AI for Revenue Cycle Management: A Cost-Per-Transaction View

Healthcare revenue cycle management (RCM) is the most opaque cost center in healthcare operations. A single transaction—a prior authorization request, a denied claim, a charge entry with wrong CPT code—carries hidden cost because the transaction lives in three different teams (billing, coding, denial management) and multiple IT systems (EHR, billing system, payer portal). AI can optimize single steps of this workflow, but RCM leaders often don't know what they're paying per step, so they can't measure what AI saves.

The work-item economics of RCM

RCM spans multiple work items: prior authorization requests, claim denials, charge entry, coding verification, denial appeals. Each has different cost structure and different AI applicability.

Prior authorization: A staff member requests pre-authorization for a procedure, tracking eligibility and benefits through the payer's system. Routine request: 8–12 minutes on phone or portal. Cost per request: $10–$25. High-complexity requests (off-label use, bundled procedures, experimental) take 30–60 minutes. AI vendors (Notable, Pearl Health, Optum's AI layer) use knowledge of payer-specific rules to pre-fill authorization requests, look up eligibility rules, and identify missing documentation. Cost per AI-assisted request: $2–$5 for routine cases, $15–$20 for complex (human still required). Payback: 50–80% on routine requests.

Denial management: A claim is denied (invalid provider ID, medical necessity not met, deductible not satisfied). Staff review the denial, gather evidence, submit appeal. Time per denial: 15–40 minutes depending on reason. Cost: $25–$80 per denial appeal. AI pre-screens denials, identifies which ones are appeals (reversible with documentation), which are write-offs (not worth appealing). Cost per denial pre-screen: $0.50–$2. Payback on triage: 60–80%. Payback on appeal success: 0–30% (depends entirely on whether you can find the missing documentation).

Charge capture and coding: A clinical user documents a procedure (broken arm treated and set), and a coder assigns the procedure code (CPT 99213) and diagnosis code (ICD-10 S52.001A). Manual coding: 3–8 minutes per encounter depending on complexity. Cost: $0.50–$2 per encounter. AI with confidence scoring (highlight high-risk codes for auditor review, auto-assign routine codes) reduces manual review load by 30–60%. Cost with AI: $0.25–$1.20 per encounter. Payback: 40–50% on high-volume, straightforward encounters.

Days in AR (accounts receivable): If the average claim sits 45 days before payment, your cash flow is delayed by 45 days of revenue. AI denial management and prior auth acceleration can reduce that to 38 days. Reducing AR by 7 days on $50M annual revenue = $972k in one-time cash release. That's a non-operating benefit; it doesn't reduce headcount, but it improves working capital.

The hidden iceberg: RCM optimization sounds like "pay 30% less per transaction" but actually means "pay slightly less per transaction AND improve first-pass acceptance rate." A claim denied because of a coding error costs $50 to correct (rework). An AI-assisted charge capture system that catches 80% of coding errors before claim submission saves $40 per prevented denial on a portfolio of 100,000 annual claims. That's $4M in rework cost avoidance, way more valuable than the $500k in direct cost reduction.

Where AI really helps RCM

Prior authorization for routine procedures. Orthopedic surgery, imaging studies, behavioral health referrals—procedures with clear medical necessity criteria and predictable payer requirements. AI automates 70–85% of routine auth requests. Payback: 12–18 months.

Denial triage and appeals prioritization. Classifying 500 monthly denials into "appeal immediately," "gather more docs then appeal," "write off," "provider contract issue." AI does this in seconds per denial. Payback: 6–12 months on supervisor cost reduction + faster appeal processing.

Preventive error detection. Finding duplicate claims, invalid provider IDs, or obvious coding mismatches before submission. This is pattern matching. Payback: 6–9 months.

Charge capture for high-volume, low-complexity care. Primary care visits, routine imaging, uncomplicated ED visits. AI handles 60–70% of coding; auditor reviews flagged cases. Payback: 12–18 months on coder cost reduction.

Where AI doesn't help RCM

Complex clinical documentation. Surgeon documents a three-stage tumor resection with reconstruction and complication (infection). The coder must interpret clinical nuance, apply medical necessity rules, cross-reference payer policy. AI hallucination risk is high (it might code the complication as a secondary diagnosis when it should be primary, changing reimbursement). Human still required. Cost: $10–$30 per encounter. AI adds minimal value.

Payer negotiations and contract disputes. A payer denies a claim and cites contract language that your legal team interprets differently. Resolving this requires human judgment, legal review, and negotiation. AI triage is useful; but AI settlement is too risky.

Outlier revenue recognition. Insurance adjustments, hospital bad debt reserves, charity care classification. These require judgment and audit trail. AI adds no value.

