Revenue Cycle Management (RCM) has historically been a chore. Claims go out, some are paid; others are denied; staff members spend countless hours on the phone with insurance companies; and, at the end of the month, you review an aging report and find that there are 90 day old accounts remaining unpaid. This has been the experience for most medical practices for decades.

Something new is happening. Not hype, not sales pitches from vendors. A practical approach to using technology to be more productive is taking place among billing offices and RCM firms. Artificial Intelligence (AI) has transitioned from being a buzzword on conference slide presentations to being a tool used by Billing Directors and Practice Administrators to reduce denial rates, accelerate cash flow, and decrease the amount of manual labor that burns out quality billing employees.

Let’s discuss where AI is currently making a difference in the RCM today, and what should your practice or billing office consider.

The Reason Why the Revenue Cycle Needed Help

To understand why AI is needed, let’s first describe the complexity of the U.S. medical billing system. There are thousands of CPT codes, tens of thousands of ICD-10 diagnosis codes, hundreds of commercial payer agreements each with its own rules, annual changes to CMS regulations, and the prior authorization process which consumes a significant portion of the time of staff members in all specialties.

The data illustrates the magnitude of the challenge. The AMA states that prior authorization alone utilizes approximately 16 hours of staff time weekly per physician practice. This equates to almost one-half of a full-time employee to manage prior authorization requests. Denial rates for claims submitted initially range from 5% to 10%, and many of these denials remain unaddressed as a result of physicians’ inability to provide sufficient resources to pursue recovery of such denials due to insufficient staff resources..

On top of that, healthcare billing errors cost the system billions every year. Undercoding, upcoding, missing modifiers, mismatched diagnosis codes, eligibility errors caught only after the claim denies. These are not exotic problems. They happen in every practice, every day, and they happen because humans processing high volumes of claims are going to make mistakes.

That is the environment AI stepped into. Not to replace anyone, but to handle the volume and speed of routine work that humans cannot do consistently at scale.

Where AI Is Actually Being Used in RCM Today

The term AI gets used loosely. So let’s understand about the actual applications that are live in RCM operations right now and what each one does.

Eligibility Verification and Pre-Authorization

The first place a claim can go wrong is before the patient even walks in the door. If the patient’s insurance has lapsed, if they switched plans, if the procedure requires prior auth and nobody caught it, you are setting up a denial before the appointment happens.

AI tools in this space pull eligibility data in real time from payer systems and flag issues automatically. Some platforms now run eligibility checks the day before an appointment and again on the day of the visit, then alert staff to any coverage gaps or auth requirements that need to be handled. What used to take a front desk employee 10 minutes per patient now happens in seconds, for every patient on the schedule, without anyone manually touching it.

On the prior authorization side, AI tools are learning payer-specific auth requirements and automatically preparing authorization requests with the clinical documentation most likely to get approved. Some platforms can submit auth requests directly to payer portals without staff involvement. The time savings are significant. More importantly, the authorization is in place before the service happens, which eliminates an entire category of downstream denials.

The Council for Affordable Quality Healthcare (CAQH) found that automating prior authorization alone saves healthcare organizations an average of $3.80 per transaction compared to manual processing. For a practice doing 500 authorizations a month, that adds up to well over $20,000 a year in staff cost savings.

Coding Assistance and Code Accuracy

This is one of the areas where AI has moved the needle most noticeably for practices with good EHR integration. AI-assisted coding tools read the clinical documentation in real time and suggest CPT and ICD-10 codes based on what the physician documented. They flag potential upcoding or downcoding before the claim is submitted. Some tools specifically check whether the diagnosis codes selected actually support the procedures billed, which is a major source of medical necessity denials.

Remember, these tools don’t replace medical coders. A good coder with clinical knowledge still catches things an algorithm misses. But they reduce the cognitive load on coders processing high volumes of notes and they catch the routine errors that happen when someone is working fast. A missed modifier, a bundling conflict, a deleted code that was not updated in the chargemaster. These tools catch those before the claim goes out.

For physician practices where the doctors are doing their own coding or where non-certified staff are assigning codes, AI coding assistance is genuinely valuable. It acts as a guardrail against the most common errors without requiring every physician to become a certified coder.

