PwC's June 2026 survey of actuaries at 27 U.S. health plans projects a 9.0% group medical cost trend for 2027 (PwC, June 2026), the highest reading since 2010. Seventy percent of those plans ranked provider AI revenue optimization tools among their top three cost drivers, a structural inflator that compounds across consecutive rating periods without adding clinical volume.

In large-group health rate filings, the provider AI coding signal was visible at the encounter level before it accumulated large enough to appear in aggregate statistics. The evaluation-and-management level distribution was drifting toward CPT 99215 at rates no underlying acuity shift could explain: contracted fee schedules rising 3% to 4% annually, but average allowed per professional claim growing at 7% to 10%. Revenue cycle vendors deploying computer-assisted coding tools were optimizing E&M level selection in real time, and the delta between the fee schedule increase and the average allowed growth was billing precision, not clinical intensity. PwC's Behind the Numbers 2027 report is the first industry-level confirmation that this effect has concentrated large enough to rank alongside GLP-1 pharmacy spending and No Surprises Act dispute costs in the 9% composite.

The five structural drivers in PwC's report are not equally tractable. GLP-1 pharmacy cost has a plausible deceleration scenario via biosimilar entry. Behavioral health utilization growth is supply-constrained and MHPAEA-driven, making it more predictable than cyclical. No Surprises Act adjudication creates a systematic ratchet on out-of-network reimbursement. General fee schedule inflation reflects provider consolidation leverage that neither network design nor prior authorization addresses. Provider AI revenue optimization is distinct from all four: it extracts incremental reimbursement per existing encounter without any change in the volume or acuity of services rendered, and it compounds forward as tool adoption continues to expand across hospital systems. That is the inflator plan actuaries have the fewest established methods to isolate, measure, and price separately from trend noise.

How Provider AI Revenue Tools Work at the Billing Layer

The tools fall into four distinct categories, each targeting a different phase of the revenue cycle and each adding an independent increment to the average reimbursement per encounter.

Computer-assisted coding (CAC) systems use natural language processing to parse clinical documentation and suggest or assign the highest clinically defensible procedure and diagnosis codes. Coding errors, including undercoded encounters, account for 40% to 80% of billing accuracy problems across hospital systems and represent the largest single category of revenue leakage hospitals face (Combine Health, 2025). A missed comorbidity in a hospital chart, a chronic condition documented in the clinical note but not translated into a billable ICD-10 code, or an office visit coded at 99213 when the documentation supports 99215: CAC tools identify these at the encounter level and flag them for correction before claim submission. Forty-six percent of hospitals already use AI in revenue cycle functions, and 74% use some form of RPA or automation in this space (MicroscopeHC, 2025). As adoption extends from large academic health systems into community hospitals and employed physician groups, the coding precision effect spreads across a larger share of the claims population.

Prior authorization prediction AI analyzes historical payer adjudication decisions to estimate which authorization requests are likely to be approved, which will require additional clinical documentation, and which should be contested. Providers use this to pre-sequence requests, front-load supporting clinical data for complex procedures, and route authorization requests through the pathways that minimize first-pass denial. The average initial claim denial rate across commercial payers runs at 11.65%, meaning more than one in nine claims is rejected on first submission (industry, 2025). Prior auth AI compresses this rate by modeling the payer's adjudication logic and giving the provider an informational advantage in the authorization sequence, pulling forward both approval timing and claim submission dates.

Discharge optimization AI identifies inpatient cases where clinical documentation supports additional covered hospital days, suggesting defensible length-of-stay extensions that maximize facility revenue per admission. Denial prevention AI monitors claims in real time before submission, applying predictive scoring models trained on the specific payer's historical denial patterns to flag any claim at elevated risk and recommend pre-submission corrections. Vendors including SmarterDx, Waystar, and Availity now offer these systems integrated directly into EHR platforms including Epic, Oracle Cerner, and MEDITECH (Combine Health, 2026). The combined effect across all four categories is higher average allowed charges per encounter, faster claim submission, and improved first-pass acceptance, all of which flow through to payer cost trend without any change in the underlying volume of clinical services delivered. Health plans view this as a fundamental transparency problem: "cost management success," PwC's report states, "depends on pairing disciplined contracting with advanced capabilities in coding-intensity surveillance, severity-shift monitoring, and payment integrity to distinguish changes rooted in documentation from shifts in patient complexity" (PwC, June 2026).

