Hyland Connect

Biraj_Bhushan_C

Enterprise knowledge is fragmented and AI cannot fix what it cannot fully see

Enterprise knowledge is trapped in fragmented ecosystems: structured data in systems of record, semi-structured files in shared drives and unstructured content in content services platforms. These sources differ not just in format, but also in metadata standards, access controls and update cycles. As a result, enterprise information remains siloed, inconsistently tagged and poorly integrated. That fragmentation becomes more visible as organizations try to unlock enterprise data and use AI to improve decisions.

Decision systems cannot deliver meaningful gains if they are forced to work from incomplete, isolated or stale data. Cross-functional questions go unanswered because relationships across systems are not captured. Compliance risk grows because content lacks traceability and role-based access enforcement. And personalization and reuse opportunities fail to materialize because the enterprise lacks a unified, reusable context layer.

That’s because any decision system needs two things:

A unified view of the information required to make the right call

A reasoning layer that can replay why that call was made

Healthcare reveals the cost of missing context faster than most industries

Healthcare exposes this gap quickly because both the evidence and the governing logic are complex. The evidence needed to support a decision may be spread across clinical notes, scanned referrals, outside records, lab PDFs, imaging reports and prior treatment history. Those fragments may live in EHRs, payer utilization management platforms, fax queues, portals and document repositories often across organizations that do not share a common data model.

Healthcare also imposes a second requirement that many industries can avoid: decisions must be justified against changing plan benefit rules, medical policies, clinical guidelines and regulatory requirements and that rationale must remain defensible later.

The evidence is clear: prior authorization is a context problem

In the fall of 2021, a patient in central Ohio developed classic heart failure symptoms. His doctors requested a left heart catheterization to check for blocked arteries. The request was denied as “not medically necessary,” and a nuclear stress test was approved instead. Thirty-six hours after the stress test, the patient went into cardiac arrest at home.

No single case proves causality, but it does expose a structural failure mode: the system produced an outcome without a legible reasoning trail showing which criterion failed, which evidence mattered, which policy version applied and what would have changed the result.

The aggregate numbers point to the same failure mode at population scale. When prior authorization requests are denied, only a small percentage are appealed yet the vast majority of those appeals are overturned. For example, in 2024, Medicare Advantage plans processed 52.8 million prior authorization decisions, denied 7.7% of them and overturned 80.7% of appealed denials even though only 11.5% of denials were appealed in the first place.

For COOs, this is not just a utilization-management story. It is an operational execution problem. It suggests the system is not consistently capturing, packaging and communicating the information needed to get the decision right the first time. Nor is it returning denial information in a way that makes correction, resubmission and appeal efficient a point AHIMA explicitly supports. When that happens, denials increase, appeals rise, workloads expand and operational costs grow regardless of how much information is available.

The operational burden makes the failure worse on both sides of the transaction

On the provider side, the workload of assembling evidence has become structurally unsustainable: practices complete an average of 39 prior authorizations per physician per week, physicians and staff spend an average of 13 hours per week on prior authorization and 40% of physicians report having staff who work exclusively on prior authorization tasks, according to the American Medical Association.

On the payer side, the evidence packet is still frequently arriving through non-standard paths. Only about 31% of prior authorizations are fully electronic (ASC X12N 278), while 32% are partially electronic and 37% are fully manual forcing payers to separate, classify, extract and normalize documentation under tight SLA clocks and regulatory scrutiny before policy criteria can even be evaluated consistently.

The impact is direct: 93% of physicians report that prior authorization delays care, 82% say it can lead to treatment abandonment and 29% report a serious adverse event associated with prior authorization, according to the American Medical Association.

Regulation is now pushing the ecosystem toward explicit reasoning artifacts under tighter operational clocks. Under the CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F), operational requirements began January 1, 2026 including 72-hour timeframes for expedited requests, seven calendar days for standard requests and a requirement that impacted payers provide a specific reason for denials regardless of submission channel. For providers, this shifts denial communication toward actionable rationale, which is critical when care decisions are time sensitive.

Taken together, these signals show the same underlying problem: fragmentation. Evidence often arrives through fragmented channels. Rationale often remains trapped in reviewer notes, delegated vendor systems or call logs. And the denial artifact returned to the provider may contain little more than “not medically necessary,” without a structured explanation of what failed and what would have changed the result. That creates avoidable rework for providers and escalating appeals, compliance and rationale-quality risk for payers.

