Healthcare AI Progress Depends on Data Strategy First

Healthcare’s AI investment surge is colliding with a familiar constraint: the industry is trying to deploy advanced automation on top of fragmented, aging, and poorly synchronized data systems. That mismatch may explain why so much AI activity feels busy but not transformative.

The message from executives speaking at the HFMA annual conference was direct. Oracle Health, HCA Healthcare, and BJC HealthCare leaders framed the problem less as a shortage of AI tools and more as a shortage of usable, real-time, connected data. The distinction matters because healthcare organizations can buy algorithms faster than they can rebuild the infrastructure those algorithms need to perform safely.

A health system can deploy AI for clinical support, claims management, documentation, scheduling, coding, denials, and payer communication. But if those tools are working from incomplete records, delayed eligibility data, inconsistent pharmacy information, disconnected financial systems, and poorly governed operational feeds, the result is not intelligence. It is automation layered over uncertainty.

AI Cannot Fix Data It Cannot Trust

The most important AI question in healthcare is not which model performs best in a demonstration. It is whether the data available at the moment of use is accurate enough, current enough, complete enough, and governed well enough to support the decision being made.

A medication recommendation may require clinical history, allergy data, formulary status, prior authorization rules, drug availability, patient affordability, and interaction risk. A discharge planning tool may need bed capacity, post-acute options, payer requirements, transportation barriers, and home support. A revenue cycle tool may need documentation, eligibility, contract terms, payer edits, medical necessity criteria, and denial history.

Most health systems still struggle to unify these layers. Clinical data may sit in the EHR. Coverage data may sit with payers. pharmacy data may sit elsewhere. Operational data may live in staffing, scheduling, and capacity systems. Financial data may move through separate billing, claims, and contract management platforms. AI can identify patterns, but it cannot safely infer missing context that the organization failed to make available.

This is why data strategy has to precede AI strategy. The Office of the National Coordinator for Health Information Technology has continued to emphasize nationwide interoperability through standards such as the United States Core Data for Interoperability and exchange frameworks such as TEFCA. Those efforts are not abstract policy projects. They are part of the foundation healthcare AI needs before it can move from narrow tasks to coordinated care improvement.

Bot Versus Bot Is a Warning Sign

The “bot versus bot” image in revenue cycle describes a real risk. Providers are deploying automation to code claims, prevent denials, appeal rejections, and recover payment. Payers are deploying automation to review claims, detect improper billing, manage utilization, and deny payment when documentation or policy conditions are not met.

That arms race may increase throughput without improving trust. If provider bots generate claims that payer bots deny, and payer bots generate denials that provider bots appeal, healthcare has not solved administrative waste. It has industrialized it.

The financial stakes are substantial. Revenue cycle friction consumes staff time, delays cash flow, increases administrative expense, frustrates patients, and diverts leadership attention from care delivery. AI can reduce that burden only if organizations use it to clarify rules, improve documentation, reduce avoidable errors, and resolve disputes earlier. If AI simply accelerates adversarial behavior between payers and providers, cost shifts rather than falls.

This is especially important as the Centers for Medicare & Medicaid Services pushes electronic prior authorization and broader data exchange through its Interoperability and Prior Authorization Final Rule. The rule’s API requirements may improve transparency and reduce manual processes, but they will not produce meaningful relief if payer and provider systems continue to operate from different definitions, incomplete data, and weak workflow integration.

Clinical AI Needs the Strongest Data Foundation

The prioritization argument raised by health system leaders is sound: clinical systems should come first. AI that touches care delivery carries a higher risk profile than tools used only for internal productivity. A coding recommendation can create compliance risk. A clinical recommendation can affect diagnosis, treatment, safety, and outcomes.

Clinical AI must therefore be evaluated in context. Performance in a vendor test environment does not guarantee performance in a local hospital, outpatient practice, emergency department, or specialty clinic. Data quality varies by site, specialty, documentation pattern, patient population, device integration, and workflow design.

