A Johns Hopkins-led study in JMIR Aging finds that AI health tools for older adults are being pulled in incompatible directions by the people expected to use, buy, build and deploy them—a divide that can turn technically capable products into expensive, inaccessible deployments.
As reported by Medical Xpress and EurekAlert, researchers from Johns Hopkins University, the University of Iowa and Washington University in St. Louis interviewed 49 stakeholders across six groups: older adults, care partners, clinicians, health-system leaders, payers, developers and investors. All cited cost, usability and value as decision factors. They did not mean the same thing by any of those terms.
For older adults and care partners, the immediate concerns were out-of-pocket expense and whether a product works with physical or sensory limitations. Clinicians emphasized affordability for patients and avoiding another workflow burden. Health systems and payers looked at interoperability, return on investment and whether a tool could reduce costly health events. Developers and investors, meanwhile, focused on addressable market, scale and margins sufficient to carry regulatory and development risk.
The study’s most important finding for technology teams is that usability cannot be treated as a late-stage interface pass. Older users may need accessible interaction models, clearer onboarding, lower-friction authentication and support for sensory or mobility limitations before an AI-driven service has practical value.
That creates a familiar enterprise-IT problem. A platform can perform well in a pilot, integrate with a clinical backend and still fail in the real world if its user experience assumes high digital confidence, modern hardware, reliable connectivity or an ability to absorb recurring subscription costs.
The researchers described end-user frustration with tools seen as “solutions in search of a problem.” That criticism is pointed at a common AI product-development pattern: adapting an existing model or platform to a health-care category instead of starting with the specific tasks, constraints and priorities of the people who will rely on it.
Developers and investors told researchers that regulatory timelines of four to seven years, alongside substantial financial risk, create pressure to pursue large returns. The result can be a product strategy optimized for scale and margins rather than narrow but important needs in aging care.
For IT leaders evaluating AI-enabled health products, that makes total cost of adoption more useful than a standard licensing comparison. The calculation should include patient-facing fees, accessibility accommodations, staff training, clinical workflow changes, identity management, data integration and the support burden created when users cannot successfully operate the tool on their own.
That recommendation is less about making every project slower and more about preventing expensive rework. In practice, an AI health pilot for older adults needs older participants and care partners involved before requirements harden—not simply as a final usability panel after the underlying product, pricing and data flows are already fixed.
The immediate lesson is straightforward: AI health tools for aging populations will not succeed merely because they are accurate or well integrated. They must be affordable to patients, workable for clinicians, defensible for health systems and designed around the people expected to use them.
For older adults and care partners, the immediate concerns were out-of-pocket expense and whether a product works with physical or sensory limitations. Clinicians emphasized affordability for patients and avoiding another workflow burden. Health systems and payers looked at interoperability, return on investment and whether a tool could reduce costly health events. Developers and investors, meanwhile, focused on addressable market, scale and margins sufficient to carry regulatory and development risk.
The accessibility requirement is not a feature request
The study’s most important finding for technology teams is that usability cannot be treated as a late-stage interface pass. Older users may need accessible interaction models, clearer onboarding, lower-friction authentication and support for sensory or mobility limitations before an AI-driven service has practical value.That creates a familiar enterprise-IT problem. A platform can perform well in a pilot, integrate with a clinical backend and still fail in the real world if its user experience assumes high digital confidence, modern hardware, reliable connectivity or an ability to absorb recurring subscription costs.
The researchers described end-user frustration with tools seen as “solutions in search of a problem.” That criticism is pointed at a common AI product-development pattern: adapting an existing model or platform to a health-care category instead of starting with the specific tasks, constraints and priorities of the people who will rely on it.
ROI calculations can obscure the patient bill
The study also exposes a tension that health-system procurement teams will recognize. A tool that improves operations or reduces downstream costs can look attractive at the system level while remaining unaffordable or difficult to use for an individual patient.Developers and investors told researchers that regulatory timelines of four to seven years, alongside substantial financial risk, create pressure to pursue large returns. The result can be a product strategy optimized for scale and margins rather than narrow but important needs in aging care.
For IT leaders evaluating AI-enabled health products, that makes total cost of adoption more useful than a standard licensing comparison. The calculation should include patient-facing fees, accessibility accommodations, staff training, clinical workflow changes, identity management, data integration and the support burden created when users cannot successfully operate the tool on their own.
Early design reviews may matter more than model selection
Lead author Zhang Zhang and senior author Nancy L. Schoenborn argue for earlier engagement among stakeholders, clearer public education around AI, and public-private partnerships that reduce risk in early development. The Johns Hopkins Artificial Intelligence and Technology Collaboratory for Aging Research is one example cited by the team.That recommendation is less about making every project slower and more about preventing expensive rework. In practice, an AI health pilot for older adults needs older participants and care partners involved before requirements harden—not simply as a final usability panel after the underlying product, pricing and data flows are already fixed.
The immediate lesson is straightforward: AI health tools for aging populations will not succeed merely because they are accurate or well integrated. They must be affordable to patients, workable for clinicians, defensible for health systems and designed around the people expected to use them.
References
- Primary source: Medical Xpress
Published: 2026-07-30T15:20:05+00:00
- Independent coverage: EurekAlert!
Published: 2026-07-30T15:20:31.654897
