Short answer: around 2–6 hours a week per teacher, once it's embedded. That figure comes from schools using AI consistently on the right tasks — planning first drafts, resource creation, parent letters, feedback phrasing — after proper training. The range is wide because usage is: a teacher using AI daily on real tasks saves hours; one who tried it twice in September saves nothing.
Where the hours actually come from
The savings concentrate in tasks that are repetitive and text-heavy: a letter that took twenty minutes takes five; a differentiated worksheet set that took forty-five minutes takes fifteen; a bank of report comment structures takes an evening once instead of every time. None of these touch the parts of the job that need to stay human. Our guides to planning, resources and feedback break down each one.
The honest caveat
The gap between potential and actual saving is training. One-off exposure produces a novelty spike that fades by half-term; embedded workflows produce the numbers above. If your school wants the 2–6 hours, budget for the training, not just the tool. Our guide to building an AI workflow shows what embedding looks like.