PS² Practice Management published an AI consulting guide for plastic surgery practices; we wrote the revenue-cycle guest section. This is the companion piece.
RPA belongs on deterministic work — eligibility, claim status, posting. Generative AI belongs on judgment work — reading op notes and proposing codes. Most practices buy them backwards.
Neither works on top of vague documentation. Fix the workflow and the op note first; then AI translates at scale, with a coder signing off on every claim.
PS² Practice Management’s AI consulting guide for plastic surgery practices
Most AI advice for practices stops before it reaches the part where the money leaks.
PS² Practice Management — the consulting firm of the American Society of Plastic Surgeons, run by Amanda Taylor — published a guide to AI consulting for plastic surgery practices this week. It opens with the most useful sentence in the entire AI conversation: AI is not a strategy.
“Successful AI adoption depends less on software and more on operational consistency. The strongest results come when AI supports well defined workflows, clear documentation, and accountable teams.” — Amanda Taylor, PS² Practice Management
I wrote the guest section of that piece — the revenue cycle blind spot. PS² covers the operational core: SOPs, reporting, staff training, patient-coordinator workflows. Coding and collections is the layer next door, and it runs on the same principle they teach: get the workflow and the documentation right first.
This post is the expanded version of that argument. Read theirs for the operational half. This is the money half.
The Wrong Robot
RPA vs. generative AI in the revenue cycle: rules work vs. judgment work
Buy them backwards and you either automate errors or pay premium prices for button-pushing.
Two different technologies get sold under one AI label, and they are not interchangeable. RPA — rule-based process automation — executes if-then logic. Generative AI reads unstructured text and makes judgment calls. One is a button-pusher. The other is a translator.
The revenue cycle has plenty of both kinds of work. Eligibility checks, claim status lookups, payment posting, routing: deterministic, repeatable, RPA territory. Reading an operative report and proposing the CPT, modifier, and diagnosis combination: judgment on unstructured text, which is exactly what a large language model is built for and exactly what RPA cannot do.
Buy them backwards and you get one of two failure modes. RPA pointed at judgment work automates errors — it will push the wrong button ten thousand times before lunch without a flicker of doubt. Generative AI pointed at deterministic work is an expensive way to do what a rule does for free.
Matching the tool to the task is most of the battle, and the vendors won’t do the matching for you. We ran that test against the specialty EMRs in our 2026 vendor-by-vendor AI scorecard — most of what ships under the AI label is workflow scripting with an AI sticker on it.
Where Claims Die
Plastic surgery coding: the cosmetic-vs-reconstructive line is the leak
High-value claims get payer scrutiny; vague op notes get denials.
Every procedure a surgeon performs has to be translated from the operative report into the right CPT, modifier, and diagnosis combination — right order, right units — before a payer reimburses the claim. In plastics that translation is unusually hard. Cosmetic versus reconstructive is a coverage line, not a clinical one. Staged procedures stack global periods. Payer-specific preferences change the right answer by contract.
The code set underneath keeps moving. The current ICD-10-CM set carries 70,000-plus diagnosis codes — your coder isn’t memorizing them, they’re navigating them under payer scrutiny. CMS refreshes its NCCI bundling edits quarterly — a code pair that paid clean in Q1 can deny the same claim in Q3.
These are high-value claims, so payers look closely. A small coding miss becomes a denial, and generally speaking the leak isn’t fraud — it’s under-coding and avoidable denials. You code once; you follow up ten times. We see it in plastics AR every week: the practice that under-codes a free-flap reconstruction doesn’t get a warning letter. They get paid less, quietly, forever.
The people who do this work well are scarce, expensive, and hard to keep current — the rules change mid-year and staying current is a discipline of its own. That labor problem is the one AI actually addresses. How we structure the coding function sits on the plastic surgery billing hub.
Garbage In, Faster
AI medical coding only works on clean operative documentation
A model reading a vague op note doesn’t get confused. It gets confident.
The working design is simple. The model reads the operative note and proposes the codes. A qualified coder reviews the proposal and signs off. AI proposes, a human disposes — the coder spends less time retyping and more time supervising. Administrative burden drops without removing the person who owns the outcome.
The catch is upstream. A model reading a vague op note doesn’t get confused — it gets confident. It produces incorrect codes faster than your old process produced correct ones. The readiness test is simple: if two coders reading the same note would land on different answers, the note isn’t ready for a model either.
That’s why the operational work PS² does comes before any of this, and it’s why documentation standards live inside our billing service rather than being treated as the surgeon’s problem. Fix the note and the workflow. Then let the model translate at scale.
RPA (rule-based automation)
Generative AI (LLMs)
Built for
Deterministic if-then steps
Judgment calls on unstructured text
Revenue-cycle fit
Eligibility, claim status, posting, routing
Op note → CPT/modifier/ICD-10 proposal, appeal drafts
Fails when
The rule changes and nobody updates it
The documentation underneath is vague
Human role
Exception handling
A qualified coder reviews and signs off
Order of Operations
Fix the workflow first, then let AI translate the op note at scale
Skip the first step and you’ve bought faster denials.
Generally speaking, the practices that get paid for AI adoption run the same order of operations. Number one: operative documentation structured enough that two coders reach the same answer. Number two: coding and denials run as a measured process — clean claim rate, days in AR, denial rate by payer — not a back-office afterthought. Number three: the right tool on each task, RPA on the deterministic steps, generative AI on the judgment work. Number four: every claim has an accountable owner, and every AI proposal gets a coder’s signature before it ships.
PS² builds the operational foundation — the workflows, the documentation, the accountability. When that’s solid and the question becomes where coding and collections are leaking, that’s the conversation we run. Read their full guide first. Then count your denials.
FAQ
Frequently asked questions
Quick answers to the questions we hear most.
Q
What is the difference between RPA and generative AI in medical billing?
RPA (robotic process automation) executes rule-based, deterministic tasks such as eligibility checks, claim status lookups, and payment posting. Generative AI (large language models) handles judgment-based work on unstructured text, such as reading an operative note and proposing CPT, modifier, and diagnosis codes. RPA follows rules; generative AI makes proposals that a qualified human reviews.
Q
Can AI do medical coding for plastic surgery?
AI can read an operative report and propose CPT, modifier, and ICD-10 codes, including the cosmetic-versus-reconstructive distinctions that make plastic surgery coding difficult. It works reliably only when the operative documentation is clear and a qualified coder reviews and approves every proposal before the claim goes out.
Q
Should a plastic surgery practice fix workflows before adopting AI?
Yes. AI built on inconsistent workflows or vague documentation produces errors faster, not fewer. Practices see the strongest results when standard operating procedures, documentation, and team accountability are consistent first, and AI is then added to speed up specific, well-defined tasks under human review.
Q
Does AI help with out-of-network and IDR claims?
Out-of-network and IDR (independent dispute resolution) claims carry the most manual work and the most revenue risk in plastic surgery. Mapping that workflow first, then using AI for the repetitive parts — assembling documentation, drafting appeals — under human review is where practices recover the most time and revenue.
Talk to a specialist
Not sure where your revenue cycle stands?
If your clean claim rate or days in AR aren’t where they should be, that’s a conversation worth having. We’ll look at your numbers and tell you straight.
This article explains how these codes tend to get paid — it is not coding, billing, or legal advice. Some of the codes here are bundled into the main procedure, or fall into grey areas that vary by payer. Before billing anything in a grey area, confirm it with your own coding and compliance team and the current guidance from your specialty society, Medicare (CMS), and AMA CPT. When in doubt, don’t bill it.
