AI Is Creating a New RCM Responsibility: Who Will Own Token Economics?
AI is creating real value across the healthcare Revenue Cycle. As adoption accelerates, another crucial operational responsibility is emerging:
understanding how AI is actually being consumed at scale.
Artificial intelligence is moving rapidly into core Revenue Cycle workflows. Denials, prior authorization, patient access, eligibility verification, claims processing, appeals, analytics, and patient financial interactions are increasingly driven or augmented by generative models and intelligent agents.
To date, the industry conversation has understandably focused on capability and output-what AI can accomplish, how fast it operates, and how much manual work it eliminates.
However, as health systems transition from isolated point solutions to enterprise-wide adoption and autonomous agentic workflows, a fundamental operational question arises:
Who is responsible for tracking, optimizing, and accounting for the volume of AI an organization actually consumes?
At most health systems, this mandate will not start with a new job posting on day one. It will begin as an expanded operational responsibility for existing analytics, IT, or Revenue Cycle leaders. Eventually, it may mature into a dedicated discipline of its own.
At RCR|HUB, we believe the most fitting title for this emerging role is the AI Utilization Analyst. But we are asking the broader RCM CommUnity to weigh in and help us define it.
First, What Are AI Tokens?
To manage this emerging responsibility, healthcare leaders must first understand the basic unit of generative computing: the token.
Large Language Models (LLMs) do not process information by standard word count. Text is parsed into smaller units called tokens. As a general rule of thumb for English text, tokens are often estimated at roughly four characters or 0.75 words, although the actual token count varies based on the model, tokenizer, and content.
When denial letters, prior authorizations, or claim files are transmitted to an AI model, input tokens are consumed. When the model formulates a response, logic summary or appeal letter, output tokens are generated. Depending on the underlying architecture, organizations may also incur computational overhead for model API calls, context window caching, vector database queries, and cloud infrastructure.
For a single patient interaction, that token footprint may be minimal. But Revenue Cycle operates at significant transactional volume: large volumes of claims, authorization requests and patient interactions. Furthermore, multi-step AI agents may perform numerous behind-the-scenes actions to complete a single task.
This shift transforms AI oversight from a static software licensing conversation into a dynamic consumption model.
Tokens Are Only Part of the Story
While measuring tokens provides technical visibility, the eventual role will require an understanding of broader infrastructure dynamics far beyond raw token counts.
AI pricing architectures can encompass API request volumes, compute time, cloud hosting, specialized software vendor licenses, and recursive agentic actions.
This evolution mirrors the growth of FinOps (Financial Operations) in general cloud computing. As AI becomes a larger component of technology consumption, FinOps practices are increasingly extending financial accountability to AI workloads, including model and token usage, to connect technical resource consumption to business value.
For healthcare, this distinction is critical:
AI Consumption → Revenue Cycle Workflow → Operational Outcome
Counting raw tokens delivers limited insight to an RCM executive. The true value lies in mapping AI consumption to financial and operational outcomes.
Why Revenue Cycle Is Uniquely Positioned for This
Revenue Cycle management is inherently a discipline of rigorous measurement. Leaders continually manage complex operational key performance indicators (KPIs):
Clean Claim Rate (CCR)
Initial and Final Denial Rates
Days in Accounts Receivable (DAR)
Cost to Collect
Net Collection Rate
Prior Authorization Turnaround Time
AI introduces a new layer of operational telemetry that can integrate directly into these existing metrics. Instead of reviewing high-level enterprise AI spend, RCM leaders may want granular visibility at the workflow level:
How much AI activity supports denial management vs. front-end registration?
What is the token consumption associated with a successfully overturned denial?
Which autonomous agents generate the highest volume of model calls during patient intake?
Are reasoning-heavy models being deployed for simple tasks that smaller, specialized models could handle faster and more efficiently?
Asking these questions is not an argument against AI adoption. It is the natural progression of operating complex technology at enterprise scale.
The Monthly Invoice Is Not Enough
Managing AI cannot simply mean reviewing an aggregated monthly invoice from a software vendor or cloud provider.
Key Distinction: AI spend tells you what you paid. AI consumption data explains what actually happened.
As generative tools become deeply embedded across hospital IT ecosystems, health systems may require more granular operational visibility, including:
Model Routing: Which specific LLM or specialized model performed the task?
Workflow Source: Which department, user, or automated trigger initiated the request?
Execution Depth: How many intermediate reasoning steps or API calls were executed?
