Coding analysis — the E/M distribution behind the margin

The coding analysis behind the margin

REV's economics are not a generic "AI will fix billing" claim. They rest on a specific, observable fact about primary-care coding: the visit mix is a narrow, routine, repetitive distribution — a handful of established-patient office-visit codes carry almost all the volume. A narrow, routine code set is exactly the kind of problem AI codes accurately and scrubs cleanly. That is why REV's ~98.5% capture and ~78% gross margin are credible for primary care — and why we say plainly that other specialties and non-ambulatory settings will almost certainly be lower margin.

Primary-care coding is narrow, routine, and highly automatable. Across primary care, two codes — 99213 and 99214 — account for roughly 80%+ of established-patient office visits, with 99214 now the plurality. Very little volume sits in 99211 or the high-intensity 99215, and new-patient visits are a minority of the schedule. A distribution this concentrated is the best-case setting for accurate automated coding and clean claims — which is the engine behind REV's capture and margin. The honest corollary, stated below: that engine is strongest here, in ambulatory primary care.

~80%+99213 + 99214 of est. visits

98.5%REV capture target

~78%gross margin (PCP case)

$89,250collections / provider / mo

$595 + 4.9%REV all-in price

These are at-scale figures — the system earns its accuracy as it learns. Year 1 runs below ~$750K collections per provider and ramps toward the ~$1,071,000/provider at-scale level as the model learns from every encounter and every corrected mistake — the ~98.5% capture and ~78% gross margin build over time, they are not switched on day one.

How the $1,071,000 / provider is built (benchmarked, not assumed)

At-scale collections decompose into three independently benchmarked inputs:

28 visits / day × 255 clinical days / yr × $150 collected / encounter = $1,071,000 / provider / yr.

~28 visits/day — above the AAFP / Physicians Foundation norm of ~20–25 patients/day (survey average ~20.2), and the lift is by design: REV reclaims admin time rather than shortening visits. Across ~8 clinical hours, that is still ~17 min/visit — at or above the average primary-care face-to-face visit (~15–18 min).
$150 collected/encounter — MGMA average revenue per primary-care outpatient encounter (collected, not billed; specialty runs ~$233).
255 clinical days — ~49 working weeks net of PTO/holidays.

This is not assumed on day one. Year 1 runs below ~$750K — about the PCP median, where schedule gaps and collection leakage hold practices under the arithmetic — and ramps to $1,071,000 at scale. The platform reclaims admin time (ambient AI finishes the note in-room, cutting the ~1.4 hrs/day of after-hours EHR work) and packs the schedule per-minute, lifting throughput above the raw 20–25/day norm without shortening visits; it also lifts capture (clean claims + fewer denials collect closer to the full $150). So the at-scale figure is benchmarked inputs lifted by removing administrative drag, not by rushing patients.

On the 90% RCM attach: the base is a captive / partner-acquired network placed onto REV's platform, so RCM attach is near-full by design — not a cold-sell ramp from zero.

Sources: AAFP — family-medicine encounter volumes · MGMA — avg revenue per outpatient encounter (~$150) · full month-by-month derivation in the model's Collections tab.

The primary-care E/M distribution

Office/outpatient evaluation-and-management (E/M) codes split into established-patient visits (99211–99215) and new-patient visits (99202–99205). For family medicine and internal medicine, the overwhelming majority of office volume is established-patient, and within that, level 3 (99213) and level 4 (99214) carry almost everything. The table shows two cited anchor points: the older CMS Medicare Part B cut (where 99213 was the plurality) and the current era after the 2021 E/M guideline change (where 99214 has risen to roughly half of established family-medicine visits). The illustrative current-era mix in the chart is a midpoint synthesized from these cited sources — labeled as such, not a single published table.

