Why we built Zorva (and not a billing service)

Most AHCIP billing software falls into one of two categories. The first is a managed submission service — a human or vendor who takes your claims, scrubs them, and pushes them to Alberta Health. The second is an EMR billing module — a checkbox on the encounter screen that drops a fee code on the claim.

Neither of them reads the clinical note against the AHCIP fee schedule. Neither of them asks: given what the doctor actually documented, is the code on the claim the code that should be on the claim?

We built one that does.

The problem we kept seeing

In 2025 we ran shadow audits on 19 encounters pulled from a working Alberta family-medicine practice. Every claim was a real AHCIP submission with real fee codes and a real clinical note. We asked an LLM auditor: is this claim correctly billed?

Across those 19 encounters, the auditor found 23 findings the biller had missed. Eleven of those findings translated to missed revenue — underbilled modifiers, missed add-on codes, shadow-billed services where the work was documented but the code wasn't on the claim. The remaining 12 were denial-risk signals — the kind of pattern that gets a claim kicked back on first submission and burns biller hours on appeal.

None of those 23 findings were caught by the EMR. None of them were caught by the submission service. The doctor had documented the work; the biller had submitted the claim; the revenue had quietly leaked out the door.

Why a decision-support tool, not a submission service

We could have built a managed billing service. The economics are attractive: clinics outsource, we capture a percentage of recovered revenue, and the business compounds with every new clinic signed. But that model has a fundamental conflict of interest — the more denials the clinic sees, the more appeals we bill for. We make more money when the clinic is worse off.

A decision-support tool inverts the economics. We charge a flat monthly fee. Your recovery rate is yours. We get paid the same amount whether Zorva finds 100 errors or 1,000 — so the incentive to under-audit is gone.

Why we don't proxy the LLM call

Other AHCIP tools route your claim data through the vendor's OpenAI key. That means the vendor's account sees your patient data, your provider numbers, and your billing patterns. We think that's a non-starter for healthcare.

Zorva is LLM-agnostic. You bring your own API key — OpenAI, Anthropic, Gemini, or any Ollama-compatible local endpoint. The auditor runs against your provider, in your tenant, and your claims never touch our LLM account.

What's next

We're running a 60-day no-cost pilot with two Alberta family-medicine clinics. After the pilot, the clinic either moves to a paid tier or the data is purged. The full pilot terms are at /pilot.

If you want to see what Zorva catches on your own claims, send 100 de-identified claims. We'll run the auditor and send back a 1-page finding-by-finding report. No fee, no follow-up unless the report is useful.