Yield Correlation: Did the AI interaction result in a paid claim, a cleared authorization, or a resolved billing query?
This telemetry can provide baseline data for departmental budgeting, forecasting, Business Partner evaluation, and workflow design.
Who Owns This Today?
Currently, this oversight may be fragmented across multiple enterprise functions:
Enterprise Role
Current Operational Focus
IT / MLOps / LLMOps
Monitors model performance, API uptime, latency, token throughput, and system architecture.
Finance & FinOps
Tracks IT budgets, software licensing, cloud spend allocation, and financial reporting.
AI Product Managers
Manages application functionality, user adoption, and specific feature roadmaps.
RCM Operations & Analytics
Tracks claim flow, denial trends, staff productivity, and net revenue optimization.
AI Governance & Compliance
Oversees data privacy, HIPAA compliance, security protocols, and responsible model use.
As AI adoption expands, the central challenge is bridging these disciplines. Health systems may require a dedicated function that understands both sides of the equation: what the AI is doing operationally and how it is being consumed computationally.
AI Agents Accelerate the Need for Oversight
The deployment of autonomous AI agents makes operational oversight increasingly important.
When a staff member uses a generative AI tool directly, consumption is generally constrained by the pace and frequency of human interaction.
An AI agent, however, can potentially execute multiple actions after receiving a task. An agent assigned to work an unpaid claim might retrieve relevant information, review payer policy documentation, query another system or API, evaluate response options, execute additional model calls, and draft an appeal without requiring a new human prompt at every step.
Traditional software metrics such as active seat licenses or user logins—may not fully reflect this multi-step background activity.
Organizations may increasingly need to measure the underlying computational work taking place within the workflow itself.
A New Set of RCM AI Metrics May Be Emerging
As healthcare organizations mature in their AI governance, Revenue Cycle teams may begin tracking operational unit economics that pair computational consumption with financial and operational performance:
AI Consumption per Clean Claim Processed
Token Consumption per Successfully Overturned Denial
Computational Cost per Prior Authorization Cleared
Model Call Ratio per Patient Access Interaction
AI Consumption by Specialty / Service Line / Care Setting
When computational usage is evaluated alongside traditional financial metrics like Cost to Collect, token data can transform from a raw technical metric into actionable business intelligence.
Changing the Business Partner Procurement Conversation
This evolution could change how healthcare systems evaluate and procure AI software partners.
While core evaluations around functionality, EHR integration, HIPAA security, and user experience remain paramount, procurement teams may introduce new requirements during Business Partner evaluations:
Consumption Telemetry: Does the Business Partner provide dashboards showing token consumption and API call volumes by workflow?
Model Transparency: Which underlying LLMs or specialized models are utilized for specific tasks, and can model routing be customized?
Pricing Safeguards: Are token, context-caching, and API compute costs fixed within the contract, or do they scale with volume?
Agentic Auditability: How are multi-step autonomous agent actions logged, reported, and audited for operational efficiency?
From AI Experimentation to AI Operations
Healthcare is transitioning from AI experimentation to core enterprise operations.
Foundational technology transformations often develop their own governing disciplines.
The expansion of cloud computing gave rise to FinOps.
Enterprise cybersecurity led to dedicated CISO and security operations structures.
The growth of data created specialized health informatics and analytics functions.
AI adoption across Revenue Cycle could follow a similar trajectory. Managing AI consumption may become a standard operational capability starting as an expanded responsibility within existing teams and potentially maturing into dedicated organizational roles.
So What Will the RCM CommUnity Call This Person?
Technical usage has limited meaning in healthcare unless it can be connected to operational outcomes.
That is why we believe AI Utilization Analyst may be the most fitting and scalable title.
There is also a familiar parallel within healthcare.
Much like Utilization Review (UR) developed to evaluate the appropriate use of healthcare services and resources, an AI Utilization Analyst could help organizations understand how computational resources are being deployed across Revenue Cycle workflows.
Other possible titles include:
AI FinOps Analyst — focused on cloud and LLM cost allocation
Revenue Cycle AI Product Manager — focused on AI applications and workflow integration
RCM AI Value Architect — focused on business value and enterprise strategy
AI Operations & Performance Manager — focused on AI and agent performance
But perhaps the RCM CommUnity will come up with something better.
How is your health system tracking AI consumption? Who currently owns this responsibility? And what should we call the person who will connect AI utilization with Revenue Cycle performance?
AI is changing Revenue Cycle operations and it may also be creating a responsibility that didn't exist before.