Code	Service (established patient, office)	CMS Part B, CY2003 (via AAFP)	Current era, post-2021 (family medicine)	Illustrative mix used in chart
99211	Minimal / nurse visit, no physician	~1–2%	low single digits	2%
99212	Straightforward, brief	~9–13%	declining	6%
99213	Low-to-moderate complexity — the classic routine visit	~61%	~35–45%	38%
99214	Moderate complexity — now the plurality	~23%	~50% (above 50% Medicare)	50%
99215	High complexity — uncommon in PCP	~3–5%	low single digits	4%
99213 + 99214 combined: ~84% (CY2003 anchor) and ~85–90% in the current-era illustrative mix. New-patient visits (99202–99205) are a separate, minority share of the primary-care office schedule; within them 99203/99204 are the plurality.

Established-patient visit mix — illustrative current-era distribution

Chart values are the illustrative current-era midpoint from the table (sums to 100%). The hard-cited anchors are: 99213 ≈ 61% and 99214 ≈ 23% in CMS Part B CY2003 (per AAFP), and 99214 approaching/above 50% of established family-medicine visits today (AAFP, 2021). See Sources.

The real engine: we stop denials before the patient is seen

Clean coding matters, but the bigger lever is the front door. The cheapest denial is the one you never incur. REV gates the encounter before it happens instead of appealing it afterward:

Real-time eligibility (X12 270/271). We check coverage and benefits the moment a visit is requested — not at check-in. No active coverage, no surprise at the back end.
100% pre-approval. If the visit needs prior authorization, we require it up front (X12 278). Encounters hit the calendar already cleared to pay.
Predict the code from scheduling. The reason-for-visit and intake tell us the likely E/M / CPT before the patient ever arrives — so we already know whether it will be paid.
Delay or decline the bad ones. Encounters we can see will be denied — or payers that won't cover that visit — get held (indefinitely) rather than performed for free. We don't do unpaid work.
Cashless copay pre-pay. Patient responsibility is collected up front, digitally — which kills the patient-side bad debt that quietly eats collections.
Card on file — with a collections backstop. The patient authorizes a card at booking (cardless, digital). The copay is taken up front, and — worst case — if insurance hasn't paid the claim within a set window (e.g. ~45 days), the patient-responsibility balance is charged to the card on file automatically. A slow or denied claim converts to cash instead of aging in A/R: the visit still collects, days-in-A/R drop, and patient-side bad debt is largely engineered out rather than chased.

Benchmark: PCP first-pass approval at an athena-class incumbent runs about 92.8% (~7% denied) before any front-door gating. Shift-left gating is how REV designs the captured share up toward ~98.5% — the denials are engineered out, not won back on appeal.

These are goals, not guarantees. 98.5% capture / ~78% margin are the conservative, fully-gated, primary-care targets. Push into riskier work — complex specialties, urgent/unscheduled visits, or looser gating to chase volume — and the number falls. But the economics are built so that even if we're off by a meaningful margin, we still crush it: see the downside sensitivity — even at well below plan the returns stay strong.

The economics of chasing a denial — and knowing when to stop

A little behavioral economics, applied carefully. A denied primary-care claim collects about $150; REV's RCM fee on it is ~4.9% — roughly $7.50. A labor-heavy shop spends $25–50+ to rework a denial by hand — frequently more than the claim is worth — so it either eats the loss or chases at a loss. Automation changes that math, and so does knowing when to quit.

Pursue

We can chase what they can't afford to

Our marginal cost to work a denial is near-zero — the rework is AI-drafted. So the math that makes a $150 claim “not worth chasing” for a $25–50 manual team doesn't bind us. We keep working it, cheaply, and for longer.

Stop

But we stop where it stops paying

Past a rational threshold — a few automated attempts, aging beyond a set window, or a denial reason with low recovery odds — the probability-weighted recovery falls below our return bar. Chasing further destroys value, so the system stops. Most billing shops either over-chase or give up too early; we do neither.

Recover or exit

Recapture next visit — or dispose cleanly

Because REV is one longitudinal system, at that threshold we have better exits than a billing shop: recapture on the patient's next encounter (the balance follows them; with a card on file it's collected at the next visit), or, if recovery is unlikely, discharge or sell the balance and move on — no open-ended, value-destroying chase.

