AI charge capture

Closes the 5-15% revenue leak from missed charges that outpatient practices typically have — without changing the provider workflow. Lives at /admin/charge-capture.

How it works

After an encounter is signed off, Pollen8 scans:

The SOAP note text (procedures performed, materials used).
Observation resources (point-of-care labs, vitals if billable).
Procedure resources already coded.
MedicationAdministration (injections / infusions).
Condition list (associated Dx for medical-necessity gating).

The AI proposes additional billable CPT codes the provider didn’t code. Each suggestion ships with:

The exact evidence excerpt from the note text.
Supporting ICD-10 codes for medical-necessity.
A confidence score (high / medium / low).
A rationale string the coder can paste into the audit log.

The queue

/admin/charge-capture is a worklist:

Encounter	Patient	Provider	Suggested CPT	Evidence	Confidence	Status
23048	Jane Doe	Dr. Smith	96365 IV infusion, init	”Patient received IV fluids”	high	pending

Coder reviews → accepts / rejects / skips. Accepted suggestions flow back into the encounter as Procedure resources and feed the next 837 build.

Why it works without auto-bill

The AI never auto-bills. The coder is always in the loop — same as today’s workflow with a chart auditor — but the AI surfaces what to look at instead of making them read every chart top-to-bottom.

Code suggester (at sign-off)

A related but distinct surface — when a provider signs off an encounter, the code suggester proposes E/M level + ICD-10 dx + CPT procedure codes derived from the SOAP body. Each code rationale links back to the supporting SOAP text span (pairs with audio-span citations for the full audio → SOAP → code chain).

This runs during sign-off; charge capture runs after. They catch different misses — the suggester catches under-coded encounters; charge capture catches missed line-item charges.