AI in education: evaluate the workflow beyond the demo.

A convincing demonstration is a starting point. An education organisation also needs to understand whether a tool fits its curriculum, staff capacity, learner needs and responsibilities for data.

Choose one workflow

Lesson preparation, feedback support and administration have different users and risks. Select a bounded task and identify who will review the output. For lesson preparation, that might mean a teacher checking curriculum alignment, factual accuracy and suitability for the class.

Distinguish efficiency from learning

Saving preparation time does not, by itself, demonstrate better learning. Evaluate staff efficiency and educational outcomes separately. Claims about improved learning require an appropriate study design and evidence beyond user satisfaction.

Include the difficult cases

Test ambiguous questions, different levels of prior knowledge and incomplete information. Check whether the tool gives useful help when it is uncertain. Involve educators in designing the evaluation and interpreting the results.

Plan for responsible use

Establish what information may be used, who can access it and how long it is retained. Begin with synthetic or suitably de-identified examples where possible. Make the tool's limits visible to staff and keep consequential judgments with the appropriate people.

These questions form an initial research agenda. They do not establish that a particular tool improves teaching or learning.