Claim Scrubbing and Pre-Submission Edits

Every RCM platform has had claim scrubbing for years. But older scrubbers checked against static rule sets. They would catch a missing date of birth or a procedure code that does not match the patient’s gender. What they could not do was learn over time.

Modern AI-powered claim scrubbers are different. They analyze your practice’s specific claim history and learn which code combinations, which payer-specific rules, and which documentation patterns lead to denials at each individual payer. Over time, the system gets better at predicting which claims are likely to be denied before they go out and flagging them for human review.

This predictive capability is what separates AI claim scrubbing from traditional rules-based scrubbing. A traditional scrubber knows that CPT code X is always bundled with CPT code Y per NCCI. An AI scrubber knows that Payer Z specifically denies CPT code X when billed with diagnosis code A, even though that combination clears NCCI, because that payer has an internal policy that the system has learned from your denial history. That is a meaningful difference.

A 2023 report from Waystar found that practices using AI-powered claim scrubbing reduced their initial denial rates by an average of 23% compared to practices using traditional rules-based scrubbers. For a practice collecting $5 million a year, a 23% reduction in denial rates represents hundreds of thousands of dollars in previously lost or delayed revenue.

Denial Management and Appeals

Denial management is where most practices hemorrhage money without realizing it. The average denial write-off rate across the industry is around 3 percent of net patient revenue. That sounds small until you do the math on a $4 million practice. That is $120,000 a year in written-off claims that could have been recovered.

The problem is not that practices do not want to work their denials. It is that working denials manually is time-consuming, and when staff are stretched thin, the oldest or smallest denials tend to sit until they hit the timely filing limit and become uncollectable.

AI-driven denial management tools change how this works. They automatically categorize every denial by reason code, payer, provider, and service type. They identify patterns in real time. If 40 of your denials in the last two weeks came from one payer citing the same reason code, the system flags it as a systemic issue rather than letting 40 individual billing reps work 40 separate appeals without anyone connecting the dots.

Beyond pattern recognition, some platforms are now drafting appeal letters automatically based on the denial reason and the patient’s clinical record. A human reviewer reads and approves the letter before it goes out, but the drafting work happens automatically. That alone can cut the time to file an appeal from 45 minutes to under 10.

Payment Posting and Reconciliation

Manual payment posting is one of the most error-prone parts of the entire revenue cycle. An EOB comes in with 200 line items. A billing specialist manually keys each one into the practice management system. At that speed and that volume, mistakes happen. Payments get posted to the wrong patient. Contractual adjustments get applied incorrectly. Underpayments slide through without anyone catching them.

AI-powered payment posting reads EOBs automatically, whether they arrive electronically via ERA or as paper documents scanned and converted, and matches each payment to the original claim. The system applies the correct contractual adjustment based on the payer contract on file and flags any line where the payment does not match the contracted rate. Human staff review only the exceptions, not the entire batch.

The accuracy improvement here is significant. But the bigger benefit for most practices is time. What used to take a full day for a billing specialist now processes in the background while that person works on higher-value tasks like denial appeals and patient account resolution.

Patient Financial Experience and Self-Pay Collections

The patient payment side of the revenue cycle has historically been the weakest link. After insurance adjudicates, the patient balance goes out on a statement and then you wait and hope. Collection rates on patient balances after the fact are often below 50 percent, and that number drops sharply as time passes.

AI tools in this space do something different. They analyze each patient’s payment history, their insurance plan, the amount owed, and other behavioral signals to predict the likelihood that a patient will pay and the best way to reach them. Some patients pay immediately if you send a text with a payment link. Others need a phone call. Others need a payment plan offered proactively before they disengage entirely.

Systems using this kind of patient-level analysis are seeing measurable improvements in self-pay collection rates. Patients are contacted through the channel most likely to get a response, at a time they are most likely to engage, with an offer that fits their situation. It is not magic. It is using data your practice already has about payment patterns and applying it consistently at scale.

MGMA data shows that the average medical practice collects only 50 to 65 cents on every dollar of patient self-pay balance. Practices that have adopted AI-driven patient payment tools report collection rates 15 to 25 percentage points higher than their previous baseline, according to vendor-published case studies from platforms like Waystar, Availity, and R1 RCM.

