UnitedHealth Says It Will Spend $1.5 Billion on AI in 2026 as Optum Real Targets 2.5 Billion Healthcare Transactions
UnitedHealth CEO Stephen Hemsley told investors the insurer will spend roughly $1.5 billion on AI in 2026, with a target of 2:1 returns, even as the STAT+ investigation puts the total bet at $3 billion and questions what the shift means for patients.
Overview
The largest private health insurer in the United States is making one of the largest AI bets inside the U.S. healthcare system. On its first-quarter 2026 earnings call on April 21, UnitedHealth Group said it expects to spend roughly $1.5 billion on AI in 2026, routing roughly one-third through Optum Insight’s transition to an AI-first software and services business and the remaining two-thirds across internal processes like claims handling, member services and back-office functions, PYMNTS reported. The company also raised its full-year adjusted earnings forecast to more than $18.25 per share, up from a prior outlook of more than $17.75, according to UnitedHealth Group’s official first-quarter release carried by BusinessWire.
Earlier in April, STAT News described the investment as a “$3 billion bet on AI” and reported that UnitedHealth employs 22,000 software engineers, with more than 80 percent of them using AI either to write code or to build new agents. The two figures are not inconsistent: the $1.5 billion disclosed on the earnings call is the 2026 budget, while the STAT framing captures a broader multi-year effort that now touches essentially every part of the company.
What We Know
The centerpiece of the 2026 build-out is a pair of production systems that are already processing transactions at scale.
The first is a real-time claims and reimbursement platform the company calls Optum Real. On the earnings call, executives told investors the platform is on track to handle about 500 million transactions in the first half of 2026 and 2.5 billion by year-end, according to PYMNTS. UnitedHealth is pitching Optum Real as a system that collapses multi-day reimbursement cycles into near-instant system-to-system exchanges, with AI embedded in claims adjudication, prior authorization, pharmacy approvals and provider payments. A new Optum Insight digital prior authorization solution is already showing 96 percent first-pass approval rates on early submissions, Fierce Healthcare reported.
The second is Avery, a generative AI assistant for UnitedHealthcare members that the company launched in late March and expects to reach more than 20 million members by the end of 2026, Fierce Healthcare reported. Avery is designed to answer coverage and billing questions, check claim approval status, surface cost estimates and schedule appointments without handing the call to a human agent.
Behind those two products sits an internal engineering stack. In a May 2025 interview with Fortune, Chief Digital and Technology Officer Sandeep Dadlani said UnitedHealth had roughly 1,000 AI use cases already in production and that about 20,000 engineers had accepted more than 60 million lines of AI-generated code through an internal environment called United AI Studio. Dadlani also told Fortune that UnitedHealth had set up a responsible AI board of internal and external technicians, clinicians and legal experts that reviews hundreds of use cases each month before they are promoted to production.
The investment is pitched as a margin story. On the earnings call, Dadlani told investors the company expects “a return conservatively of 2 to 1” on the AI programs over the next few years, with many individual use cases paying back within 12 to 18 months, according to PYMNTS’s coverage. The first-quarter numbers are consistent with that thesis even before most of the spending has landed. UnitedHealth’s Q1 release shows consolidated revenues of $111.7 billion, adjusted earnings of $7.23 per share and a medical cost ratio of 83.9 percent, down 90 basis points from a year earlier — figures CNBC confirmed in its report on the Q1 headline beat.
The financial scope of the 2026 push is only part of the story. STAT’s reporting, published before the earnings call, framed the broader shift in more pointed terms. “The effort involves building up engineering teams to reinvent how billions of medical claims are processed and audited, automating everything from fraud detection, to clinical documentation, to the selection of billing codes that determine how much a given medical encounter costs — and who pays,” the outlet reported. STAT’s piece, part of its “Health Care’s Colossus” series, also quoted Dadlani saying the company has “doubled down on training, on investments, on driving meaningful use cases” since the advent of generative AI. Independent coverage of the earnings call by Managed Healthcare Executive and Hospice News confirms the $1.5 billion figure and the same one-third / two-thirds allocation between Optum Insight’s software push and core enterprise processes.
