Loudoun Medical Group, P.C.

Corella Lumpkins

Revenue Cycle Integrity Advocate

Artificial Intelligence (AI) can help draft a clinical note, recommend codes and prepare a claim. However, an unsupported detail can move through that same sequence just as efficiently. Responsible innovation requires leaders to understand both what is being accelerated and whether the results remain accurate.

Ultimately, my perspective comes from 38 years across coding, billing, auditing, compliance and provider education. The one principle that I found to remain consistent: documentation must faithfully reflect the care delivered and coding must follow that documentation and applicable guidelines. AI creates opportunities to strengthen this foundation, along with new responsibilities for protecting it.

The best starting point is a defined operational problem. AI can identify denial patterns, organize work queues, flag missing information and draft appeals for review. These capabilities can free experienced professionals to address complex cases. Success should include time saved across the entire process, accounting for verification, corrections and rework. Faster initial processing offers limited value if another team inherits the cleanup.

Leaders must also distinguish documentation accuracy, correct coding and payer coverage. A paid claim does not prove coding accuracy and a denial does not automatically establish an error. Independent audits should evaluate the actual record and applicable requirements. CMS reinforces that providers remain responsible for correct coding even when automated NCCI edits do not identify an inappropriate combination.

A less obvious risk emerges when multiple AI tools reinforce the same mistake. Consider an ambient scribe that inserts an unsupported diagnosis, followed by a coding tool that recommends a corresponding code. Agreement between those systems is not independent validation. A qualified reviewer must establish that the documentation accurately represents the encounter before downstream decisions rely on it.

“We should use the time AI saves to strengthen professional judgment, improve provider understanding and build a more reliable revenue cycle for the people we serve.”

Safeguards should begin before procurement. Define the intended use, accountable owner, review requirements and escalation authority. Validate performance using representative specialties, patients and workflows. Protect patient information through appropriate access controls, data-use restrictions, risk assessment and business associate arrangements where required. Contracts should address vendor updates, audit rights, incident support and rollback. Preserve enough evidence to reconstruct recommendations and final decisions.

Provider education makes these controls practical. Physicians need concise guidance on checking generated notes for inaccurate diagnoses, medications, findings, time statements and descriptions of services. They must also recognize omissions and confirm that clinical reasoning is preserved. Longer or more polished documentation cannot make unsupported information reliable. A signature sign-off should follow every substantive review.

Short, specialty-specific examples can show how one inaccurate detail affects coding, risk adjustment or an appeal. Coders and billing staff need training to challenge recommendations and report anomalies. Repeated corrections should inform education and system improvement. Low override rates may indicate strong performance or uncritical acceptance; leaders should examine review quality and give staff time and authority to exercise judgment.

Integration also requires attention to what models learn. A denial prediction tool trained primarily on past payer decisions could reproduce disputed outcomes or discourage legitimate appeals. Predicting payer behavior should remain separate from evaluating coding correctness. As tools gain authority to submit appeals or release claims, organizations must define permitted actions, approval requirements and who can stop the process.

The business case should include integration, monitoring, downtime, vendor dependence and ongoing review costs. Maintain fallback procedures and staff competencies. Examine whether automated collection priorities and communications create unfair burdens, particularly for patients facing language barriers or seeking financial assistance. Financial performance should be evaluated alongside the patient experience.

I always advise leaders to begin with a controlled pilot, establish baseline accuracy and workload and set conditions for expansion or suspension. Integrate AI into existing compliance policies, training, reporting, auditing and corrective action. Monitor both overcoding and undercoding, unsupported documentation, correction rates and patient concerns. Please note that recommendations that reduce revenue deserve the same scrutiny as those that increase it.

Overall, the lasting value of AI will depend on whether organizations can explain their decisions, protect sensitive information and correct problems promptly. We should use the time AI saves to strengthen professional judgment, improve provider understanding and build a more reliable revenue cycle for the people we serve.