Presentation (CTI)

OBJECTIVES: Claims that AI saves time in HEOR are common, but most count only the junior analyst hours removed from a task, and miss two silent costs: the compute (token) cost of running the AI, and the senior-reviewer time AI quietly adds. We quantified the net economics of AI-assisted HEOR once these are counted, and argue the findings break the hour-based pricing model on which HEOR consulting still runs.
METHODS: Across completed AI-assisted deliverables spanning SLR, cost-effectiveness modelling, model validation, and technical reporting, we measured effort and cost with and without AI. For each we recorded gross analyst time saved, then subtracted the offsetting costs: token spend, and the added senior review, validation, and sign-off time, which AI shifts from junior production toward expensive expert verification. Net cost was gross saving minus these. We examined whether time-based billing still reflects the value of AI-assisted work.
RESULTS: Counting only analyst hours overstated the benefit in every deliverable. AI compressed junior production time sharply but reallocated effort upward: a senior health economist spent more time reviewing AI output than before, and token cost was non-trivial at scale. For repeatable, high-volume, specified tasks net economics were strongly favourable, as token and review cost amortised across many items. For one-off, judgement-heavy tasks, a deliverable produced in an hour still carried hours of senior verification and real compute cost, so true cost and billed time diverged sharply.
CONCLUSIONS: When a task is delivered in an hour but consumes senior expert review and real compute, hour-based billing misprices the work, undercharging for embedded expertise and value. The economics point toward outcome- and value-based pricing in HEOR: charging for the validated deliverable and the decision it supports, not hours at the desk.