Mercer’s National Survey of Employer-Sponsored Health Plans, released August 31, 2026 from over 1,800 employer respondents, projects an 8.2% average increase in health benefit costs per employee for 2027 — the largest single-year jump since 2003. The headline is already getting attention. The finding with direct IBNR implications is buried in the driver decomposition.
Mercer isolates three contributors that each add approximately one percentage point above the underlying trend: GLP-1 weight-management drugs, No Surprises Act independent dispute resolution (IDR) process costs, and AI-enabled physician billing software. The third driver — AI billing — is the one that has not yet surfaced in most IBNR reviews, because it has no analog in the historical paid data that anchors professional services development factors.
What AI billing software actually does to claims
Physician practice management platforms have integrated AI tools to optimize evaluation and management (E/M) code selection since 2024. The software reviews encounter documentation and flags cases where a higher-acuity E/M code is defensible under the 2021 AMA office visit guidelines. The effect is a systematic shift in the submitted code mix toward level 4 and level 5 visits, which carry higher allowed amounts than the level 3 visits they replace.
For a self-funded employer plan, the result is that the same patient panel, at the same utilization rate, generates higher professional services claims than it did in 2022 or 2023. This is not a trend problem. It is a base problem. A plan that applies a 7% or 8% professional services trend to a 2023 PMPM benchmark is growing from a starting point that understates the current claim level by roughly the amount of the code mix shift.
That distinction matters for IBNR. Most self-funded plan actuarial opinions set the expected claim ratio by trending historical PMPMs forward to the current period. If the historical PMPM predates the AI billing adoption curve (broadly: anything before mid-2024), the expected ratio is understated before the actuary applies a single development factor.
Who it affects
Self-insured employers who use actuarial opinions for IBNR, particularly those whose most recent reserve study used 2022 or 2023 as the primary credibility period. Plans renewing stop-loss coverage for 2027 using pre-2024 experience to set aggregate attachment points carry the same exposure: the attachment is calibrated to a plan cost baseline that no longer exists.
Hospital systems, universities, and public entities running self-funded health benefits through a captive or a trust face the same base-calibration gap. The Business Group on Health 2027 Employer Healthcare Strategy Survey, which drew from 127 large employers representing 11 million covered lives, projects a median 8.5% to 9.2% health care cost trend for 2026 into 2027, corroborating the Mercer finding that five consecutive years of elevated growth are not reverting.
The compounding problem
AI billing does not operate in isolation. The Mercer survey identifies two other one-point drivers alongside it. NSA IDR process costs are a separate structural addition: providers initiating arbitration months after initial payment can reopen claims that a development triangle would already show as closed. (The mechanics of the IDR late-development tail are covered in the August 21 piece on Health Plan IDR Rulings Add $5B to Costs, Skewing IBNR Models.) GLP-1 coverage decisions introduce a third layer. Plans that retained obesity coverage for 2027 carry higher near-term pharmacy PMPM; plans that dropped it face deferred diabetes and cardiovascular costs 24 to 36 months out. (See the earlier GLP-1 Cost Surge Exposes Pharmacy and Stop-Loss Planning Gap.)
If all three drivers are additive and a plan’s IBNR model still uses pre-2024 professional services experience, the cumulative understatement of 2026 ultimate plan cost could reach 2 to 3 percentage points before plan design offsets are applied. Against a typical self-funded IBNR margin of 3 to 5% of estimated incurred claims, that is a material gap. On a 10,000-member plan spending $14,000 per employee, a 2-point understatement of ultimate costs is roughly $2.8 million of unrecognized liability.
What this means for your next review
Ask your actuary which credibility period anchors the professional services expected PMPM and whether the 2024 and 2025 diagonals have been weighted heavily enough to reflect the AI billing shift. If 2022 or 2023 is the primary credibility anchor, request a sensitivity run using only post-2023 experience. The gap between that run and the base opinion is a reasonable proxy for the exposure the AI billing driver creates.
For context on the development factors that compound this base problem, the IBNR for self-funded health plans explainer covers how professional services development patterns are constructed and where a code-mix shift would appear in the triangle diagnostics.