Most medical practice leaders have heard the pitch by now: AI that reads clinical documentation, suggests codes, flags missed charges, and catches errors before claims go out the door.

Ambient clinical documentation — AI-generated visit notes — has gotten most of the attention, and MGMA has polled on it before. But a different category of AI is gaining ground in the middle of the revenue cycle, and it deserves its own conversation.

AI-assisted coding and AI-driven charge review are distinct from ambient scribing. They sit downstream of documentation and upstream of claims submission — the stretch of the revenue cycle where coding accuracy, charge capture completeness, and compliance risk converge. For midsized practices especially, this is often the part of the cycle where small inefficiencies compound into real money.

Our April 7, 2026, MGMA Stat poll found that only 33% of practice leaders say they are using AI for coding (9%), charge review (4%), or both (20%), while 68% are not. The poll had 256 applicable responses.

Practice leaders report that AI’s biggest impacts on coding and charge review workflows were significant gains in speed and accuracy, driving higher productivity, better capture, fewer edits, and reductions in denials while improving compliance. Secondary impacts include improved ability to compare payer policies to documentation and identify missed charges or risks, though several groups note it is too early to tell or that use remains limited or subject to governance approvals.

Among the majority not currently using AI for coding or charge review, many expect to evaluate it within the next year, largely driven by potential cost savings, revenue recovery, efficiency gains, and ROI, with several already actively reviewing platforms or planning future implementations. Others remain hesitant due to financial constraints, competing priorities, limited vendor capabilities, or EHR integration barriers, particularly where current options lack specialty nuance or demonstrated value. Interest is widespread but conditional, with adoption hinging on clear economic justification, accuracy, reliability, and seamless integration.

If you are part of that majority, let’s take a closer look at how to effectively evaluate whether these tools are worth the investment, as well as which ones actually fit.

What AI-assisted coding and charge review do

It helps to separate the two functions, even though some vendors bundle them.

• AI-assisted coding typically works alongside or after a coder reviews documentation. The AI reads the clinical note — often an EHR-generated or ambient-scribed note — and suggests CPT, ICD-10, and modifier selections. In some configurations, the AI suggests codes before a human coder touches the chart; in others, it acts as a second pass. The value proposition centers on coder productivity, coding accuracy, and reduction in lag between encounter and claim.
• AI-assisted charge review operates at a different point. It looks at the charges attached to an encounter and compares them against the documentation, payer rules, and historical patterns. The goal is to catch missed charges, flag documentation gaps, identify audit risk, and reduce charge leakage — revenue that was earned but never billed. Some tools also flag potential upcoding or unbundling issues before a claim goes out.
• Clinical documentation integrity (CDI) overlaps with both. AI-driven CDI tools can prompt clinicians or coders to clarify documentation that does not support the specificity a code requires. For practices moving toward value-based arrangements, CDI also matters for risk adjustment accuracy.

The common thread is that all three sit in the mid-cycle — after the patient encounter, before the claim drops. That is precisely the zone where many practices have the thinnest staffing and the least automation.

Why smaller practices face a different calculus

Large health systems typically have dedicated coding departments, CDI teams, and compliance staff who can absorb and govern a new AI tool. Small practices may not have the volume to justify the cost. Midsized groups often occupy an uncomfortable middle: enough volume for coding and charge errors to matter financially, but not enough infrastructure to evaluate and manage AI tools the way a system can.

That creates a specific set of evaluation challenges.

1. Staffing leverage matters more. A midsized practice with three coders is not looking for marginal productivity gains. It is looking for the difference between keeping three coders or needing to hire a fourth — or the difference between keeping up with volume and falling behind on time-to-bill. The evaluation question is not just, "does the AI code accurately?" It is, "does it reduce enough coder time per chart to change our staffing math?"
2. Integration complexity is real. Most AI coding and charge review tools need to connect with the practice's EHR, practice management system, and sometimes a clearinghouse. For system-owned groups, IT support and integration resources are usually available centrally. For independent and private practices, integration is often the bottleneck, not the AI itself. Before evaluating accuracy claims, a practice leader should ask: How does this connect to our existing systems, and who manages it once it is live?
3. Compliance governance looks different. A hospital-owned group typically has a compliance department that will audit AI-suggested codes and build oversight protocols. A private midsized practice may need to build that oversight from scratch. That means the evaluation is about the tool and about whether the practice has a plan for validating AI output on an ongoing basis. Blindly accepting AI-suggested codes may seem like a workflow improvement before it devolves into a compliance liability.

