If you can’t open your inbox or attend a conference without being told about artificial intelligence revolutionizing healthcare, you’re not alone.
Every startup now claims to be “AI-powered,” every legacy vendor has an “AI suite,” and every deck has a slide about machine learning transforming the industry.
Let’s be honest: some of it is real, much is inflated and a fair portion is pure nonsense. The problem isn’t AI’s power, it’s that most of what’s being marketed is still in pilot, not scalable nor economically viable for the real world of healthcare operations.
So let’s separate what’s here today, what’s coming soon and what’s still the stuff of pitch decks and press releases.
What’s Real Today
AI is fundamentally changing how we process and summarize qualitative and quantitative information. In a health system, Generative AI copilots, when integrated responsibly into electronic health records (EHRs) and workflow platforms, are empowering staff to be more productive and reduce administrative time through natural language models that draft appeal letters, clinical summaries and financial correspondence. Predictive models are becoming better at identifying denials before they occur, and computer vision is reliable for document classification.
The catch? Technology tools work best when problems are well-defined, the data is structured and there’s a clear feedback loop. They’re task automation and reasoning assistants, not omniscient “revenue cycle robots.”
What’s Being Piloted
Agentic AI, systems that reason across multiple steps and act semi-independently, are being tested to orchestrate hospital workflows, make routing decisions and self-correct claims. These are promising but still fragile, working beautifully in controlled pilots, then collapse when faced with payer rule changes, missing documentation or hospital-specific quirks.
Large-scale predictive and reasoning systems across the entire revenue cycle are also being tested, with few, if any, have proven scalable results across multiple clients, payers and care settings.
What’s Still Fantasy (for Now)
Vendors claiming “end-to-end AI-driven revenue cycle management” across every payer, state and service line are overselling. Current AI capabilities consist of very good reasoning models that communicate with each other and complete many actions based on model conclusions.
Human in the loop inclusion remains essential — especially navigating complex payer, processes and exceptions of health systems. Even advanced systems require expert intervention to manage exceptions, ensure compliance and maintain operational integrity.
If presented with “self-learning” models that retrain in real time without human oversight? Run. Particularly in healthcare, that’s not innovation; it’s a compliance risk dressed up as tech swagger.
What Health System Leaders Should Be Asking
You don’t need to be a data scientist to evaluate an AI vendor. Just ask disciplined, evidence-based questions. Here are six core ones that separate substance from spin:
1. How big, diverse and clean is your data?
AI learns from data, not PowerPoint. Understand the size and composition of what their training data looks like — how many encounters, payers and claim types. Is the data representative of your environment?
A model trained on one hospital’s experience will not generalize to yours. And if the data is heavily synthetic or curated from perfect conditions, you’ll be buying a lab experiment, not an operational tool.
2. How do you unify data across the revenue cycle?
Healthcare’s dirty secret is that data silos still rule. Eligibility, coding, clinical documentation, payment posting — all different systems, formats, owners.
Vendors who can’t explain how they normalize, link and reason across sources aren’t building intelligence; they’re adding complexity. The best partners demonstrate how their models connect pre-service, mid-cycle and post-service data to identify systemic leakage.
3. How do you safeguard PHI and ensure cybersecurity compliance?
Security protocols should be detailed, including encryption, access controls, endpoint coverage, breach monitoring and compliance with HIPAA and HITRUST. They should proactively be running independent exercises including third-party penetration tests.
4. How long have you been doing this?
The AI hype cycle makes it easy to forget that experience is invaluable. Companies that have survived multiple payer cycles, ICD changes and regulatory shifts have scars that matter.
If an 18-month-old organization claims, “billions in recovered revenue,” dig deeper. Are they extrapolating? Are those real, auditable results or projections based on subsets of easy wins?
5. What is your actual revenue cycle expertise?
PhDs and data scientists are indispensable, but healthcare’s complexity will chew up even the brightest teams without understanding payer behavior, DRG creep or prior authorizations.
Ask who built the product not just who coded it. Are experienced revenue cycle operators involved in feature design and testing? Do they have EHR data integration experience? If not, you’ll end up being a beta client, not their customer.
6. How are your results proven at scale?
This one separates the adults from the interns. It’s easy to get 95% accuracy on 5,000 records in a pilot. It’s another to sustain performance across hundreds of payers, dozens of specialties and millions of claims.
Ask for production metrics about where has the model been deployed in the wild, and the performance deltas before and after. Are results independently validated? Vague answers like “we’re seeing great engagement,” are code for “we don’t have data yet.”
The Right Kind of Skepticism
Successful leaders aren’t buying flashy demos, they’re asking rigorous questions about data quality, workflow integration and economic ROI.
Demand clarity on how models are trained, what guardrails exist, and how performance is measured with system and environmental context — not in a lab. And when a vendor says, “Our model learns as it goes,” your question should be, “How do you keep it from learning the wrong thing?”
Bottom Line
AI’s impact is real and already making the revenue cycle faster, more consistent and less dependent on brute-force labor. But the transformative stuff, the autonomous, cross-payer, self-healing revenue engine isn’t ready for prime time.
Until it is, focus on the vendors who respect the science and the operations; and understand that healthcare is a living system of patients, clinicians, coders and payers.
The smartest AI doesn’t replace human expertise; it amplifies it. The rest is just noise and there’s plenty of that.
Jim Gaffney is the Chief Strategy Officer at Ensemble