What AI Cannot Do in Revenue Cycle Management

You must know the limits of AI here because vendor marketing often glosses over them and practices end up frustrated when the tool does not perform the way it was sold.

AI tools in RCM are only as good as the data they work with If you have poor clinical documentation, old codes in your chargemaster, incorrect payer contract information loaded into your practice management system, no AI tool will be able to resolve these issues going forward. “Garbage in, garbage out” is just as applicable in AI-assisted billing as it is in every other application.

AI cannot replace a coder’s or biller’s clinical judgment. It supports them. The most effective use of AI tools for billing occurs when experienced coders and billers work with the tools to review exceptions, train the systems on the nuances of edge cases and identify the unique situation that are missed by pattern matching algorithms. Organizations which approach AI billing as simply setting it and forgetting it will likely find themselves facing a new set of problems than the ones they started with.

Ultimately, regulatory compliance remains a human responsibility. AI tools can flag potential upcoding patterns or unusual billing behavior, but the compliance officer, the coder, and the physician are still the ones who own the accuracy of what goes out under their NPI and their practice’s tax ID.

The regulatory risk of billing errors is no less, regardless of whether an automated software solution caused the error, rather than a human.

A 2023 OIG report notes that use of AI-assisted coding tools, when used without sufficient human oversight, will continue to create and increase the number of coding errors as much or more than manual coding.

Practical Steps for Practices Evaluating AI in RCM

If you’re considering the use of AI in your RCM, consider developing an implementation governance strategy will be necessary, not simply purchasing a software product.

Start With Your Denial Data

Before you look at any tool, pull your denial data by reason code, by payer, and by CPT code for the last 12 months. Where are your denials actually coming from?

If 60 percent of your denials are eligibility-related, AI-driven eligibility verification should be your first investment. If your denial rate is low but your self-pay collection rate is terrible, the patient payment tools are more relevant. Match the tool to your actual problem, not to what the vendor leads with on their demo.

Evaluate Integration Before Features

The most common implementation failure I see with RCM technology is buying a tool that does not integrate cleanly with the practice management system or EHR already in place. Data has to flow between systems automatically for AI tools to work. If your staff still has to manually export files and import them into a separate platform, the efficiency gains disappear fast. Before anything else, ask whether the tool has a certified integration with your current system and get references from practices using the same combination.

Measure Baseline Performance First

Know your numbers before you go live. Document your current first-pass clean claim rate, your denial rate by category, your days in accounts receivable, and your self-pay collection percentage. These are your baseline metrics. In order to determine if an AI RCM (Revenue Cycle Management) tool is working, you must have defined metrics to compare to six months into the relationship.

Without defined metrics, you cannot evaluate whether the investment is providing a return.
You can’t say that vague descriptions by vendors of “transformative” improvements mean anything without knowing what the baseline was for comparison.

Train Your Staff on the New Workflow

Most of the ROI lost due to poorly implemented AI RCM tools results from poor staff adoption. The tool is live; however, staff continue to perform around the tool because they either do not believe in it or do not understand how to operate it.

Plan for formalized training during your implementation of the new workflow. Ensure that your billing team understands not just how to operate the tool, but also understand the logic behind it and how to manage exceptions identified by the tool.

Staff will use the tool more effectively, and identify more problems if they understand the logic behind the tool rather than treating it like a “black box.”

A fair benchmark for a well-implemented AI RCM tool would be a 15 to 25 percent reduction in denial rates within six months of full staff utilization, a 20 to 30 percent reduction in days in AR within the first year, and documented staff time reductions on manual posting and eligibility functions within 90 days.

If a vendor claims to deliver greater ROI than this without documentation, then they are likely claiming unsubstantiated ROI and you should ask more questions

Final Thoughts

The revenue cycle has always been complex. AI makes it no simpler. What AI does is takes the high volume/low value, rule-based portions of the process and performs them faster and more consistently than any human team can at scale.

That allows your billing staff to focus on the parts of the process that require judgment (complex denials, payer negotiation, compliance reviews, etc.) and/or require communication with patients.

Done correctly, AI in the revenue cycle is not about replacing humans – it’s about allowing them to do their jobs more effectively.