AI in Healthcare: To AI or Not

Contributors: Wren Keber and Lisa Soroka
To learn more about Wren and Lisa, click here.

 

  Source: Bigstock

Introduction: Setting the Stage

Artificial Intelligence (AI) has moved rapidly from experimentation to operational reality across the healthcare ecosystem. Providers, payers, and vendors are deploying AI-enabled tools at unprecedented and unforeseen speed and scale; drawn by the promise of improved efficiency, reduced administrative burden, and more consistent decision-making. At the same time, this pace of adoption has outstripped the industry’s ability to fully identify and assess downstream impacts on care quality, cost, fairness, and the human healthcare experience, and mitigate risks associated with exceptionally fast-paced disruption.

This tension between promise and necessary prudence frames the purpose of this four-part series. Rather than focusing solely on the technology itself, we examine how AI is reshaping real-world healthcare operations and what that means for providers, payers, patients, and the broader ecosystem. The central question is not whether AI will influence healthcare, but “How it can be harnessed in ways that enhance quality and outcomes without compromising both patient and provider trust, accountability, and well-being?”

Our perspective here is intended to be pragmatic. AI has shown it can be a powerful positive force when thoughtfully designed, deployed, and rigorously governed. However, the cautions surrounding its use, particularly in high-stakes administrative and financial workflows, are very real and deserving of careful scrutiny. We strongly believe balancing innovation with broad responsibility among stakeholders should not be optional.

The Rise of AI in Healthcare Administration

One of the most visible areas of AI adoption has been on the administration/business side of healthcare. Organizations are increasingly using AI-driven tools to support medical coding, automate prior authorization workflows, adjudicate claims, and streamline revenue cycle processes. These applications are often positioned as solutions to longstanding challenges, including administrative complexity, workforce shortages, rising costs, and clinician burnout driven by non-clinical work.

We understand the motivation; administrative tasks continue to consume an inappropriately significant share of overall healthcare resources. Even marginal efficiency gains can translate into substantial financial impact at scale. AI proponents promise faster turnaround times, more efficient, accurate, and/or standardized processes, and the ability to sift through vast volumes of data more quickly than all-human teams alone, presumably leading to better informed decisions.

Yet, while these goals are laudable, the reality on the ground is more nuanced. Administrative decisions, especially those tied to billing, reimbursement, and even coverage, carry significant clinical, financial, and ethical implications. As AI systems become more embedded in administrative workflows, and sophistication evolves, we believe the question shifts from “Can this work?” to “At what cost, and under which safeguards?”

Coding and Claims: Opportunities and Pitfalls

1. Billing and Coding

AI-powered coding tools are increasingly used to assist or, in some cases, perform medical coding, by analyzing clinical documentation and suggesting diagnosis and procedure codes. In theory, these tools can help improve accuracy, reduce lag time, and ensure that services provided are fully captured.

In practice, however, challenges quickly emerge. AI systems may suggest additional diagnoses or procedures based on contextual clues in documentation, sometimes without the clinical nuance or judgment applied by certified human coders. Without appropriate review and oversight, this can lead to inappropriate code selection, perceived upcoding, and/or claims that fail to align with payer policies.

Unlike human coders, AI systems are not credentialed, certified, and/or professionally accountable. Their “training” derives from utilizing historical data patterns, which may themselves reflect variability, bias, and/or outdated practices. Fully AI-autonomous coding without essential human review prior to claim submission, potentially introduces unmitigated financial and regulatory risk many organizations may underestimate.

These dynamics raise important questions including but not limited to:

  1. What safeguards are in place to prevent erroneous and/or non-compliant coding generated by AI?
  2. How do systems distinguish between legitimate documentation improvement and inappropriate revenue capture?
  3. How do widely publicized examples of AI used to detect fraud and abuse influence payer scrutiny of otherwise legitimate claims?

There is also an often overlooked patient-facing dimension.  As AI enables more granular charge capture, patients may experience higher financial responsibility for services that previously went unbilled or uncollected. While one could argue this represents more complete accounting of the care delivered, the real-world impact can feel very different, which could reinforce patients’ perceptions that “every penny is being tracked” in an already perceived complex, uncaring, and stressful healthcare financial system.

2. Human vs. Machine: Expertise and Accountability

The expanding role of AI in coding and claims processing highlights a broader tension between human expertise and machine-driven automation. Experienced coders develop judgment over years of navigating clinical ambiguity, evolving regulations, and payer-specific nuances. That expertise is difficult to fully replicate through algorithms trained on static or imperfect datasets.

In some cases, removing humans from technically complex workflows may reduce immediate labor burden but introduce new vulnerabilities. Errors may be harder to detect, accountability may become diffused, and organizations may struggle to explain or defend AI-driven decisions during audits, appeals, and/or legal challenges.

Legal and/or ethical liability is another increasingly emerging consideration. When AI-generated outputs contribute to payment disputes, compliance findings, and/or patient harm, where does responsibility ultimately lie? Is it with the technology vendor, the deploying organization, or the individual(s) who failed to intervene? These questions, among others, remain largely unresolved.

Additional considerations around consent, transparency, and disclosure, particularly in settings involving caregivers, parents, or legal guardians exist. As AI increasingly participates in interpreting clinical records and influencing financial outcomes, health care organizations must consider whether and how patients and/or families are informed about the role these tools play in the patient journey.

Conclusion: Charting a Thoughtful Path Forward

AI has undeniable potential to improve, streamline, and/or optimize healthcare operations, reduce administrative burden, and help enable more consistent decision-making. At the same time, its use in areas such as coding and claims introduces real financial, regulatory, and human risks that cannot be ignored and must be timely addressed.

The path forward requires much more than technical capability. It demands intentional thoughtful governance, clear accountability, and a commitment to keeping human expertise integrated into high-stakes processes. Stakeholders should be asking not only what AI can do, but how it should be deployed, monitored, and constrained to align with the broader healthcare mission of optimal/appropriate high-quality patient care and outcomes/values.

We will discuss pre-authorization/utilization management, clinical workflows/decision-making and website design in the next three articles.

As AI continues to expand into adjacent areas - such as prior authorizations and increasingly automated payer-provider interactions - the need for this balance will only intensify. Thoughtful adoption today will help shape trust, equity, and sustainability across the broad healthcare ecosystem tomorrow.


Contact Wren at: [email protected]
Contact Lisa at: [email protected]



-