Carefully stated — illustrative. $150/encounter (MGMA); ~$7.50 = REV's ~4.9% fee on it; $25–50+/claim manual rework is an industry estimate (MGMA/HFMA). The stop thresholds and the recapture-vs-dispose logic are design choices (pre-GA); patient balances are handled with the patient's card-on-file authorization and standard AR disposition. The point isn't to chase harder — it's that automation lets us economically work the long tail at near-zero cost, and a rational stopping rule means we stop spending exactly where it stops paying.

Why this distribution supports REV's model

Three claims, each tied directly to the distribution above. None of them require REV to "win" on aggressiveness — they follow from the fact that the primary-care code set is narrow and routine.

Capture

A narrow code set is easy to code right — closing the leakage

When ~80%+ of visits resolve to two codes, the AI is choosing the documentation-supported level on a small, well-understood surface, then scrubbing for the routine edits that cause denials. That is the opposite of a sprawling surgical or modifier-heavy code space.

The market's own data shows independent practices lose a documented 5–10% of collections to under/miscoding and unworked denials. A concentrated, routine distribution is precisely where accurate coding + automated scrubbing can drive that leakage down toward REV's modeled ~1.5% leakage (98.5% capture).

Low fraud risk

Concentrated in 99213/99214, not aggressive 99215

The mix sits in level 3 and level 4 — not the high-intensity 99215. REV codes to documentation-supported specificity, not "highest defensible." The goal is to capture work already performed and documented, explicitly not to upcode.

This matters because 99214 is one of HHS's most error-prone / improper-payment E/M codes — so the safeguard is documentation, not aggression. Contrast the cautionary case: eClinicalWorks paid $150M+ to the DOJ (2017) over misrepresented software behavior. A narrow coding surface plus documentation-tethered automation is a low False-Claims-Act-risk posture, by design.

Margin

Routine + automatable → low cost-to-serve

Because the work is repetitive and rules-friendly, the marginal human touch per claim falls as volume grows. RCM cost-to-serve trends toward roughly ~1.3% of collections at scale, while REV prices at 4.9% — that spread is the gross margin.

On $89,250/provider/mo of collections ($1,071,000/yr), REV's all-in $595 + 4.9% supports the modeled ~78% gross margin. Critically, that ~78% is a primary-care figure — it is the direct consequence of this distribution, not a blanket claim across all of healthcare.

Apply REV's automation to each function — the math

This only works because primary care isn't a series of one-off, custom cases. The economics here do not hold if the job is to code rare, complex, bespoke procedures from day one — that is not the bet. They hold because ~94% of primary-care visits resolve to just three routine codes — 99212, 99213, 99214; because those claims are denied for a small, known set of reasons (eligibility / coverage, registration / missing information, and documentation / coding level); because REV prevents those exact causes upstream (eligibility at booking, no re-keying, coding generated from the encounter); and because we automate most of what remains (AI-drafted denials and appeals, continuous AR). On a surface that narrow and that repetitive, an AI-native system beats a rules-based one — which is how REV can credibly out-perform even a best-in-class incumbent. It is also why we say plainly: this is the primary-care case, not a claim about every specialty.

PCP-first means one workflow and one code set, so each RCM function automates harder. Index athena's labor-heavy RCM effort at 100; residual = athena pts × (1 − automation). The bar shows REV's residual inside athena's labor.

▮ bright green spine = REV's residual labor · colored funnel = athena's labor by function (fades to dark at the edges) · exact numbers in the table below.

RCM function	athena (index pts)	REV automates	REV residual	athena ▮ vs REV ▮	how
Eligibility & registration	15	92%	1.2		270/271 at booking; digital intake → claim
Coding & charge capture	14	88%	1.7		ambient AI coding at point of care
Claim scrub & submission	9	88%	1.1		auto-scrub, clean first pass
Denial rework	20	82%	3.6		~72% prevented; rest AI-drafted
Appeals	9	78%	2.0		AI-drafted appeal packets
Payment posting (ERA)	10	90%	1.0		ERA auto-posting + auto-reconcile
AR follow-up / collections	17	78%	3.7		continuous automated AR worklist
Patient statements / support	6	85%	0.9		cardless digital statements
Total RCM manual labor	100	~85%	~15	→ REV ≈ 15% of athena's labor

Add it up: the residuals sum to ~15 — REV runs RCM at roughly 15% of athena's manual effort. That ~85% labor reduction (at scale) is what drives RCM cost-to-serve toward the ~1.3% of collections in the Margin claim above.