What We Don’t Know
Several important questions remain open.
The most basic is the boundary between automation and clinical judgment. UnitedHealth has a history here: in the May 2025 Fortune interview, Dadlani said flatly that “AI is never used to deny a claim,” explaining that roughly 90 percent of claims are auto-adjudicated by rules-based software and that of the remaining 10 percent, 98 percent are ultimately approved, with denials driven by ineligible benefits or human clinical determinations. Independently verifying that boundary — especially as prior authorization, pharmacy approvals and provider payments move onto the Optum Real substrate — requires disclosures the company has not yet made public.
The STAT investigation explicitly flagged that gap. Its framing asks what the new stack “means for patients,” noting that the systems increasingly decide how much a medical encounter costs and who pays for it. The full investigation sits behind STAT+, so the specific case studies and documents underpinning that framing are not publicly available, but the headline claim — a multibillion-dollar reorganization of how a major insurer processes care decisions — is.
A second open question is the quality and reach of Avery. Neither the company’s disclosures nor outside coverage provide independent benchmarks on how often the assistant resolves member questions correctly, how frequently it escalates to human agents, or how error rates compare to existing call-center performance. The target of reaching 20 million members by the end of 2026 is ambitious; the performance characteristics of the system once it is operating at that scale are not yet observable.
A third is how much of the 2026 spend survives beyond UnitedHealth’s walls. Dadlani told investors that the third of the $1.5 billion flowing into Optum Insight is intended to produce software and services that can be sold to other healthcare organizations, per PYMNTS. Whether hospitals and rival payers will buy AI tooling from the largest commercial payer in the market — and what governance conditions they will attach — is an open commercial and political question.
Finally, the company’s own disclosures leave the full multi-year envelope ambiguous. The $1.5 billion figure is the 2026 budget; STAT’s $3 billion number appears to capture a broader multi-year commitment. UnitedHealth has not published a consolidated, audited disclosure of the total AI investment across years, which makes direct comparisons with other hyperscale enterprise AI programs — from OpenAI’s enterprise push to the agentic deployments announced by rival payers and providers — difficult to calibrate.
Analysis
Two things are simultaneously true about UnitedHealth’s April disclosures.
The first is that, measured against the benchmarks its own executives are willing to defend on the record, the deployment is concrete in ways most enterprise AI announcements are not. Optum Real is an active claims pipeline with a transaction count and a year-end target. Avery is a member-facing assistant with a specific rollout schedule and a defined scope of questions. The 2026 spending number is framed around a specific return profile — 2:1 conservatively, with individual programs paying back in 12 to 18 months — that investors can eventually measure against the medical cost ratio, the Optum segment’s operating margin and the company’s per-member cost to serve.
The second is that the same disclosures leave the governance question almost entirely to the company. The Responsible AI board Dadlani described to Fortune is internal. The publicly disclosed audits of specific use cases — whether on fraud detection, billing code selection, clinical documentation or prior authorization — are limited. The STAT investigation framed the tension directly: a private actor is rewriting large pieces of U.S. healthcare’s administrative spine on a budget comparable to the AI commitments of frontier model labs, with patient-facing consequences that are structurally hard for outsiders to observe.
That combination — visible deployment metrics, opaque decision boundaries — is likely to define the policy conversation around AI in health insurance for the rest of 2026. Regulators in both the Centers for Medicare and Medicaid Services and state insurance departments have already spent the last two years tightening disclosure around algorithmic decision-making in coverage determinations. The Q1 earnings call, by quantifying the investment and the expected return, effectively sets a floor on how hard those bodies will have to push to get matching disclosure on outcomes.
For now, the clearest read on the bet is financial. UnitedHealth raised guidance, its medical cost ratio improved, and it is pointing investors at AI-driven efficiency as a load-bearing piece of its 2026 turnaround story, per the Q1 release. Whether the same systems deliver a symmetric improvement for the 50 million-plus members whose claims they increasingly mediate is, as STAT’s framing implied, the question the rest of the pipeline will have to answer.