A patient asks, “So how much is this going to cost me?” and no one can give a straight answer.

Not the care team.
Not billing.
Not even the vendor running eligibility checks.

These aren’t disengaged patients. These are patients referred by their doctors… who want care… and still don’t move forward.

And more often than not, the reason isn’t motivation.

It’s uncertainty about cost.

It Didn’t Look Like a Billing Problem at First

We were seeing lower-than-expected conversion rates. Patients were getting referred, expressing interest, and then… stalling.

At first, we assumed it was an engagement issue. Messaging. UX. Timing.

It wasn’t. The biggest barrier wasn’t behavioral.

It was structural.

Behind the scenes, invisible system constraints… billing logic, coding rules, inconsistent service classification, limited visibility into allowed amounts… were shaping whether patients felt comfortable moving forward.

We couldn’t reliably answer a simple question:
“What will this actually cost me?”

And that uncertainty was enough to stop people from enrolling.

The Problem Isn’t the Tools

The first instinct is to blame the tooling.

Fragmented systems. Bad vendors. Incomplete data.

That’s the wrong diagnosis.

We already use the same systems everyone else uses… eligibility checks, billing platforms, clearinghouses. This is a mature ecosystem. There are entire companies built around “real-time, accurate cost estimates.”

On paper, the infrastructure is solved.

But in practice, it breaks exactly where it matters most.

Not because the data is missing.

Because the logic isn’t consistent.

The Real Problem: Payers Don’t Agree on What Care Is

The same service can be categorized completely differently depending on the insurer, the plan, and even the state.

Evaluation & Management (E/M) visits are predictable. They follow a familiar structure. Most plans handle them the same way. Patients get a copay. It’s understandable.

Remote patient monitoring (RPM) is not.

The exact same RPM codes can be classified as specialty care, telehealth, or something else entirely depending on the plan.

And those classifications aren’t cosmetic.

They determine everything…
whether it’s covered,
whether it hits the deductible,
whether it’s coinsurance,
how the claim adjudicates.

Now layer this into the actual patient experience.

Most programs start with an E/M visit, then transition into ongoing RPM.

Clinically, that’s one continuous care journey.

Financially, it’s not.

It’s two completely different billing models stitched together.

So when a patient asks, “What will this cost?”
There isn’t a single answer.

There are multiple possible answers… all dependent on how their payer decides to classify the service.

That’s not a tooling problem.

That’s a systems problem.

Why Cost Estimation Fails (Even When Everyone “Does It Right”)

Even if you try to estimate cost, your inputs are unstable.

Three variables control everything:

• The payer
• The plan
• The state

Change any one of them, and the same service can behave differently.

We’ve seen the same RPM codes… 99453, 99454, 99457, 99458… get covered, denied, or bundled within the same payer network depending on the plan.

Most eligibility tools don’t even query at the CPT level. They query by service category.

So if your system checks “telehealth,” but the payer classifies RPM as “specialty care,” you get back: “not covered.”

Even when it is.

That’s how two vendors can return completely different answers for the same patient.

They’re not wrong.

They’re just operating on different assumptions.

So We Stopped Trying to Be “Right”

Once we understood the problem, our approach changed.

We stopped trying to generate a single “accurate” number.

Because the system itself isn’t consistent enough to support one.

Instead, we designed for how the system actually behaves.

We:

• Split estimates across E/M and RPM instead of forcing one number
• Gave care teams language to explain variability without needing billing expertise
• Built feedback loops using real claim outcomes
• Started tracking patterns across payer, plan, state, and CPT combinations

Not to eliminate uncertainty.

But to make it legible.

Where AI Comes In… And How We Get There

This is exactly the kind of problem AI should solve.

But only if the foundation exists.

Right now, most systems are trying to predict cost without the right underlying data. They rely on eligibility responses and generalized rules… not on how claims actually adjudicate in the real world.

That’s why they break down.

What matters isn’t just access to data… it’s having the right data.

We’re building toward a structured repository of adjudicated claims outcomes.

For each patient, that includes:

• The payer, plan, and state
• How specific CPT codes were processed
• Whether cost-sharing was applied as copay, coinsurance, or deductible
• What the allowed amounts actually looked like

Not what should happen.

What did happen.

Once that dataset exists at scale, AI becomes meaningful.

It can identify patterns across payer-plan combinations.
It can predict how services are likely to adjudicate.
It can surface variability before it impacts the patient.
It can translate complexity into something understandable.

But AI isn’t the starting point.

The data is.

And getting that layer right is what makes everything else possible.

The Part Most People Miss

If patients don’t understand cost, they don’t enroll.

If they don’t enroll, care doesn’t happen.

And if care doesn’t happen, everything upstream… referrals, programs, product… doesn’t matter.

This isn’t a billing edge case.

It’s a system design failure.

And until we design for how payers actually behave… not how we wish they behaved… we’ll keep building solutions that look right on paper and fail in reality.