A new framework gives hospitals a way to weigh AI beyond cost

Hospitals have started adopting clinical AI faster than they can evaluate it, and most of that evaluation still runs through one filter: cost. A framework published this month in npj Digital Medicine by R. Andrew Taylor of the University of Virginia and Arwen B.L. Declan of Clemson gives administrators a structured way to ask what cost-only analysis hides.

The framework is called Total Mission Value (TMV), and it is built for the procurement meeting as much as for the strategy retreat. In a Perspective published 11 June 2026, the two emergency-medicine researchers layer five domains onto a hospital's existing mission: Patient Care, Staff Experience, Operations, Economic, and Education and Research. The structure borrows from the Balanced Scorecard and the Quintuple Aim, two tools that already circulate in hospital boardrooms, and adds the AI-specific tradeoffs those tools were not built to capture.

The design is a pyramid. Patient care and patient experience sit at the top. Ethics is the foundation. Economic sustainability is the supporting base. The placement is intentional: an AI tool can clear the budget filter and still fail the patient or the staff, and TMV is built so that failure shows up before a contract is signed. The authors' stated constraint is that AI should support, not replace, care providers, and that constraint runs through every layer.

Taylor is vice chair of research and innovation in UVA's Department of Emergency Medicine. Declan is a clinical assistant professor in Clemson's School of Health Research and is also affiliated with Prisma Health–Upstate and the University of South Carolina School of Medicine Greenville. The pair's stated motivation is the gap between AI's adoption curve and the slowness of hospital governance. Standard technology-evaluation playbooks, they write, center on cost, and the things cost analyses tend to render invisible are the things AI introduces first.

Those four risks are bias, opacity, workforce displacement, and erosion of the patient-clinician relationship. None of them shows up in a vendor's price-per-prediction slide. The framework's contribution is putting all four in the same room as that slide, with a structure for weighing them against a hospital's mission. A tool that meets its accuracy target and adds hours to the nursing workflow, for instance, would pass a cost-only review on Economic and Patient Care and fail under TMV, because the Staff Experience domain would force the workflow impact into the decision.

The framework is a proposal, not a validated tool. The authors are explicit that TMV draws on existing management frameworks rather than introducing new empirics, and it has not been tested in a procurement committee. The most concrete evidence of uptake is regional: 29news, a Charlottesville ABC affiliate, picked up the announcement on 21 July, and an Augusta Free Press reprint carried it as well. The coverage trail shows the framework is being read, not yet being run.

The competing-interest disclosure belongs on the same page as the recommendation. As the paper notes, Taylor has received a Beckman Coulter grant to evaluate TriageGo at Yale New Haven Health System and serves as an advisor to VeraHealth; Declan declares no competing interests. The disclosure matters because TMV is a vendor-agnostic decision aid, and the lead author's industry ties should travel with the framework. Neither tie is to a TMV product; both are disclosed in the published paper.

For hospital administrators, the practical read is short. The next AI pitch that lands on a procurement committee does not have to be evaluated on cost, throughput, and accuracy alone. TMV gives a structure for the question the standard playbook dodges: what does this tool cost the mission, not just the budget.