Function by function — exactly how each one automates

Eligibility & registration — athena 15 → REV 1.2 (~92% target). A labor-heavy shop verifies coverage at check-in and re-keys demographics off a clipboard, then chases the ~quarter of all denials that trace to bad eligibility or registration. REV fires a real-time 270/271 the moment a visit is booked, and digital intake writes the patient's details straight into the chart and the claim — captured once, never re-keyed. Coverage problems surface days early, not on the remit. Residual: mid-cycle coverage changes and secondary coordination-of-benefits a person still confirms.
Coding & charge capture — athena 14 → REV 1.7 (~88% target). Legacy coding is a separate, after-the-fact pass — a coder or the physician revisits the note hours later and, to stay safe, often under-codes. REV's ambient scribe documents the visit as it happens and the same model assigns the documentation-supported level in the room, on a narrow 99213/99214 surface. Coding is a byproduct of the encounter, not a second job. Residual: visits with ambiguous documentation a human reviews.
Claim scrub & submission — athena 9 → REV 1.1 (~88% target). Incumbents run claims through scrub edits and fix rejects before they go out. Because eligibility and coding were already correct upstream, REV's claim is clean by construction and submits automatically at a ~98% first-pass target. Residual: the occasional payer-specific edit a person clears.
Denial rework — athena 20 → REV 3.6 (~82% target). This is the biggest bucket at an incumbent — a billing team works every denial by hand. REV prevents ~72% of denials before submission (front-end and coding-level fixed upstream), and for the ones that still land, the rework packet is AI-drafted — a person reviews and sends rather than investigating from scratch. Residual: the genuinely contestable denials.
Appeals — athena 9 → REV 2.0 (~78% target). Appeals at incumbents are manual letters assembled from the chart. REV auto-assembles the appeal with the supporting documentation attached; a person approves and submits. Residual: medical-necessity appeals that need human judgment (the hardest, lowest-automation bucket — on purpose).
Payment posting (ERA) — athena 10 → REV 1.0 (~90% target). Manual remit posting is a standing cost line. REV auto-posts the 835/ERA and reconciles against the expected amount; only exceptions — underpayments, unexpected adjustments — reach a person. Residual: those exceptions.
AR follow-up / collections — athena 17 → REV 3.7 (~78% target). Incumbents work aging AR in a monthly human sweep. REV works the AR worklist continuously in the background, prioritizing what will actually collect and auto-following-up; and the card-on-file backstop converts unpaid balances to cash once the insurance window lapses, so far less ever reaches a person. Residual: the high-dollar or genuinely stuck accounts.
Patient statements / support — athena 6 → REV 0.9 (~85% target). Paper statements and a billing phone line are labor. REV collects patient responsibility up front (cardless, at booking) and sends digital statements with self-serve pay — fewer balances and fewer calls. Residual: the patient-support contacts a person handles.

Basis — design targets, not measured. REV is pre-GA, so the per-function automation rates are design targets, informed by our sales calls + public RCM benchmarks (front-end ≈ half of denials; cost-to-collect ~2–4% for a good operation, in-house up to ~13.7%). The "athena = 100" index and function allocation are illustrative, not athena's audited figures (their ~4–8% is service price, not internal cost). Reached on a ramp, not on day one.

The honest caveat — this is the primary-care case

Other specialties and non-ambulatory settings will almost certainly be lower margin.

The ~78% gross margin and ~98.5% capture are the primary-care beachhead case. They are credible because ambulatory primary-care coding is the narrow, routine distribution shown above. We are not claiming these economics hold everywhere — and investors should not read them that way.

