Why practices need a stronger data foundation before relying on AI tools

Physician practices are being inundated with claims from AI tool developers that promise easier access to relevant clinical information and automated documentation.

Physicians and practice staff feel this need acutely. Growing documentation requirements, increasingly complex workflows, and expanding expectations for data exchange make any technology that reduces friction especially appealing. These pressures explain why conversational AI tools are gaining attention. They hold the potential to simplify daily tasks by allowing clinicians to retrieve information with a brief spoken or typed question rather than navigating multiple screens.

However, what may be obscured by these claims is that the value of AI depends on the underlying data being complete, structured, and clinically accurate. By strengthening that foundation first, emerging technologies can operate safely and consistently, delivering meaningful benefits to practices and patients.

Promise of conversational ai falls short today

The primary reason conversational AI is currently falling short of its promise is that large language models do not correct the inconsistencies or gaps that appear in routine clinical documentation. They reflect whatever data exists in the record, whether it is structured or not. When notes are unstructured, fragmented, or scanned, AI tools may miss essential details or, at worst, provide invented answers or “hallucinations.”

Practices already encounter these issues every day due to human error and technological limitations. For example, a referral note may arrive with limited structure, a medication list may require multiple rounds of reconciliation, or key details of an encounter may be buried in narrative text. These challenges influence clinical decisions, care coordination, quality scores, and the usefulness of any AI tool that depends on this information.

These everyday documentation issues become even more significant as expectations around data exchange continue to rise. Federal pressure around interoperability is increasing, including stepped-up enforcement against information blocking, which means practices are receiving more external data from more sources.

As this volume grows, the accuracy and structure of that incoming information become critical. After all, a polished conversational interface can automate the management of this data, but it cannot compensate for gaps that originate in documentation or coding.

Unstructured clinical information burden

The primary challenge facing these tools is that a large share of clinically relevant information remains as unstructured text, making it difficult to extract, standardize, or share meaningfully across systems. This creates friction for physicians who must maintain accurate problem lists, reconcile medications, and manage chronic conditions under tight time constraints.

Even widely used standards, such as Fast Healthcare Interoperability Resources, are of little help because they tend to support technical processes rather than clinical ones. They help systems move data but do not improve the accuracy or completeness of what is exchanged. When that content is inconsistent or incomplete, the receiving system or clinician must still interpret or restructure it, adding time and diminishing clinical value.

Strengthening data at the point of care

One promising approach to removing this barrier is a universal medical coder that converts free-text documentation into structured clinical concepts during routine workflows. This type of tool maps information to vocabularies such as ICD-10, Systematized Nomenclature of Medicine Clinical Terms, and Logical Observation Identifiers Names and Codes while maintaining clinical meaning.

While such a universal coder supports billing and compliance, its broader value lies in its ability to create a consistent, clinically usable data foundation. Clearer histories, more reliable summaries, and fewer documentation inconsistencies directly support better care. Consistently and accurately structured data also strengthens decision support, population health analytics, care coordination, and the performance of emerging AI tools.

Several steps can help practices move toward that foundation:

1. Begin by strengthening documentation workflows, so problem lists and medication records remain accurate and up to date. This creates a more reliable starting point for any technology that depends on those details.
2. Build on that foundation with tools that identify clinical concepts during documentation, reducing the amount of cleanup required later and keeping information consistent across encounters.
3. Use this improved consistency to evaluate AI solutions based on how effectively they maintain or enhance data quality, since reliable inputs help determine long-term clinical value.
4. Extend this evaluation by asking vendors to explain how their systems ensure structured, accurate documentation, which helps reveal which solutions are designed for sustainable improvement rather than short-term convenience.

A clearer path to safe and effective ai adoption

Despite the limitations and risks, natural language AI systems are already reducing the documentation burden in practices and health systems across the U.S. Yet their success has been mixed due to the high variability in data consistency and quality across information systems.

Strengthening that environment can, in turn, improve the tools’ performance and, more importantly, help protect patient safety, advance care quality, and reduce physicians’ time spent on data management and documentation.

When the data foundation is strong, the dazzling new tools built on top can finally deliver consistent and clinically meaningful value.

David Lareau is president and CEO of Medicomp Systems.