What to evaluate and what to ask vendors

Practice leaders considering AI for coding or charge review should resist the temptation to lead with accuracy percentages. Vendors will quote impressive numbers, but accuracy depends heavily on specialty mix, documentation quality, payer rules, and how "accuracy" is defined. A tool that is 95% accurate on E/M codes in primary care may perform very differently on surgical coding in a multispecialty group.

Instead, start with operational fit.

• Workflow integration: Where does the AI sit in your current process? Does it replace a step, add a step, or run in parallel? If coders have to toggle between the AI's suggestions and the EHR, the productivity gain may be smaller than the demo suggested. Ask for a workflow diagram specific to your EHR, not a generic one.
• Specialty and service mix coverage: Most AI coding tools perform best on high-volume, well-documented encounter types — primary care E/M visits, for example. Performance may drop on procedural coding, multispecialty groups with heterogeneous documentation styles, or ancillary services. Ask which specialties and CPT ranges the tool has been trained and validated on.
• Denial and audit impact: For charge review tools, ask specifically about denial rate reduction, not just charge capture improvement. A tool that finds more charges but increases denial rates is not helping. Similarly, ask whether the tool flags potential compliance risks or just optimizes revenue.
• Measurement and reporting: Can the tool report its own performance in terms you can act on? Coder productivity per chart, time-to-bill changes, denial rate shifts, charge capture variance — these are the metrics a practice leader needs. If the vendor cannot show you a dashboard or report that maps to your existing KPIs, that is a warning sign.
• Cost structure: Pricing models vary: per-encounter, per-provider, flat monthly, or percentage of recovered revenue. For a midsized practice, the per-encounter model often makes the most sense for predictability, but a percentage-of-recovery model may align incentives better for charge review tools. Either way, calculate the break-even point against your current cost per coded encounter.

Private practice versus system-owned: different entry points

For a private or independent practice, the key concerns tend to be cost justification, integration burden, and whether the tool reduces daily operational friction. Private practices are more likely to pilot a tool on a single specialty or location before committing, and they should. A 60- to 90-day pilot with clear before-and-after metrics is a reasonable ask of any vendor.

For a system-owned or hospital-owned group, the decision often involves IT, compliance, revenue cycle leadership, and sometimes a systemwide procurement process. The AI tool may need to meet enterprise security and interoperability standards that add months to evaluation. But system-owned groups also have an advantage: they can compare performance across multiple sites and specialties, which gives them better data to judge whether the tool delivers.

In both cases, the evaluation should not end at go-live. AI coding and charge review tools learn and update. The practice needs a plan for ongoing validation — periodic audits of AI-suggested codes against coder judgment, tracking of denial patterns, and regular review of whether the tool's performance holds as documentation patterns, payer rules, or provider mix change.

The question behind the question

This week’s poll shows where practices stand today on AI-assisted coding and charge review, but the more useful conversation is about what leaders should be asking before they adopt or expand a pilot into a full deployment.

The right AI tool in the mid-cycle can meaningfully improve coder productivity, reduce charge leakage, shorten time-to-bill, and lower compliance risk. The wrong tool — or the right tool without governance — can introduce new errors, create audit exposure, and frustrate the staff it was supposed to help.

Don’t start by asking, "which AI is best?" Always build a foundation around, "what problem are we solving, how will we know if it is working, and who owns the oversight?" Answer those questions first, and the vendor evaluation gets a lot more productive.

Join the conversation

• MGMA Stat polls are conducted weekly to give medical practice leaders a pulse on the latest trends in healthcare management. To participate, sign up for MGMA Stat at mgma.com/mgma-stat.
• Has your organization added AI for coding and/or charge review? Share your story in the MGMA Member Community or email us at connection@mgma.com.