Health AI offers solid ROI, but threatens workforce cuts

Where AI ROI impact will be most keenly felt

AI is providing strong returns in healthcare faster than expected.

Last year, healthcare executives said that it would take at least two years for AI to produce returns. But now, they are saying that the ROI is being realized in about one year. The ROI averages 3.5 times the original investment, exceeding expectations in about 40% of use cases.

"That survey finding is consequential, as nearly every AI budget approved over the past two years was built on a payback assumption that turned out to be wrong by half," the report authors noted. "Organizations still modeling two-year paybacks are underinvesting against their own results."

In some areas, AI ROI is even higher. The survey shows that in the provider revenue cycle arena, ROI is 4 times the initial investment, and 67% of respondents report using semi- or fully autonomous AI agents for revenue cycle management. Another 49% of respondents said they use semi- or fully autonomous AI agents for front-office operations. In contrast, only 4% said they are using these types of agents for clinical work.

AI adoption for payer contracting management grew the most among providers, doubling from 19% of respondents using AI for contracting last year to 37% this year. AI use for prior authorization also grew from 32% to 46%.

Payers are seeing the strongest ROI in claims management, with returns of 3.4 times the original investment. Member engagement is proving to be a popular area for running semi- or fully autonomous AI agents, with 62% of respondents saying they do so.

AI utilization also grew significantly for provider credentialing and enrollment, from 35% of respondents using AI for this in 2025 to 56% in 2026.

However, the strengthening of AI's ROI could be bad news for healthcare workers. Organizations across the provider, payer and pharma segments expect to cut an average of 8% to 13% of their workforce in the near future due to AI implementation.

More than half of small (64%), mid-sized (52%) and large (64%) providers plan to reduce their workforce. These reductions are expected to be concentrated in the back office, with 73% of providers saying they plan to cut RCM and medical billing roles, followed by 49% citing cuts in clinical research roles.

More mid-sized payers (69%) than small (46%) or large (36%) payers said they plan to cut jobs, with most aiming to cut workers involved in claims processing, management and operations (71%). About 68% also plan to cut member or customer service roles.

AI adoption is moving past experimentation

More broadly, healthcare AI is crossing over from the proof-of-concept phase to pilots and scaled initiatives, the report stated.

In 2026, 28% of providers are in the pilot phase and 12% in the scaled implementation phase for AI tools, compared with 23% in the ideation phase and 26% assessing proof of concept. Among payers, 31% are in the pilot phase and 5% are in the scaling phase in 2026, versus 29% and 28% in the ideation and proof-of-concept phases, respectively.

Across both providers and payers, AI use cases for patient-facing services accounted for the largest share of scaled AI implementations. These include AI-driven front office, RCM and member services.

On the other hand, clinical AI use cases only reach the scaled implementation phase about 50% of the time, even though they reach the proof-of-concept phase at similar rates to front-office AI. Regulatory and compliance uncertainty is a major factor holding clinical AI back from broad deployment, with 36% of providers reporting they have delayed or canceled clinical AI implementations due to this uncertainty.

About half of the provider respondents also said that liability and legal exposure from AI-based clinical care are barriers to scaling clinical AI from proof of concept to full deployment.

Anuja Vaidya has covered the healthcare industry since 2012. She currently covers healthcare IT and innovation, including artificial intelligence, digital healthcare, EHRs and interoperability.