Payers are now naming clinical AI as a top cost driver. The 2027 PwC numbers are uncomfortable for everyone.

PwC’s annual Behind the Numbers medical-cost-trend report is the one document the actuarial side of U.S. healthcare actually reads. For 2027, PwC projects a 9% medical cost trend in the group market and 8.5% in the individual market — the highest single-year projection in roughly seventeen years.

The drivers PwC lists are familiar: pharmacy spend (anchored by GLP-1s), behavioral-health utilization, and out-of-network disputes under the No Surprises Act. The new entry — and the one getting the attention of the trade press — is provider use of AI.

Specifically: nearly 70% of plan actuaries surveyed by PwC ranked AI-enabled documentation as one of the top three inflators for 2027, and about 20% identified it as the single largest driver. That is the first time clinical AI has registered as a first-order cost story in a payer trend report, and it changes the political economy of the technology in ways that the engineering and clinical-informatics conversations have not yet caught up to.

What the actuaries are actually claiming

The PwC narrative is straightforward enough. Ambient-scribe tools, AI-augmented coding assistants, and revenue-cycle automation let providers capture diagnoses and procedure complexity that, under purely manual documentation, often went unrecorded. The same encounter, more thoroughly coded, generates a higher allowed amount under fee-for-service contracts and a higher risk-adjusted payment under value-based contracts.

In actuarial language, this is upcoding without proportional intensity — patients getting more specifically documented care, but not measurably more care. The unit of care delivered is unchanged; the unit of payment is higher.

It is important to be clear about what this claim is and isn’t.

It is not a claim that AI documentation is fraudulent. The diagnoses being captured are, in most cases, ones a sufficiently diligent human coder would also have captured. The shift is that AI makes diligence the default rather than a sometimes-funded ideal.

It is not a claim that AI doesn’t reduce clinician burnout. The ambient-scribe deployment literature on burden reduction is by now unusually well-evidenced for a clinical-AI category — the 40–60 minute daily savings per primary-care physician shows up consistently across systems and study designs.

It is a claim that two things can be true at once: clinical AI reduces clinician administrative burden and increases payer cost. The reduction-in-burden case is what providers and vendors talk about; the increase-in-payment case is what actuaries are now beginning to publish.

Why this changes the conversation

For most of the last three years, the public narrative around clinical AI has been framed around clinician time, patient experience, and access. The PwC report changes the lens.

If 70% of plan actuaries think AI documentation is in their top three cost drivers, several things follow over the next 12–24 months.

Coverage and contracting language will tighten. Expect payer-provider contracts to introduce documentation-density adjusters, audit clauses for AI-generated note content, and explicit definitions of which AI tools are considered part of normal documentation versus revenue-enhancement. A handful of large commercial plans have already begun negotiating this; PwC’s framing accelerates that across the industry.

Risk-adjustment models will be re-tuned. CMS-HCC, the Medicare Advantage risk-adjustment model, and the commercial HHS-HCC variant were not calibrated for an era in which AI captures every documentable comorbidity at every encounter. The actuarial community will push for recalibration that strips out AI-driven documentation intensity from the risk score. That is a multi-year regulatory fight, and it will be ugly.

Some AI documentation tools will be re-classified. Tools whose marketing literature openly promises higher reimbursement — as opposed to those positioned around clinician time savings — sit in the most exposed position. Expect OIG, MA plan audits, and commercial-plan audits to start treating those vendors’ workflows as a category worth scrutinizing.

The ROI conversation inside provider organizations will shift. Right now, ambient-scribe and AI-coding deployments are typically justified to provider leadership in three buckets: clinician retention, encounter throughput, and incremental revenue capture. The third bucket is the one that the PwC report puts at risk. CMIOs and CFOs who modeled clinical-AI ROI heavily against revenue uplift will have to model against the medium-term scenario in which payers price the uplift out.

What the report does not settle

A few important questions remain genuinely open.

How much of the AI-attributed inflation is actually AI? AI documentation tools spread fastest in organizations that were already investing in revenue-cycle modernization, value-based-care infrastructure, and physician-incentive realignment. Disentangling the AI effect from the broader investment effect is hard, and PwC’s report is a survey of actuarial attribution, not a measurement of isolated AI impact.

What is the patient-outcomes counterfactual? If a documented comorbidity changes downstream care — earlier referral, more aggressive risk-factor management, more accurate transitions of care — then the “no proportional intensity increase” claim is incomplete. There is essentially no published outcomes data on this yet, and the absence is itself a finding.

What does the equity story look like? AI documentation deployment is currently concentrated in well-capitalized health systems. If those systems extract more reimbursement per encounter, the gap between them and resource-limited safety-net systems widens. That is a known dynamic in U.S. healthcare; AI accelerates it but does not create it.

The point

For three years, clinical AI has been mostly a story about clinicians and patients. With the 2027 PwC numbers, it becomes — also, additionally, not instead — a story about who is paying.

The cleanest summary of where this leads: the next generation of contracting between health plans and provider organizations will explicitly price AI documentation. The deployments that survive are the ones whose business case is robust to that pricing. The ones positioned principally around “capture more codes” will find themselves on the wrong side of a payer market that has now seen this curve and named it.

The Behind the Numbers report itself is a forecast, not a verdict. But it is the first time an actuarial document this widely read has put clinical AI in the same paragraph as GLP-1s and the No Surprises Act. That is, on its own, a meaningful event in the industry conversation.