Adaptive platforms promise to route every learner to the right next task. But the routing is only as good as the questions it routes on.
An adaptive engine learns from item-level data: who answered what, how confidently, how quickly. If those items test recall rather than the real task, carry ambiguous wording, or sit untagged in a bank nobody has audited, the system will still make confident decisions. It simply makes them from weak evidence, then scales the error learner by learner.
That is the quiet risk in AI-driven personalisation: it amplifies your assessment design rather than correcting it.
- Trainers and subject experts: write items from real work decisions, not from slides.
- Instructional designers: specify what each item measures, plus difficulty and tags.
- L&D managers: audit the item bank before you buy the adaptivity.
- HR leaders: accept a slower rollout for a defensible one.
Before you personalise the path, ask what the path is built on.