A sepsis model, for example, may depend on timely vitals, lab values, medication history, and documentation. A readmission model may require social determinants, prior utilization, discharge disposition, and payer data. A clinical documentation tool may affect how diagnoses are represented, how risk is captured, and how downstream care teams interpret the record.

The National Institute of Standards and Technology provides a useful governance structure through its AI Risk Management Framework, which emphasizes mapping, measuring, managing, and governing AI risks. In healthcare, that means validating AI not only for technical performance, but also for workflow fit, clinical relevance, bias, safety, monitoring, and escalation when outputs are wrong.

Legacy Systems Are a Capital Strategy Problem

Technical debt is not only an IT problem. It is a capital allocation problem. Aging interfaces, redundant platforms, custom integrations, brittle data feeds, and manual workarounds all limit what health systems can safely automate.

Replacing legacy systems is expensive, disruptive, and politically difficult. Many organizations have years of workflow adaptation embedded into older platforms. Departments may depend on local configurations that are poorly documented but operationally essential. A new AI initiative can appear less expensive than foundational modernization, but that comparison is misleading if the AI tool underperforms because the infrastructure remains weak.

CFOs need to evaluate AI against the cost of the data foundation required to make it useful. A narrow tool may produce quick savings if the use case is bounded and the data is reliable. A broader enterprise AI strategy may require investment in cloud architecture, interoperability, data governance, master patient indexing, terminology management, consent management, cybersecurity, and analytics operations.

The return on AI depends on this hidden infrastructure. Without it, organizations may keep buying solutions that work around fragmentation rather than reducing it.

Patients Experience the Data Gaps

Patients rarely see the technical architecture, but they experience its failures. A patient may receive an AI-assisted message that lacks context about a recent hospitalization. A payer denial may arrive because documentation did not flow cleanly. A medication may be recommended before coverage or availability is known. A scheduling tool may route a patient incorrectly because referral data is incomplete.

These failures can undermine trust even when the algorithm itself is not the root cause. Patients do not distinguish between bad AI, bad data, poor workflow, or system fragmentation. They experience a healthcare organization that seems unable to connect what it already knows.

That is why patient impact should be part of AI governance. Health systems should monitor not only efficiency metrics, but also access delays, care coordination failures, complaint patterns, portal confusion, denied services, and equity effects. AI that reduces internal work while increasing patient burden is not a success.

Regulatory Pressure Will Favor Evidence

Regulators are moving toward stronger expectations for transparency, interoperability, and responsible AI use. The creation of the CMS Office of Health Technology and Products signals a more active federal posture on digital tools, data exchange, AI implementation, and product strategy across public programs. At the same time, federal and state policymakers are scrutinizing AI in prior authorization, claims review, medical billing, cybersecurity, and clinical decision support.

Healthcare organizations should expect more questions about how AI tools are selected, validated, monitored, and governed. Vendor assurances will not be enough. Leaders will need inventories of AI tools, documented use cases, data lineage, access controls, performance metrics, bias testing, human review procedures, and incident response pathways.

The industry does not need another AI pilot disconnected from enterprise strategy. It needs fewer pilots and stronger foundations.

The Real AI Strategy Is Operational Discipline

Healthcare AI progress is not stuck because the technology lacks promise. It is stuck because the operational environment is not ready to absorb that promise at scale.

A serious AI strategy begins with data governance, interoperability, identity management, cybersecurity, clinical workflow design, financial discipline, and accountability for outcomes. It treats automation as a tool for reducing complexity, not multiplying it. It asks whether the organization has the data and processes required to support a model before asking which model to buy.

The bot-versus-bot future is avoidable. Payers and providers can use AI to clarify rules, improve documentation, reduce avoidable denials, and support faster decisions. Clinicians can use AI to surface information that is timely, accurate, and relevant. Patients can benefit from systems that know enough to guide them safely.

That future requires the least glamorous part of digital transformation: fixing the data foundation. Without that work, healthcare AI may continue to accelerate the same fragmented processes that made the industry inefficient in the first place.