The vendor landscape for RCM AI

Notable (Series B funded) operates as an RCM operations copilot, focusing on prior auth and denial triage. Pearl Health (portfolio company) owns charge capture and coding optimization. Optum (UnitedHealth subsidiary) bundles AI across the entire RCM value chain—prior auth, denial management, coding—as part of their broader RCM services. Smaller vendors like Dolante or Fluxus focus on specific RCM steps.

The competitive axis is not just cost per transaction; it's integration depth. A prior-auth-only vendor requires manual handoff to your claims system. A full-stack vendor that plugs into your EHR and billing system can optimize the entire flow end-to-end. Integration cost and implementation time vary by vendor and your systems architecture.

Most RCM leaders buy best-of-breed (one vendor for prior auth, one for denial management) rather than full-stack, but integration burden increases.

The cost attribution challenge in RCM

RCM cost is typically buried across multiple departments. Billing staff, coding staff, authorizations coordinator, denial manager, revenue cycle director. No one person owns the total RCM P&L. Finance sees "billing and collections staff" as a cost line, but can't break it down to prior auth cost vs. denial cost vs. coding cost.

This creates blind spots. A vendor that cuts prior auth cost by 30% might increase denial rate (by being too aggressive on what it authorizes), which increases denial management cost 20%. Net RCM cost might increase. But since those costs live in different ledgers, no one sees the tradeoff.

The proper model: build a cost-per-claim view. From claim entry (charge capture + coding, $0.50–$2 per claim) to prior auth (if needed, $2–$15 per authorization) to claim submission to denial management (20–30% of claims are initially denied; triage + appeal cost $10–$50 per denied claim). Sum those and divide by net collected revenue. That's your RCM margin. Then measure how each AI vendor improves that metric.

RCM cost benchmark table

| RCM function | Work unit | Manual cost | AI-assisted cost | Payback horizon | | --- | --- | --- | --- | --- | | Prior authorization | 1 auth request | $10–$25 | $2–$5 (routine) | 12–18 mo | | Denial triage | 1 denied claim | $25–$80 (appeal) | $0.50–$2 (triage) | 6–12 mo | | Charge capture | 1 charge/code | $0.50–$2 | $0.25–$1.20 | 12–18 mo | | Coding audit | 1 audited claim | $1–$3 (reviewer) | $0.25–$0.75 | 6–12 mo | | AR days reduction | 1 day of float | 7-day DSO | 38-day DSO | One-time cash | | Denial appeal success | 1 appealed claim | $50–$150 (effort) | $30–$100 (with AI help) | Margin only |

The CFO playbook for RCM AI

1. Establish baseline RCM cost and yield. Calculate: (1) total RCM team cost (all staff, all departments: $2M–$5M depending on revenue size), (2) total claims processed annually, (3) cost per claim (total team cost ÷ claims), (4) net revenue per claim (gross charges minus denials minus adjustments, divided by claims). This is your anchor. Example: $2.5M RCM cost ÷ 200,000 annual claims = $12.50 per claim; net revenue $95 per claim; RCM margin = $82.50 per claim.

2. Map RCM cost by function. How much of the $2.5M goes to prior auth, how much to coding, how much to denial management? Finance doesn't usually know this. Time-log one team per function for 4 weeks to get real data. You'll find that prior auth and denial management eat 40–50% of RCM cost combined.

3. Run vendor pilots sequentially, not in parallel. If you deploy prior auth AI and denial AI simultaneously, you won't know which one is driving results. Start with prior auth (usually highest ROI): 3-month pilot on all routine auth requests, measure time per auth and first-pass authorization rate. Then denial management, then coding.

4. Measure denial rate and appeal success rate separately. Some RCM leaders have a 25% denial rate and 40% appeal success (meaning 10% of original claims are appealed and paid). Others have 18% denial rate and 60% appeal success. AI that improves either or both is valuable. AI that reduces denial rate by catching errors earlier is better than AI that improves appeal success (because it avoids rework).

5. Model the cash impact, not just the cost impact. If AI reduces your AR days from 45 to 38, that releases $972k in working capital for a $50M revenue practice. That might be more valuable than the $300k annual cost reduction from faster authorization processing.

6. Require the vendor to guarantee accuracy thresholds. Before implementation, lock in: prior auth success rate (percentage of authorizations that are accepted by payer) at 92%+, denial appeal success rate at 55%+, coding accuracy at 97%+ (auditor review). If the vendor hits those, expand. If not, pause.

7. Set an 18-month payback requirement by function. Prior auth and denial management should pay back in 12–18 months. Coding optimization in 18–24 months. If the vendor can't commit to that timeline given your specific volume, the math doesn't work for mid-market.

For CFOs at health systems and payers, RCM AI is one of the highest-ROI IT investments in healthcare. Prior auth acceleration alone can reduce DSO by 5–7 days, unlocking $500k–$2M in one-time cash. To model the true RCM cost structure and where AI intervention pays back most, talk to Runrate to establish work-item-level attribution across your revenue cycle.