Content Intelligence Reduces the Cost of Assembling Evidence

This is where Hyland Content Intelligence helps creating a unified view of the enterprise by treating evidence not as a pile of attachments, but as a governed set of proof artifacts that can be assembled, validated and mapped to the policy criteria being applied.

For providers, this improves first-pass submissions and reduces pend cycles. For payers, it standardizes review packets and reduces variance, rework and handle time without lowering clinical rigor. The practical mechanism is a criteria ledger: a canonical mapping in which each policy clause has a preferred evidence artifact and the extracted fields that clause tests. In this model, policy becomes a series of executable gates. Evidence becomes the structured proof each gate requires from benefit and eligibility confirmation, to administrative completeness and coding validity, to clinical indication, severity thresholds, prerequisite history, safety constraints and site-of-care justification.

Each gate is then paired with its best proof artifact: eligibility and benefits data for coverage, forms and orders for completeness, clinical notes and consults for indication, imaging and lab reports for severity, medication and treatment history for prerequisites, allergy and safety data for contraindications and site-of-care documentation for setting decisions. With that structure in place, the system can extract the precise fields the clause tests and produce an outcome that states without ambiguity what failed and what would change it.

Decision intelligence preserves the “why” behind every outcome

Decision intelligence captures how the enterprise decides by recording reasoning when judgment occurs instead of trying to reconstruct it later from emails, memory or vendor outputs. When a decision is made inside the workflow, decision intelligence writes a structured decision record that binds the outcome to the evidence reviewed, the policy or guideline version in force, the criteria evaluated and how they were interpreted, the human judgment applied through overrides or escalations, the reviewer or delegated vendor path that produced the decision and the downstream result.

That changes the operating model. The enterprise is no longer guessing why something happened. The decision trace becomes part of how work is done, and it can be reused when the same request returns through correction, resubmission or appeal.

At scale, those records become institutional memory. They reveal where policies are ambiguous, where evidence capture is systematically weak, where reviewer variance is high, where vendor decisions correlate with overturn rates and where decision patterns diverge from downstream outcomes. For providers, this turns resubmissions and appeals into targeted fixes instead of guesswork. For payers, it turns medical-necessity determinations into more consistent, audit-ready rationales.

When content intelligence and decision intelligence work together, explanation becomes a query rather than an investigation. Now you can answer in structured terms which policy version was applied, which evidence was used, which criterion failed, who made the call and what would have changed the result.

For COOs, the shift is fundamental. Operational excellence is no longer defined solely by process efficiency. Increasingly, it is defined by the quality, consistency and explainability of the decisions made inside those processes. Organizations have spent decades optimizing workflows. The next frontier is optimizing judgment itself. The ability to make better decisions faster, while maintaining consistency and compliance, is becoming a core operational capability.

The Hyland Context Engine™: Connect content, decisions and systems of record

At Hyland, we see these capabilities converging. Building decision intelligence at enterprise scale requires three things simultaneously: a consolidated view of enterprise data to support sound decisions, the ability to capture decision moments as they happen in workflows and the ability to link those decisions to the actual content and evidence that informed them all while anchoring the work to canonical business objects across multiple systems of record.

That matters in healthcare because the same request often spans EHR workflows, payer utilization management platforms, delegated vendors and external documentation sources. Most vendors have one of these capabilities, perhaps two. Point solutions in workflow automation do not have content depth. Content platforms do not orchestrate decisions. Integration middleware does not understand either.

Hyland operates at the intersection of all three, particularly in healthcare, because we are already present where evidence is accessed and decisions are executed:

When clinical documents are opened

When work is escalated

When approvals or denials are recorded

The Hyland Enterprise Context Engine is a technology that represents our architectural vision for unifying content intelligence, decision intelligence and unified querying across both so enterprises can make defensible decisions at scale thousands per day with confidence in their consistency.

The strategic question: how do you scale sound judgment with confidence?

Every enterprise deploying AI in complex workflows faces the same scaling question: how do we apply sound judgment consistently across thousands of decisions each day and know we are getting it right?

Organizations that build memory of both facts and reasoning can answer that question more confidently. They can see where judgment is consistent, where it drifts and where AI can safely augment human expertise. They can meet regulatory demands not as a separate compliance burden, but as a natural byproduct of better-instrumented operations. And they can deploy AI as a managed capability, not a leap of faith.

For healthcare organizations navigating AI in clinical and administrative workflows, this is no longer a future-state idea. It is becoming an operational necessity.

Are you trying to increase authorization throughput without sacrificing clinical appropriateness, consistency or compliance? Please leave a comment and join the conversation.