Healthcare and the AI velocity gap

<h2><span class="h6">Health systems aren’t short on AI ambition; they’re short on the capacity to absorb AI into constrained, regulated, daily work. That gap is where the next phase will be won or lost.</span></h2>

<p>Healthcare is no longer waiting for better AI. It is waiting for operating models that can survive contact with it.</p> <p>This may sound backward in a year when models keep improving, agentic systems have entered the enterprise vocabulary and every major technology provider presents healthcare as the natural proving ground. Yet the constraint inside most health systems is not awareness. Leaders already see the promise across access, revenue cycle, documentation, contact centers, care coordination and payer friction that consumes too much labor for too little return.</p> <p>The harder question is whether the organization can absorb the technology into the work.</p> <p>The Centers for Medicare &amp; Medicaid Services’ CMS-0057-F interoperability regulation makes that question concrete. Finalized in 2024, with key operational provisions beginning in 2026 and API requirements following in 2027, the rule compresses prior authorization decision timelines, requires specific denial reasons, expands public reporting and pushes the industry toward Fast Healthcare Interoperability Resources (FHIR)-based exchange.</p> <p>Technically, much of the rule is aimed at payers. Operationally, providers will feel it through intake, documentation, evidence assembly, follow-up, denials, appeals and the daily work of moving patients through care without additional administrative drag.</p> <p>No AI model decides whether an organization meets that standard. A workflow does.</p> <p>Someone must change the process, watch outputs, handle exceptions and own the result when the system is incomplete, uncertain or wrong. This is the work regulation exposes, and it is also the work already-stretched teams are least able to absorb.</p> <p>This is the <a rel="noopener noreferrer" href="/content/cognizant-dot-com/us/en/insights/insights-blog/closing-gap-between-ai-infrastructure-investments-and-business-value-realization.html" target="_blank">AI velocity gap</a>: AI capability advancing faster than healthcare organizations can safely and consistently operationalize it.</p> <p>The gap is easy to underestimate because early AI use cases appear reassuringly concrete. A model can summarize a record, review a packet, draft an appeal, aggregate pre-visit information or support a contact-center agent. These are important capabilities, but they are not yet an operating model.</p> <p><a rel="noopener noreferrer" target="_blank" href="https://blog.providence.org/national-news/providence-study-finds-ai-ambient-listening-tool-modestly-reduces-documentation-burden-improves-provider-efficiency">Providence’s recent real-world ambient AI study</a> makes the point: reduced documentation burden matters, but the relevant signal was a statistically significant relative value unit increase without additional appointments. Recovered time only becomes strategic when the operating model can convert it into capacity.</p> <p>The more difficult work begins when this output must move through handoffs, policies, audits, exceptions and teams already operating near their limit, without creating new ambiguity. The next phase of healthcare AI will not be won by the largest experimental portfolio. It will be won by the best absorption architecture.</p> <h3><span class="h4">AI builders and the healthcare AI velocity gap</span></h3> <p>That is where <a rel="noopener noreferrer" href="/content/cognizant-dot-com/us/en/insights/insights-blog/why-ai-builders-are-essential-for-large-enterprises.html" target="_blank">the AI builder model</a> earns its place. Hyperscalers provide essential infrastructure, model access and platform ecosystems. Consultancies help frame strategy, governance and transformation choices. Systems integrators connect applications, data and workflows. While healthcare requires all three, none of them alone can close the gap.</p> <p>An AI builder is capable of remaining close enough to the work both to change it and to validate whether the change performs. This is an evolution of the integrator role, not a rejection of it. In practice, that requires uniting healthcare context, engineering, workflow redesign, controls and accountability after release. The practical meaning of AI builder is connecting systems end to end; grounding the work in clinical and operational reality; and measuring whether the work performs after launch.</p> <p>The most useful opportunities are rarely the most visible. They reside in the burdened middle of the health system, where work is repetitive enough to benefit from AI, variable enough to require judgment and consequential enough to demand oversight.</p> <p>Examples include patient access, referral intake, prior authorization support, denials prevention, documentation review, discharge follow-up and contact-center operations. These activities shape how patients move through the system, define how much administrative drag clinicians carry and determine how much revenue leaks through preventable defects.</p> <p>AI helps in these workflows only when the workflow itself is made fit for purpose. Otherwise, it becomes another brittle layer on a process that already has too many. This creates the capacity question. If AI requires redesign, monitoring, exception handling, data stewardship, performance measurement and continuous improvement, where does that work live? It cannot be carried by the same teams already managing the daily load, nor will it mature through scattered pilots across facilities and markets.</p> <h3><span class="h4">The role of the healthcare GCC &nbsp;</span></h3> <p>Global capability centers deserve a serious conversation here.</p> <p>GCCs are well understood in parts of healthcare, <a rel="noopener noreferrer" href="/content/cognizant-dot-com/us/en/insights/insights-blog/role-of-global-capability-centers-in-life-sciences.html" target="_blank">particularly across life sciences</a>, payers and large technology-enabled healthcare enterprises. Provider health systems are a less mature category. There are important exceptions, but many provider organizations are only now considering whether GCCs can support clinical-adjacent services, administrative operations and AI-enabled workflows in a more structured way. That timing matters. The point is not to send yesterday’s process somewhere cheaper. The point is to build durable capability around the work AI is now exposing.</p> <p>Done well, a GCC works in tandem with the core health system rather than as a distant back office. The core site retains clinical accountability, patient-facing judgment and ownership of outcomes. The GCC provides persistent capacity for engineering, workflow redesign, monitoring, data stewardship and operational improvement that daily teams rarely have time to build.</p> <p>That model requires more than technical staffing. It requires people who understand utilization management, coding, appeals logic, interoperability standards, care-coordination workflows and the operational consequences of getting the work wrong, working alongside engineers who build around that operating reality. A GCC should never become a lower-cost destination for broken processes. That only exports the inefficiency.</p> <p>This is the practical connection between the two strategies. The AI builder supplies the method. The GCC supplies the capacity.</p> <p>Consider CMS-0057-F again. Meeting a seven-day clock at scale is not a single deployment. It is a standing operation: accepting electronic requests, routing work, drafting and checking responses, surfacing exceptions that need human review, documenting reasons and reporting performance. Built well in a GCC, that becomes a monitored, improvable line of work. Pushed back onto an already stretched utilization management, access or revenue cycle team, it becomes the next thing that breaks.</p> <p>None of this scales without governance designed at the start. Role clarity, auditability, human oversight, escalation paths, data controls and performance standards are not compliance ornaments. They are how trust survives contact with daily use. A clinician, revenue cycle leader, compliance officer or operations executive should not have to guess whether an AI output is advisory, determinative, incomplete or unsafe. Those are workflow responsibilities.</p> <p>AI without ownership is only automation with plausible deniability.</p> <h3><span class="h4">Becoming AI ready in healthcare</span></h3> <p>Healthcare leaders should be wary of viewing AI readiness as a technology inventory. The better test is whether the organization can absorb AI into real work, govern it, measure it, improve it and provide someone capable of owning the result when it does not go to plan.</p> <p>More impressive models will not close the velocity gap. The gap will close when health systems build the structure to make AI useful under the actual conditions of care: constrained labor, regulated workflows, uneven data and zero tolerance for elegant failure.</p>