EdTech products are usually sold to administrators and used by teachers, which is why so many are purchased and then quietly abandoned. The adoption question is narrow: does this remove work in the first week, or add it?
Replace a task, do not add a tool
A product that requires preparation before it helps loses to the existing method, which requires none. The products that stick take over something teachers already do — building assessments, marking, chasing attendance, writing reports — and return the time immediately.
A teacher will not invest an hour today for an hour saved next month. The saving has to be now.
Curriculum alignment must be modelled
Content that is topically plausible but not aligned to the board and grade is unusable. This means modelling the syllabus structure explicitly rather than relying on a model's general knowledge — which is exactly what we had to build for TestGenie across 18+ boards, and it is the difference between output that is used and output that is checked and discarded.
Design for shared and low-end devices
A computer lab machine used by six classes, a teacher's three-year-old phone, a school connection shared across forty devices. Heavy applications fail here, and failing here means the product is unusable for most of the market it claims to serve.
Generated content needs an edit path
Teachers will not use output they cannot change. Anything generated — a question, a report comment, a lesson outline — must be editable, and the edits should improve future suggestions. Presenting AI output as final removes the professional judgement the teacher is accountable for.
Children's data deserves a stricter default
Collect the minimum, define retention explicitly, obtain parental consent where required, and be careful about what analytics reveal about an identifiable child. The safest position is to collect less than you are permitted to, and to be able to explain every field you do keep.
Sell to administrators, design for teachers
Budget authority and daily use sit with different people, and building for the buyer produces the compliance dashboards that go unused. Serve the administrator's reporting needs as a by-product of teachers using the system, not as the primary design goal.
