Every handoff is a chance to get it wrong
The average health plan claim touches 5 to 7 people before it is paid, and each handoff adds delay, cost, and a fresh opportunity for error. In a self-funded plan you are paying for all three. Industry data shows:
- 3-14 days average turnaround for manual adjudication
- Meaningful per-claim processing cost on legacy stacks
- 40-60% auto-adjudication rate for legacy systems
- Significant rework cost when errors are caught after payment
For a TPA processing 10,000 claims per month, that adds up to substantial monthly processing costs: before counting rework, denials, and staff overhead.
What auto-adjudication actually does
Auto-adjudication isn't a black-box AI making medical decisions. The core is a deterministic rules engine applying plan benefits, fee schedules, and regulatory requirements to validated claims: the same steps a human examiner would follow, executed in milliseconds. AI-assisted edit checks (medical-necessity scoring, modifier validation, pattern recognition layered over CMS NCCI/MUE rules) catch the cases where rules-only systems would otherwise pend.
The adjudication pipeline
Step 1: Pre-Adjudication Rules
Before pricing a single line, the engine checks:
- Is this a duplicate claim? (CARC-18)
- Is the member eligible on the date of service? (CARC-27)
- Was the claim filed within the timely filing window? (CARC-29)
- Are the diagnosis codes covered under the plan? (CARC-96)
- Is prior authorization required? (Pend for review)
Step 2: Fee Schedule Lookup
Each claim line is priced against the applicable fee schedule with priority ordering:
- Provider-specific contracted rate (highest priority)
- Network-level negotiated rate
- Plan default fee schedule
- Fallback: percentage of billed charges (typically 80%)
Step 3: Cost-Sharing Calculation
The engine applies member cost-sharing in the correct order:
- Deductible (individual and family tracking)
- Copay (based on service type: PCP, specialist, ER)
- Coinsurance (plan percentage of remaining allowed amount)
- Out-of-pocket maximum protection
Step 4: Disposition and Output
Based on the results:
- Approved: All lines priced, payment determined
- Denied: Pre-adjudication rule triggered (with CARC/RARC codes)
- Partial: Some lines approved, some denied
- Pended: Requires human review (prior auth, unusual amounts)
The 85-95% Auto-Adjudication Target
Modern rules engines achieve 85-95% auto-adjudication rates because most claims are routine:
- Office visits with standard E/M codes
- Lab work and diagnostic imaging
- Preventive care and well-child visits
- Prescription refills
- Physical therapy sessions
The remaining 5-15% are pended for examiner review, typically complex cases involving:
- High-cost procedures requiring prior authorization
- Out-of-network emergency services
- Coordination of benefits with other payers
- Unusual billing patterns flagged by anomaly detection
Why Speed Matters
Sub-2-second adjudication isn't just a technical flex. It has real business impact:
- Cash flow: Faster adjudication means faster payment to providers
- Member satisfaction: Claims resolved in hours, not weeks
- Staff efficiency: Examiners focus on complex cases, not routine processing
- Error reduction: Deterministic rules don't have bad days or miss steps
- Scalability: Adding volume doesn't require adding headcount
How to test this without switching anything
The transition from manual to automated adjudication doesn't have to be all-or-nothing. A claims analysis lets you:
- Run the auto-adjudication engine on a claims extract your current administrator already produces
- Compare results line by line: speed, accuracy, cost-sharing calculations
- Build confidence before switching over
- Identify edge cases specific to your plan designs
SmartTPA's claims rerun is built for exactly this. Send a claims extract your current administrator already produces, and we run every line through the engine and report what it would have paid and how much of the file cleared without a human. No cost, no data migration, and you keep the report either way, including the version that tells you your current setup is fine. Start a claims rerun.