Cost per claim — shift-left
Why denials stay low
Three facts make primary-care RCM unusually clean for an AI-native platform: the visit mix is narrow and repeatable (99213 + 99214 are ~84% of visits — ~90% of the work is a handful of codes), those routine established-patient E/M codes are among the lowest-denial, most-automatable claims in medicine, and the denials they do get are overwhelmingly front-end — eligibility and registration, not the code itself. Handled our way, an already-low denial rate goes lower.
The easiest claims to get paid — and we make them easier
Primary care isn't a long tail of exotic, denial-prone procedures. It's a handful of routine, established-patient office visits — which makes it the cleanest, most automatable RCM in medicine, and exactly where REV's approach compounds.
~90% of the work, a handful of codes
99213 + 99214 are ~84% of visits; the established-patient E/M set is nearly all of it. A narrow, repeatable distribution is the single most automatable thing in revenue-cycle.
Lowest-denial by nature
Routine established E/M aren't the high-denial, prior-auth-heavy procedures or new-patient complexity. When they do deny, it's almost always front-end (eligibility/registration) or a documentation-level mismatch — not the code.
We cut it further — our way
Eligibility at booking removes the front-end denials; the structured encounter + ambient coding makes the documentation always support the level billed — so the documentation/level denials go too. An already-low rate goes lower, automatically.
The point: the most common denial causes for these codes are the two REV automates away at the source. So this isn't "manage denials better" — it's structurally fewer denials on the claims that are already the easiest to win.
Where denials come from
Front-end denial share per the sources below (MD Clarity / Rivet / Change Healthcare); E/M mix illustrative of the PCP established-patient distribution (see coding-deep-dive).
From claim to labor — the denial cascade
Best-guess, modeled cascade (illustrative — public benchmarks + REV's model, not athena's audited figures): where primary-care claims sit in a multi-specialty book, how few of them deny, why, and how much of that denial labor REV's AI + shift-left filtering removes.
Front-end (eligibility + registration) ≈ half of denials and is fully preventable upstream; coding/level denials are eliminated by the structured encounter. Net: ~72% of denials are AI-prevented before submission; the ~28% residual is what any human ever touches.
~72% less denial labor on the cleanest claims in medicine — and the residual is AI-assisted, not manual. That is the mechanism behind RCM cost-to-serve of ~1.3% of collections (vs a labor-heavy multi-specialty book) and the ~78% gross margin.
How to read this: figures are an illustrative model — denial reason mix and the ~30% PCP share are best-guess estimates from public benchmarks (front-end ≈ half of denials; eligibility ≈ a quarter — sources below); claim volume, the 15.8% weighted denial rate and the ~72% AI-prevention are REV's model assumptions. They are not athena's audited numbers.
Shift-left: fix the claim before it exists
The cheapest denial is the one that never happens. REV pushes every check that normally happens after the visit to before it:
Verify eligibility
Real-time insurance eligibility + benefits check when the appointment is booked — bad/expired coverage is caught days early, not at the claim.
Capture copay
Cardless copay/estimate collected pre-visit — no front-desk fumble, no missed patient responsibility.
Code it right, once
Ambient AI scribe codes the 99213/99214 visit correctly at the point of care — documentation-supported, not under-coded-and-reworked.
Clean first-pass claim
Front end + coding are already right, so the claim goes out clean — high first-pass acceptance, low denials, minimal manual A/R.
Why it compounds: because ~a quarter of denials are registration/eligibility and roughly half are front-end overall, eliminating them at scheduling removes the single biggest, most-avoidable denial bucket — before any AI even touches the coding.
What it does to the model
Low net denials
Weighted denial rate ~15.8% × ~72% AI-prevented → a small residual worked automatically. Clean-claims first pass is the norm.
RCM cost ~1.3%
Because denials/rework are prevented, RCM cost-to-serve runs ~1.3% of collections (vs labor-heavy legacy shops) → ~78% gross margin.
Falls over time
The AI learns the narrow PCP mix, so cost-to-serve per physician declines year over year while headcount still grows with the business.
See the coding analysis for the full 99213/99214 E/M distribution, and the encounter walk-through for the eligibility-at-scheduling and pre-visit-copay steps in the product.
Sources
- Registration/eligibility ≈ a quarter of denials; front-end issues ≈ half of denials: MD Clarity — Front-End Denial Rate and Rivet Health — front-end issues cause ~half of denials (citing Change Healthcare).
- PCP E/M visit mix (99213/99214 dominant) — see coding analysis (CMS/AAFP-sourced).
- Why these codes are low-denial & automatable: established-patient office-visit E/M (99213 / 99214, AMA) are routine, high-volume visits — not the prior-auth-heavy procedures that drive high denial rates. Their denials concentrate in front-end (eligibility/registration, per source 1) and documentation/level support — both addressable upstream, which is what REV's eligibility-at-scheduling + structured-encounter coding do. "Lowest-denial / most-automatable" is a qualitative characterization of the routine PCP visit set, not a single audited statistic.
- REV denial/cost figures — the model's denial derivation (weighted ~15.8% × ~72% AI-prevented) and cost-to-serve (~1.3% of collections / ~78% gross margin).