Procedural, surgical, modifier-heavy, prior-authorization-heavy, and inpatient / non-ambulatory work involves a far wider and more variable code space: bundling and global-period rules, NCCI edits, modifier logic, medical-necessity and prior-auth friction, and higher per-claim dollar stakes. More complexity and variance means more coding labor, more denials, and more variance — so both capture and margin will almost certainly be lower in those settings than in the PCP case.

That is exactly why REV's strategy is to win primary care first, where the coding economics are strongest and the model is most defensible. Expansion into other specialties is an opportunity, but it should be underwritten with its own, lower, specialty-specific assumptions — not the primary-care numbers.

Ambulatory primary care (REV beachhead)

Narrow, routine code set (99213/99214 dominate). High automation, low variance → capture and margin as modeled.

Procedural / surgical specialties

Wide CPT surface, global periods, bundling/NCCI edits, modifiers. More manual review → lower margin.

Prior-auth / infusion / high-cost

Heavy prior-auth and medical-necessity friction; large per-claim dollars. More denials/variance → lower capture.

Inpatient / non-ambulatory

Different code families, DRG/facility logic, complex documentation. Out of the modeled beachhead → lower margin.

Sources

Every figure on this page is either directly cited below or labeled as an illustrative synthesis of these sources. Where a precise single-table figure does not exist publicly (e.g., the exact new-patient share of PCP office volume), we say so rather than invent one.

AAFP / FPM — "Coding 'Routine' Office Visits: 99213 or 99214?" (2005). Cites Medicare Part B Physician/Supplier National Data, CY2003: family physicians coded 99213 for ~61% and 99214 for ~23% of established-patient Medicare visits. aafp.org/pubs/fpm/issues/2005/0900/p52.html
AAFP / FPM — "Coding Level 4 Office Visits Using the New E/M Guidelines" (Jan 2021). States the national average for family physicians' use of 99214 is approaching 50% of established-patient office visits, and is above 50% for Medicare patients. aafp.org/pubs/fpm/issues/2021/0100/p27.html
MedCentral — "The Most-Billed and Most Error-Prone CPT Codes of 2023." Confirms 99213 and 99214 are among the most frequently billed office E/M codes. medcentral.com/coding-reimbursement/the-most-billed-and-most-error-prone-codes-of-2023
Tebra / The Intake — "E/M code 99214 tops HHS' list of Medicare improper payments." Basis for the documentation-not-aggression safeguard framing around 99214. tebra.com/theintake/getting-paid/e-m-code-99214-medicare-improper-payments
STFM Family Medicine — "Coding Family Medicine Residency Clinic Visits, 99213 or 99214?" (2019). Peer-reviewed corroboration that 99213/99214 carry the bulk of family-medicine established visits and that undercoding is common. journals.stfm.org/familymedicine/2019/june/young-2018-0390/
U.S. Department of Justice — "Electronic Health Records Vendor to Pay $155 Million to Settle False Claims Act Allegations" (May 31, 2017). The eClinicalWorks settlement referenced as the cautionary contrast. justice.gov/archives/opa/pr/electronic-health-records-vendor-pay-155-million-settle-false-claims-act-allegations
AMA — 2021 office/outpatient E/M (99202–99215) code and guideline changes. Background on the 2021 time/MDM revisions that shifted volume toward 99214. ama-assn.org/practice-management/cpt/cpt-code-99214-established-patient-office-visit-30-39-minutes

Method note: The percentage ranges and the illustrative current-era mix are synthesized from the cited sources to represent a typical primary-care (family/internal medicine) established-patient distribution; they are not a single official published table, and real distributions vary by panel, payer mix, and practice. The capture, margin, collections, and pricing figures are REV's own modeled proforma assumptions for the primary-care case and are documented in the EMR/RCM deep dive. They should not be applied to other specialties or non-ambulatory settings without separate, lower assumptions.

Related: the EMR/RCM deep dive models capture and net revenue across RCM models; the EMR comparison covers all-in pricing.