-year multiple-choice-heavy module tells you very little about what will actually happen with a cohort of dissertation-length business reports. The only cost figure worth acting on is one built from your own institution’s actual numbers.

The basic structure of the calculation

The calculation has three components: the fully loaded cost of an assessor’s time per assignment under your current process, the fully loaded cost per assignment once AI-assisted marking is in place, and the difference between the two, multiplied by your actual assignment volume. Each of those needs its own honest inputs.

Current cost per assignment

Start with your assessor’s fully loaded hourly cost, salary plus employer costs, not just headline pay, divided into an hourly rate. Multiply that by the actual average time spent marking one assignment of the relevant type, timed realistically rather than estimated from memory, since most people underestimate how long thorough marking actually takes. Add any moderation or second-marking time that applies under your current process, since that is a real cost too and should not disappear from the comparison just because it is a separate line item.

Cost per assignment with AI-assisted marking

This has two components of its own: the tool cost itself, allocated per assignment based on your actual volume and pricing structure, and the reduced but non-zero human time for the review and approval step our companion piece on human-in-the-loop marking describes. That review time is not negligible, and treating it as such produces a number nobody will trust once it does not match reality. A realistic review time, timed the same way you timed the original marking, not guessed at, is what makes this comparison credible rather than a marketing exercise.

The volume multiplier

Multiply the per-assignment difference by your actual annual assignment volume for the relevant category, not your total student headcount, since the two are not the same once you account for multiple assessments per module and modules per programme.

A worked example, with placeholder numbers you should replace

To make the method concrete, here is the same calculation with illustrative inputs. Every number in it is a placeholder. Replace the hourly rate, the current marking time, the review time, and the tool cost with your own measured figures, and the structure of the calculation will hold even though the answer will not match.

Line

Current process

With AI-assisted marking

Fully loaded assessor cost

GBP 35 per hour

GBP 35 per hour

Time per assignment

25 min marking

8 min review and approval

Human cost per assignment

GBP 14.60

GBP 4.70

Tool cost per assignment

GBP 0

GBP 2.00 (illustrative)

Total per assignment

GBP 14.60

GBP 6.70

Across 3,000 assignments a year

GBP 43,800

GBP 20,100

That is a saving of just under GBP 8 per assignment, or close to GBP 24,000 annually at that volume, before accounting for the harder-to-quantify value of faster turnaround itself, which our companion piece on why slow marking costs providers goodwill argues is often the bigger prize.

What this calculation deliberately leaves out, and why that matters

This method only captures direct time cost. It does not put a number on faster feedback improving retention, on reduced marking-related staff burnout, or on the reputational cost of slow turnaround covered elsewhere on this site. Those are real, and arguably larger over time than the direct cost saving, but they are harder to quantify credibly, and bundling a soft, hard-to-verify number into a hard cost figure tends to undermine trust in the whole calculation. Keep the two arguments separate: a defensible, conservative direct cost saving, and a qualitative case for the harder-to-quantify benefits alongside it.

Frequently asked questions

Should we trust a vendor’s published cost-saving percentage for our own budgeting?

Treat it as a rough indication at best. Marking time and cost vary too much by discipline and assessment type for a borrowed figure to be reliable for your specific business case.

What is the biggest mistake institutions make when building this calculation themselves?

Treating the human review time under AI-assisted marking as negligible or zero. It is not, and a calculation that ignores it will look better on paper than it performs in practice, which damages trust in the whole business case.

Should soft benefits like retention or staff wellbeing be included in the cost figure?

Keep them as a separate, qualitative argument rather than folding them into the hard cost number. Mixing a conservative, verifiable calculation with a speculative one weakens the credibility of both.

How often should this calculation be revisited?

At least annually, and whenever assignment volume, staffing costs, or your tool’s pricing structure changes meaningfully, since all three inputs can shift the answer substantially.

Run the numbers on your own volume

Tell us your assignment volume and assessor cost and we will walk the calculation through with you. Book a demo or start free.