Automation That Elevates People: Advantum’s Position on Healthcare AI

Advantum Health graphic reading Better tools. Greater capability. Technology and expertise working together, with a glass arch and sphere

The short answer: Our position on AI is straightforward. Automation should remove the work that does not require judgment so people can spend their time on the work that does. We build AI into our own platform and our own teams operate it. Our clients do not run it. That is deliberate. The right technology in the wrong hands still produces the wrong results, and accountability is the one thing that cannot be automated. Advantum Health has spent 25 years learning that revenue cycle outcomes come from people who are persistent, informed and answerable for the result.

Why I keep saying this out loud

Every healthcare organization I talk to is being sold AI right now. Most of the pitches sound the same, and most of them are selling the same promise: fewer people doing this work.

I understand the appeal. Margins are compressed, hiring is hard and administrative cost is real. The industry is moving: more than 80 percent of health systems now prioritize agentic AI for revenue cycle work, and 37 percent already use generative AI there. But I have watched this industry chase labor reduction before, and the pattern is consistent. Cost comes out, and then a year later the denial rate is up, the accounts receivable is older and nobody can explain why, because the people who understood the payer are gone.

So I want to be clear about what we think automation is for. It is not for removing people from the revenue cycle. It is for removing the parts of the revenue cycle that were never worth a person’s attention in the first place.

What we automate, and what we do not

The dividing line we use is judgment. If a task has a right answer that can be verified against a source, it is a candidate for automation. If it requires interpreting a payer’s behavior, weighing a tradeoff or deciding what to assert, a person does it.

Eligibility verification has a right answer. Claim status has a right answer. Assembling documentation for an appeal has a right answer. Ranking a denial queue by recoverable value is arithmetic. All of that should be automated, and at Advantum it is.

What a specific payer will actually accept in an appeal is not arithmetic. Whether a documentation gap is worth escalating to the physician is not arithmetic. Whether a patient balance should go to a payment plan or to financial assistance is not arithmetic. Those are judgment calls, and judgment calls need someone who will be accountable for having made them.

This is what I mean when I say results come from people and technology supports the work. It is not a sentiment. It is an operating decision about where we put the humans.

Why we do not hand the AI to our clients

This is the part that surprises people, so let me explain the reasoning.

Our AI is built into our platform and used by our own staff. Clients see the output. They see the dashboards, the priorities and the results. They do not operate the models, and we do not position the technology as a tool we hand over.

The reason is accountability. A model handed to a client becomes the client’s problem the first time it produces a wrong answer, and the vendor’s role quietly shifts from responsible party to software provider. We are not willing to make that trade. When a claim is wrong, the answer to who is responsible should be a person at Advantum, not a license agreement.

It also produces better technology. Our staff use these tools every day across many clients and many payers. When something does not work, they tell us, and we fix it. A tool deployed into a client’s environment and reviewed at renewal does not get that feedback.

What this means for how we measure

If automation is meant to elevate people rather than replace them, the metrics have to reflect that.

We do not report on claims touched, letters generated or tasks automated. Those numbers go up whether or not anything improved. We report on net collections, denial rate, days in accounts receivable, first-pass resolution and cost to collect, because those are the numbers a chief financial officer can act on.

We also watch a metric that does not appear on most vendor dashboards: what our people are spending their time on. If automation is working, the share of staff hours going to judgment work should be rising. If it is not, we have bought speed without buying capability.

What I would ask a vendor

If you are evaluating AI in your revenue cycle, these are the questions I would want answered before anything else.

  • Who is accountable when it is wrong? Ask for a name and a role, not a support tier.
  • What does it do when it is uncertain? A system that never abstains is a system that will be confidently wrong at scale.
  • What is the baseline? Any improvement claim without a prior-period denominator and a stated payer mix is marketing.
  • What happens to my team? If the answer is vague, the plan is headcount reduction and nobody wants to say so. That may still be the right decision for your organization. You should make it deliberately.
  • Can you show me the work? If you cannot reconstruct what the system did on a specific claim, you cannot defend it to an auditor.

The part that does not change

Revenue cycle management is a persistence business. Payers change policies without notice. Denials arrive in patterns that take experience to recognize. Contracts underperform quietly for years until someone benchmarks them. None of that is solved by processing the same work faster.

Technology makes our people faster, more accurate and better informed. It does not make them unnecessary, and any partner telling you otherwise is describing a product rather than a result.

That is the standard we hold ourselves to across all nine services in the revenue cycle, and it is why our technology is built to support experienced operators rather than to replace them.

If your revenue cycle is not performing, the question worth asking is not whether you have enough AI. It is whether anyone is accountable for the outcome. We are happy to have that conversation.