Cambridge Assessment Network · 2025 to present
Keep expert judgement in the loop.
An accredited three-session series on using generative AI for formative assessment design, built so practitioners can generate items, then check them, rather than handing the work to a model.1
My role: I designed the course and have delivered four cohorts as the named Cambridge trainer. The Assessment Network owns the public programme, booking and accreditation.
The need
Assessment teams were being asked to use AI without a professional method.
The Assessment Network’s 2024 customer survey found that more than 70 per cent of assessment professionals wanted to learn more about the challenges and opportunities of AI in assessment.1
The course treats AI as a support for item generation and workflow, not a replacement for subject expertise. Security, privacy, bias and quality assurance sit inside the practice, not after it.1
Design problem
How could a short, live series give mixed-experience practitioners a repeatable way to generate assessment content with AI, then reject or repair what is not good enough?
What I designed
A sequence that ends in verification, not a demo
Each session builds on the last. Participants generate questions from structured inputs, evaluate them, and leave with a checklist for AI-generated assessment content.1
Make the model the junior partner
The public course promise is that practitioners remain the expert in the human-AI relationship. The live work is built around that claim: generate, then judge.
Use any capable model
The sessions teach principles that apply across large language models, so the method does not collapse when a tool interface changes.
Put QA in the same room as generation
Bias, curriculum alignment, distractors, feedback and limitations are practised as part of the workflow, not as a warning slide at the end.
Leave something reusable
Participants receive resources for their own context, including a checklist for AI-generated assessments, and need access to an AI tool throughout.
Course architecture
Three live workshops
Six accredited CPD hours. A certificate requires attendance at two live sessions.1
From a participant
The useful test is whether people still trust their own judgement.
“I now recognise that I am still the expert in the human-AI relationship and that, with the right guidance and some experimentation, I can become more confident in using it and reaping its benefits, whilst having the processes in place to guard against some of its challenges.”
What I took from delivering it
Three design lessons
- Confidence is a professional outcome, not a mood. People need a method they can defend to colleagues, not a set of clever prompts.
- Generation without review teaches the wrong habit. The valuable part of the course is slowing down after the model has drafted.
- A short accredited series has to travel. The same spine has to work for teachers, publishers, learning designers and assessment managers.
Attribution: Designing assessments with AI is an Assessment Network programme. This case study covers my design and delivery of the series. Booking, membership discounts and current dates are on the public course page.
Sources and further reading
Explore more work
See other programmes spanning teacher capability, system leadership and publishing.