The $1.7 Trillion Hidden Asset: What Healthcare Organizations Are Missing in Their Clinical Notes

Every major health system in the United States has, over the past decade, invested substantially in healthcare analytics infrastructure. Data warehouses have been built. Business intelligence platforms have been deployed. Population health dashboards display chronic disease prevalence, utilization patterns, readmission rates, and quality metrics in real time. The analytics capability of the average large health system in 2025 is incomparably more sophisticated than it was in 2015.

And yet the most important data in that health system — the clinical intelligence that captures physician reasoning, patient-reported experience, treatment response nuance, care coordination breakdown, and hundreds of other dimensions of clinical reality — is almost entirely absent from every one of those dashboards.

It is absent because it lives in clinical notes. In discharge summaries. In consultation letters. In radiology report narratives. In pathology case descriptions. In nursing shift notes. In patient-reported symptom descriptions. In the free-text fields of electronic health record systems that are, by clinical design, the place where healthcare providers document the things that don't fit in a structured field.

The estimate that 80% of all clinical data exists in unstructured form understates how profound this analytical gap is when you understand what that 80% contains. It is not 80% of ancillary data. It is 80% of the data that clinicians themselves use to make decisions. It is the physician's documented reasoning. The nurse's observation that the patient seems confused since starting medication. The discharge summary notation that the patient mentioned they were sleeping in their car. The oncology note that the patient's response was “much better than expected.”

The analytical value of this data — and the cost of its continued inaccessibility — is genuinely difficult to quantify. But the McKinsey Global Institute has attempted to quantify the value of AI-enabled healthcare analytics improvements, and the figures are large enough to justify this article's title. The $1.7 trillion figure represents the estimated addressable value of AI-enabled healthcare analytics improvements, including the subset attributable to unlocking unstructured clinical data.

The research literature on what structured clinical data misses — what is documented in clinical notes but never surfaces in administrative or analytical data systems — is extensive enough to constitute its own field of study. The findings are consistent, and they are striking.

Perhaps most consequentially for pharmaceutical applications: research from project-emerse.org documented that unstructured data was essential for resolving 59% of CLL trial eligibility criteria and 77% of prostate cancer trial eligibility criteria.

The patients that structured-data-only eligibility algorithms miss are not a random sample of the eligible population. They are systematically the patients whose eligibility is documented in clinical narrative rather than structured code — often patients with more complex clinical histories, more nuanced disease presentations, and greater need for the experimental therapy.

The analytical inaccessibility of clinical notes is not a new problem. Clinicians and researchers have understood for decades that free text clinical documentation contains information of enormous value. The reason most organizations have not acted on this understanding is not because they were unaware of it. It is because until recently, the technology required to extract structured intelligence from clinical free text at scale was not mature enough for production deployment.

That threshold has been crossed. The combination of healthcare-specific natural language processing models, large language models fine-tuned on clinical text, and clinical NLP infrastructure that can process millions of clinical notes with acceptable latency and accuracy has transformed the practical accessibility of unstructured clinical data between 2021 and 2025.

The key developments:

The analytical value of unstructured clinical data manifests in six application domains with the highest combination of clinical impact and commercial return.

Given the analytical value of unstructured clinical data and the maturity of the technologies required to access it, the question worth asking is why most healthcare organizations and pharmaceutical companies have not yet built substantial unstructured data analytics capability.

The answer is not primarily technological. It is organizational and infrastructural.

Clinical notes are heterogeneous in ways that create genuine engineering challenges. The free-text documentation style of a cardiologist at an academic medical center differs from the documentation style of a community hospitalist, which differs from the documentation style of a family medicine physician in a rural practice. Abbreviations, institutional conventions, dictation artifacts, and documentation culture vary substantially across clinicians and institutions.

The governance infrastructure required to process clinical notes for analytics purposes is also more demanding than the governance requirements for structured data analytics. Clinical notes contain extraordinarily sensitive patient information — mental health observations, sexual health history, substance use documentation, social vulnerability details — that requires heightened privacy protection.

And the organizational appetite for clinical NLP has, historically, run ahead of the organizational capacity to act on what NLP analytics reveals. An NLP platform that surfaces previously invisible clinical intelligence creates work, not just insight. The organizational capability to act on that work at scale requires process changes, workflow integration, and cross-functional coordination.

These are real constraints. They are surmountable constraints. The organizations that have invested in surmounting them are discovering analytics advantages that are, for now, genuinely differentiating.

For healthcare organizations and pharmaceutical companies evaluating how to begin building unstructured data analytics capability, the following principles reflect the approach of leading practitioners.

The healthcare industry's relationship with unstructured data analytics is at the inflection point that structured data analytics reached approximately fifteen years ago. In 2010, most health systems had EHR systems but had not built the data warehousing and analytics infrastructure to use them. By 2020, structured data analytics had become a baseline expectation.

The same transition is underway for unstructured data. The technology is mature. The governance frameworks are established. The use cases are proven. The competitive advantage of early movers is substantial.

That is not a technology prediction. It is an evidence-based observation about where the most important clinical intelligence in the industry already lives — and has always lived — waiting for the analytics infrastructure that makes it accessible.

The infrastructure exists. The question is no longer technological. It is organizational.

The notes are not noise. They are the signal. It is time to listen.