Provider Market Concentration as the Pricing Base

The AI revenue optimization effect does not operate in isolation. It layers on top of a provider market that has already concentrated to a degree that significantly limits payer contracting leverage, turning a billing precision gain into an amplified reimbursement extraction.

In 2024, one or two health systems controlled the entire market for inpatient hospital care in 47% of U.S. metropolitan areas. In 83% of metro areas, one or two systems controlled more than 75% of inpatient market share (KFF, 2024). Four of five metropolitan areas saw hospital market concentration increase between 2015 and 2024 (KFF, 2024). The physician market has followed the same trajectory: at least 47% of physicians were employed by or affiliated with hospital systems in 2024, up from fewer than 30% in 2012 (GAO, 2024). In concentrated markets, providers extract above-market reimbursement rates through contracting leverage at the moment of network renewal. AI revenue optimization adds an independent increment per encounter between renewals: the negotiated rate does not change, but the average allowed per claim rises because more of the clinical value documented in the record is captured in the submitted code set.

Network narrowing and enhanced prior authorization programs, the standard payer tools for managing provider cost, cannot offset a billing optimization effect that operates at the code selection layer rather than the utilization or rate layer. A plan can narrow its network to tier-one in-network facilities and still absorb the full coding precision increment from those facilities' AI tools, because the increment arrives through the charge per encounter, not through the volume of encounters. This is the compounding mechanism that makes provider AI revenue optimization a structural inflator across multiple rating periods: adoption extends, average allowed per encounter rises, and the historical regression used for trend selection reflects a mix of pre- and post-adoption claims that understates the forward run rate if the transition period is not identified and handled explicitly.

Pharmacy Running Near Double-Digit

GLP-1 prescription volumes nearly doubled across the commercially insured population between December 2024 and December 2025 (PwC, June 2026), a growth trajectory already embedded in the 2027 composite. The spending impact in employer plans is visible at the line level: per-member per-month GLP-1 expenditure jumped from $4.34 in 2022 to $27.23 in the first quarter of 2025 (SHRM/EBRI, 2025), a 6.3-fold increase in three years, and GLP-1 claims grew from 6.9% of total employer prescription spending in 2023 to 10.5% by 2025 (SHRM, 2025). More than 85% of surveyed health plan actuaries expect 2027 pharmacy trend to outpace the overall medical composite (PwC, June 2026). A plan running a 36-month log-linear regression on total pharmacy PMPM without separating GLP-1 claims is fitting a curve whose slope changes midway through the estimation window, and the resulting trend understates the forward run rate unless the adoption acceleration is explicitly modeled, as our GLP-1 trend factor analysis describes in detail.

The 2027 to 2028 window introduces genuine two-sided uncertainty. Biosimilars to injectable semaglutide and tirzepatide are expected to enter the market, and oral GLP-1 formulations began reaching patients in 2026 at initial monthly prices near $200. If biosimilar price competition materializes within the current pricing period, pharmacy trend could decelerate materially from the near-double-digit run rate currently embedded in 2027 filings. Plans that embed the higher trajectory without scenario-weighting the biosimilar pathway may overprice 2027 and find themselves competitively exposed at renewal. The direction of the uncertainty is asymmetric: the floor on GLP-1 spending is now structural given coverage commitment from large employer groups, but the ceiling depends on pricing dynamics that are genuinely unresolved.

Behavioral Health Coding Intensity and the MHPAEA Expansion Effect

Post-pandemic normalization of mental health care-seeking has driven sustained behavioral health utilization growth, and MHPAEA parity requirements have expanded the set of services covered and the network of providers available to deliver them. The number of in-network behavioral health providers increased by 48% across commercial health plans over three years as parity enforcement pushed networks to add credentialed capacity (PwC, June 2026). The DOL's 2024 Report to Congress, published in January 2025, found that none of the health plans examined in a compliance review were fully compliant with MHPAEA parity requirements, particularly around network adequacy and non-quantitative treatment limitations (DOL, January 2025). As enforcement catches up to the statute, behavioral health utilization will continue expanding into service categories currently subject to restriction.

AI diagnostic coding tools are operating in behavioral health settings in the same way they operate in inpatient medicine: identifying undercoded encounters, capturing diagnostic complexity at higher acuity levels, and improving submission velocity. In mental health outpatient settings, this creates a dual compression on trend. Encounter frequency is rising as more people seek care from a larger network of credentialed providers. Average acuity per visit is also rising, because AI coding tools capture comorbid conditions and diagnostic specificity that were previously not translated into billable codes. A plan that runs behavioral health as a single trend line sees the aggregate increase; a plan that decomposes it into encounter frequency and acuity intensity per visit can select forward assumptions for each component with more precision and document the differential clearly in the filing memorandum. The two drivers respond to different management interventions, and conflating them in a single trend factor produces a selection that is neither accurate for pricing nor defensible in a regulatory review.

