Presentation (CTI)

OBJECTIVES: A proof-of-concept (PoC) shows that an AI tool can work under favourable conditions, not that it is a viable product, a gap routinely underestimated in HEOR, where promising demonstrations are mistaken for production readiness. Drawing on patterns across our own AI tool-building, we developed a reusable maturity framework mapping the PoC-to-production journey through three gated stages.
METHODS: A structured framework-development exercise synthesised lessons from multiple AI-enabled HEOR tools spanning evidence synthesis, economic modelling, and technical reporting. For each tool, the conditions under which the PoC succeeded were contrasted with those required for dependable repeated use, and the points at which tools broke during scale-up were grouped into product-viability dimensions. These were arranged into three maturity stages: proof-of-concept (capability shown), validated tool (reliability shown), and governed production system (dependable operation shown). Each transition was defined by engineering-readiness dimensions expressed as conceptual criteria, not numeric thresholds: representativeness of evaluation inputs; robustness to messy inputs, edge cases, and drift; reproducibility and version stability; observability and traceability of outputs; integration with existing workflows and quality processes; and clear ownership, monitoring, and maintenance once live.
RESULTS: The framework specifies viability as a staged sequence rather than a single demonstration. Stage 1 establishes that the tool can perform the task; its gate rejects curated, single-run success as readiness. Stage 2 requires reliable performance on representative, unseen inputs, predictable edge-case behaviour, and reproducible outputs across runs and versions. Stage 3 requires dependable operation in real workflows: monitored for drift, maintainable, observable, and owned, with human oversight where outputs feed downstream analysis. Each gate stops a capable but unreliable tool from advancing.
CONCLUSIONS: In HEOR, a PoC proves possibility, not product viability. Mapping the journey as staged, gated progression turns ad hoc tool-building into a repeatable engineering discipline and a shared standard for when a tool is ready to rely on.