AI can’t overcome healthcare’s data integration problems on its own

Board decks are filled with demos of ambient documentation, virtual assistants, and predictive models that promise to ease clinician burnout and make care more proactive. Health systems are spending heavily to secure pilots, platforms, and partnerships—often under pressure to show momentum quickly.

Then the tools arrive and things don’t go as planned. 

Clinicians discover that AI-generated notes don’t flow cleanly into the EHR. Predictive models can’t see key lab results. Automation stalls when it hits a system that hasn’t been updated in years. What looked seamless in a demo becomes brittle in production.

The problem isn’t that the AI doesn’t work. It’s that the data can’t move.

According to MuleSoft’s 2025 Connectivity Benchmark Report, 95 percent of IT leaders say integration challenges are a primary barrier to deploying AI effectively. Healthcare’s age-old problem, that patient data remains scattered across hundreds of disconnected systems, bogs down successful AI deployment.

A sprawl no model can span

According to the report, the average enterprise now runs 897 applications, with 46 percent of enterprises managing more than 1,000. Yet 71 percent of those applications remain unintegrated. For healthcare organizations already burdened by fragmented electronic health record systems and decades of accumulated legacy technology, the result is a patchwork environment that resists automation.

Disconnected systems create blind spots that no AI model can fix. Data locked inside lab platforms, imaging archives, billing systems, and clinical notes cannot be meaningfully analyzed if it cannot move freely.

“There’s no AI without APIs,” said Karim Trojette, global MuleSoft alliance leader at Deloitte Digital. “AI can only be effective when the infrastructure supports it. That means having an enterprise-wide strategy that integrates every app and system.”

Without that foundation, even the most advanced tools stall at the point of deployment.

AI adoption is accelerating faster than organizations can handle

Healthcare AI adoption surged to 22 percent of organizations in 2025, more than double the rate seen across the broader economy, according to Menlo Ventures research. Health systems now lead adoption at 27 percent, followed by outpatient providers at 18 percent and payers at 14 percent. Spending reached $1.4 billion this year, nearly tripling investment from 2024.

Yet adoption without integration introduces new risks. MuleSoft found that 81 percent of IT leaders struggle specifically to use AI for system integrations. When models lack access to complete, real-time patient data, their outputs become partial at best. Predictive tools that cannot simultaneously draw from labs, imaging systems, and clinical documentation produce results clinicians cannot trust.

A 2024 federal analysis found that 71 percent of nonfederal acute care hospitals reported using predictive AI embedded in their EHRs, up from 66 percent the year before. But the gains were uneven. Larger health systems advanced faster than smaller, rural, or critical access hospitals. Without robust integration, the gap between well-resourced organizations and the rest of the healthcare system is likely to widen.

Automation is breaking down at the bedside

Physicians using AI-powered documentation tools often discover that automation ends abruptly at system boundaries. Notes generated in one platform must still be manually transferred into another.

A fall 2024 survey of 43 U.S. health systems found that while every organization had adopted or was piloting ambient clinical documentation tools, outcomes varied widely. Only 53 percent reported a high degree of success, and many pointed directly to integration challenges as the limiting factor.

Researchers studying healthcare interoperability reached similar conclusions. Patient data remains dispersed across institutions with incompatible systems and inconsistent standards. While Fast Healthcare Interoperability Resources has improved data exchange, it cannot resolve situations where records are fragmented across organizations using different implementations and governance models.

The result is technology that works in theory but breaks down in practice.

Stretched IT teams face mounting strain

MuleSoft reports that 86 percent of IT leaders expect workloads to increase in the coming year, and 3 in 10 say projects were delivered late in 2024. Maintaining aging systems while simultaneously deploying AI creates a cycle of delay and technical debt.

Organizations estimate that developers now spend 39 percent of their time building custom integrations simply to deliver new digital capabilities. The financial implications are significant. APIs and API-related implementations are now tied to 40 percent of company revenue, up from 25 percent in 2018. Yet 87 percent of respondents say their API management practices need improvement, and 91 percent believe working with third parties would help maximize return on investment.

AI, in other words, is amplifying existing weaknesses rather than masking them.

Autonomous agents raise the stakes

The pressure is intensifying as autonomous AI agents move from experimentation to deployment. MuleSoft found that 93 percent of IT leaders plan to implement autonomous agents within two years, and 40 percent already have them in production. These systems promise to coordinate workflows independently, scheduling care, managing claims, or routing tasks without human intervention.

But autonomy requires access. Agents must pull data from every relevant system in real time. Without seamless integration, they remain isolated tools rather than operational breakthroughs.

Healthcare faces additional hurdles. Privacy regulations impose governance requirements that other industries can sometimes avoid. Infrastructure designed decades ago for paper workflows resists modern integration patterns. Clinicians, understandably, demand reliability before trusting AI systems that influence patient safety.

Strategy beats speed in AI deployment

A 2025 World Economic Forum analysis of healthcare AI adoption found that successful deployments share a common trait: leaders paused to build foundational infrastructure before scaling AI.

The MuleSoft data suggests this shift is underway, though unevenly. About 65 percent of organizations report having complete or nearly complete strategies to enable low-code and no-code automation for nontechnical users. Roughly 70 percent have centralized IT governance for automation across the enterprise. These approaches slow early experimentation but create conditions for sustainable growth.

Only 28 percent of respondents say their leadership has implemented a clear, organization-wide API strategy. Seven percent report having no comprehensive plan at all, with each project developed in isolation.

The work ahead involves standardizing data formats, building APIs for real-time communication, and establishing governance frameworks that balance innovation with compliance. These efforts demand sustained investment and executive attention, and organizations that avoid this important work will remain stuck with AI they’ve purchased but cannot deploy effectively.