No Surprises Act Adjudication and the Dispute Learning Loop

More than 5.1 million disputes had been submitted to the No Surprises Act independent dispute resolution portal as of January 31, 2026 (CMS, 2026), and 1.2 million new disputes were filed in the first half of 2025 alone, a volume 40% higher than in the preceding six months (CMS, 2026). The IDR process was designed to resolve individual billing disputes, but at this volume it has become a systematic mechanism through which providers extract above-contract reimbursement for out-of-network services at scale.

Provider-side AI tools are now modeling payer adjudication patterns in the IDR portal: analyzing which dispute types result in awards above the qualifying payment amount, which clinical documentation profiles are most persuasive to arbitrators, and which payer dispute response patterns signal a likely settlement. This predictive capability shifts the economics of OON dispute resolution in the provider's favor. Disputes with below-expected award probability are settled or abandoned; disputes with favorable expected outcomes are contested with optimized documentation. The result is a higher average recovery rate on contested claims and a systematic ratchet on the effective OON reimbursement level over time, independent of any change in contracted in-network rates. As providers accumulate IDR award data, they also use it as a floor in in-network contract renegotiations, giving high-award-rate specialties a benchmarked reference point for what a payer can be compelled to pay. For plan actuaries, OON trend in emergency medicine and anesthesiology may be running 15% to 25% above the in-network medical composite in markets with high IDR adoption; a blended medical trend selection that aggregates OON and in-network costs will understate the prospective OON exposure, particularly for self-funded employer plans without tight OON cost-sharing structures. The CMS prior authorization metrics release from January 2026 provides denial rate data by service category that anchors the utilization management side of this calculation.

IBNR Lag Recalibration for AI-Accelerated Billing

The actuarial response to the provider AI inflator is not limited to selecting a higher trend point. It also requires recalibrating the completion factors used in IBNR development for service categories where billing submission velocity has materially changed.

Standard IBNR development assumes that the timing pattern of claim submission is reasonably stable across the estimation window, so that completion factors derived from historical paid claim runout accurately estimate the incurred liability at any given lag. AI-accelerated revenue cycle tools compress this submission cycle in two ways. Prior authorization AI and denial prevention AI both pull forward the claim submission date by eliminating the revision loops that delay initial submission after a first-pass denial. Electronic CAC tools reduce chart turnaround time, with vendors reporting 40% to 70% reductions in the time from service delivery to coded claim submission (BillingParadise, 2025). Where these tools have been adopted in concentrated specialties, particularly high-cost procedures in anesthesiology, surgical specialties, and behavioral health, the historical completion factors for those service categories may systematically understate how quickly claims are now accumulating.

The diagnostic test is a before-and-after comparison of the paid claim runout triangle for affected service categories. If the 3-month and 6-month paid percentages have shifted upward relative to the historical pattern, the completion factors need recalibration to reflect the current submission velocity. Applying completion factors developed from a slower-reporting historical window to a faster-reporting recent period overstates IBNR, carries excess reserve into premium pricing, and can produce a calculated trend that understates the true forward run rate when the excess reserve unwinds. For large-group pricing where the experience window is already short, a service-category completion factor analysis that identifies the AI tool adoption date for major in-network hospital systems and splits the runout triangle before and after that date produces a more accurate IBNR and a more defensible experience trend than a uniform historical development approach.

The broader actuarial task for 2027 rate filings is decomposing the 9% composite into four components: unit cost, encounter volume, coding intensity, and true acuity change from population morbidity shift. Provider AI revenue optimization sits inside coding intensity, not acuity. The two are not interchangeable in the trend selection or in the filing memorandum. A plan that attributes the billing optimization signal to deteriorating member morbidity will select a trend with the right headline number but the wrong mechanistic interpretation, and when regulators or clients ask which driver is expected to revert and which is structural, the filing will not have a clear answer. The rate filing benchmark methodology is covered in our companion analysis on PwC's 9% benchmark and the credibility weighting problem. What the companion piece describes as the AI coding step-change to be stripped from the experience regression is, at the mechanism level, exactly what the four tool categories described here produce at the billing layer.

Further Reading

Sources