Healthcare AI Has a Measurement Problem. Revenue Cycle May Be Where CFOs Can Solve It First

Deloitte found that healthcare organizations scaling AI fastest are often worse at proving its financial value. Revenue cycle offers something many AI use cases do not: measurable baselines, financial outcomes, and transaction-level economics.

Healthcare’s AI conversation is changing.

For the past two years, the dominant question has been whether health systems and health plans would adopt generative and agentic AI.

That question is increasingly settled.

The harder question now coming from the CFO’s office is much simpler: What are we actually getting for the money?

New research from Deloitte exposes a striking gap between AI adoption and AI accountability. In a 2026 survey of 64 U.S. healthcare CFOs and finance leaders, Deloitte classified 44% of respondents as “AI scalers”—organizations deploying generative AI more broadly across functions.

Yet only 18% of those AI scalers reported mature financial attribution capabilities. Among organizations earlier in their AI journey, the figure was actually higher at 31%.

In other words, some of the organizations moving fastest on AI are among the least equipped to prove what that AI is worth.

For revenue cycle leaders and the companies selling AI into RCM, that may be one of the most consequential signals of 2026.

Capital Is Moving Faster Than Measurement

Healthcare CFOs aren't necessarily responding to the measurement problem by cutting AI investment. They're increasing it.

Deloitte found that 75% of AI scalers expect to increase generative and agentic AI investment over the next 12 months. Eighty-five percent expect those investments to break even within five years. And 64% expect at least 5% annualized cost savings from AI initiatives within one to two years.

Those are meaningful expectations. But expectations aren't realized ROI.

Deloitte separately analyzed more than 17,000 public articles from 62 health systems and health plans. Among the technology-focused communications it examined more closely, only 19% of the 2026 narratives reflected demonstrated value—up from 9% in 2023, but still substantially behind discussion of anticipated benefits.

And even when organizations reported demonstrated value, Deloitte found that the evidence more commonly involved operational or clinical improvements than clearly attributable financial outcomes.

Why Revenue Cycle Is Different

This is where RCM becomes particularly interesting.

Many healthcare AI use cases have inherently difficult ROI equations. How much is a better clinical note worth? What is the financial value of reducing physician cognitive burden? How should a health system value a better patient experience?

Those outcomes matter enormously, but translating them into an attributable P&L impact can be difficult.

Revenue cycle starts with something different: an existing financial system full of measurable transactions.

A health system already knows—or should know—its baseline for metrics such as:

• denial rate and overturn rate
• days in accounts receivable
• clean-claim rate
• coder productivity
• cost per account or claim touched
• authorization turnaround time
• cash collections
• write-offs and leakage
• staffing requirements

Deloitte makes essentially the same point in its research, specifically identifying denial overturn rates, days in AR, coder productivity, rework, cash acceleration and cost per claim touched as potential measures for an RCM AI copilot.

That makes revenue cycle one of healthcare AI's potentially best proving grounds.

An AI agent doesn't merely have to demonstrate that it can perform a task. It can be asked to demonstrate that performing that task changed an economic outcome.

The Next RCM AI Battle May Be Attribution

That has implications for vendors.

The first generation of healthcare AI selling focused heavily on capability: our AI can code this chart, our agent can work this denial, our model can call this payer, our copilot can summarize this account.

The next generation will increasingly have to answer: What happened financially because it did?

That is a substantially higher bar.

A denial agent that touches 100,000 accounts isn't necessarily valuable because it automated 100,000 touches. The relevant questions are whether the intervention increased overturns, accelerated cash, reduced labor, prevented write-offs or produced some combination of those outcomes—and whether those benefits exceeded the fully loaded cost of running the AI.

The vendors capable of establishing that attribution could gain an important advantage as AI purchasing shifts from innovation budgets toward enterprise operating budgets.

Then There Is Token Economics

Deloitte identifies another issue that RCM buyers may soon have to confront: AI doesn't always behave economically like traditional software.

Traditional enterprise software is often priced by seat, module or annual license. Generative AI introduces a variable consumption component.

Prompts, retrieval, outputs, orchestration and interactions among agents can all generate incremental cost. In high-volume workflows—including revenue cycle—those economics can become material.

A model that saves three minutes of labor on an account isn't necessarily economical if the AI infrastructure required to process that account costs more than the capacity created.

Cost per denial prevented. Cost per authorization completed. Cost per dollar collected. Cost per chart coded correctly. Cost per account resolved without human intervention.

That is a much more rigorous way to evaluate autonomous RCM.

RCAI View

The healthcare AI market isn't entering an adoption crisis. It's entering an evidence phase.

Deloitte's findings suggest healthcare executives remain remarkably optimistic about AI. The constraint increasingly isn't willingness to spend. It's the ability to connect that spending to measurable financial performance.

That should favor RCM.

Revenue cycle sits at the intersection of administrative labor, enormous transaction volume and directly observable financial outcomes. Few areas of healthcare provide such a natural environment for measuring whether automation actually creates enterprise value.

But it also raises the standard for the RCM AI market.

The winners in the next phase of RCM AI may be the platforms that can prove, account by account and workflow by workflow, exactly what changed after their technology entered the process—and what that change was worth.

That is a much harder standard than demonstrating AI capability. It is also the standard healthcare CFOs appear increasingly ready to impose.