← Back to Cheryl
AI Pharmacist Workforce · Prior Authorization

How cheryl turns a chart
into an approvable PA letter

The end-to-end workflow — every node, in plain language. Data flows top → bottom; the learning loop feeds back up.

Intake Agent Rules Authoritative data Case law / IRO Learning loop
Live running in production
Step 1 · Intake

Unified Intake

Live

However the rep gets the chart, cheryl turns it into one block of processable text. One entry point, four sources — so every step below is written once, not per-source.

4 sourcesall merge into one intake
📋 Paste handoff

Drop in the forwarded chart text.

📖 Read the screen

One click reads any on-screen page — EHR or payer portal.

📄 Upload PDF

Read to text on-device; nothing leaves the machine.

🔎 Photo / scan

Image-only PDFs go through AI vision OCR.

all four converge on a single, encrypted, token-authenticated intake

Step 2 · Agent · GPT-4o

The Orchestrator

Live

An AI brain runs the case like a senior pharmacist: it decides what to do next — gather a missing fact, or write the letter. No keyword routing; it genuinely understands the case, then calls a standardized, auditable tool.

Infers the setup

Which drug, which payer, which scenario — new, renewal, or appeal.

Asks back

Missing facts → it questions the rep, never fabricates.

Acts via tools only

Every action is a controlled, logged tool call.

Step 3 · Rules + Agent

Hybrid Extraction

Live

Two engines run in parallel to pull the facts out of the chart — diagnoses, BMI, A1C, prior drugs tried, side effects.

two co-equal enginesparallel ∥ then merged
📏 Rules engine

Hundreds of hand-written deterministic rules. Owns the numbers & codes — stable, predictable, explainable.

🧠 LLM understanding

Reads the hard sentences: "no sleep apnea", "family history of diabetes", "tried but stopped".

🔒
Iron rule — nothing is invented. Every extracted fact must appear verbatim in the source text, or it is dropped on the spot. That's what makes every line of the letter traceable.
Step 4 · Rules

Normalization & Reconciliation

Live

The raw facts are cleaned up before use — and the order matters.

a sequenced chain
Fix time-labels

A chronic condition mislabeled "past" → set to "current".

Fix coding

Gestational / drug-induced / pre-diabetes → their own codes, not type-2.

Resolve conflicts

"present" vs "absent" → AI's full-sentence read wins.

Dedup

Rules + AI hit the same fact → merge to one. Runs last: dedup would otherwise collapse a present/absent pair before the conflict is judged.

Step 5 · Authoritative data

Authoritative Lookups

Live

Each fact is checked against public, authoritative data sources — the backbone of trust. cheryl cites; it doesn't recall from memory.

open / authoritative registries
ICD-10

Diagnosis codes — must be specific, no vague placeholders.

RxNorm

NLM's standard drug vocabulary.

SNOMED CT / LOINC

Clinical terms & lab codes (kept in reserve).

Clinical trials + PMID

Real trial evidence, each with a verifiable PubMed ID.

Payer medical policy

Per-drug, per-plan criteria — the published policy of record.

CMS Medicare Bridge

Federal eligibility & appeal criteria from CMS.

NPI / NPPES

National provider registry — verifies the prescriber is real.

IRO precedent corpus

Independent-review rulings — powers node 7.

Step 6 · Rules

Compliance & Approval Prediction

Live

The draft is checked line-by-line against this payer's hard criteria, returning three independent verdicts — not one vague yes/no.

three-axis verdict
Can we draft?

If facts can't support a real letter, refuse — don't fabricate.

Can it be submitted?

What, if anything, the rep still needs to fill in.

Will it be approved?

A probability + "how the payer may push back" + rebuttal prep.

the goal isn't a letter that looks good — it's a letter that gets approved

Step 7 · Case law / IRO

Case-Law & Appeals Engine

Live

On a denial appeal, cheryl does what others don't: it cites precedent. It pulls real Independent Review decisions where a similar denial was overturned, and argues from them — like a lawyer citing case law.

⤵ branches by plan type — and the two legal sources never cross
🏛️ Commercial → State IRO

Denied, then independent medical review. Corpus: California DMHC — 1,625 published decisions.

🦅 Medicare Bridge → Federal chain

Independent review (IRE) → administrative-law judge (ALJ) → Medicare Appeals Council. Its own precedent.

Why this is the moat — and the hard discipline behind it
Citing precedent is what turns a templated appeal into a persuasive one. But the discipline matters as much as the corpus: a commercial IRO precedent can never be cited into a federal Medicare appeal, and vice-versa. The two doctrines don't transfer, and citing the wrong authority can void the letter — so the system structurally forbids cross-lane citation rather than trusting anyone to remember.
Step 8 · Output

Two Outputs, Kept Separate

Live

The result splits into two deliverables that are physically separated in the product.

📨 The payer letter

Clean, compliant, submit-ready — only what a letter should contain.

📊 The rep dashboard

Verdict, approval-%, blockers, strategic risks & checklist — pharmacist-only, never in the letter body.

internal risk notes inside a letter read as defects to a reviewer — so they live apart

Step 9 · Agent

Completeness Dialogue

Live

Missing a key fact (no current BMI, no prior drugs tried)? Before writing, cheryl asks the pharmacist — one item at a time, from that payer's standard checklist — then completes the letter.

↻ feeds back to The Orchestrator (node 2)
Step 10 · Learning loop

The Learning Loop · ML Exit

Every rep action flows back out as a training signal, so the system gets sharper with use.

Payer outcome Live

Approved / denied / partial, logged with drug, payer, scenario — the ground truth.

Code correction Live

The code a pharmacist fixes becomes the strongest correction signal.

Accept / reject / edit Live

Each rep decision emits a PHI-free training signal into the letter-training funnel — generated → submitted → labeled.

Cross-clinic aggregation Live

Nightly roll-up across clinics; ≥3 clinics required (privacy Safe Harbor), then human-reviewed before any rule change.

↻ feeds back to Extraction & Rules (nodes 3–4) so the next letter is sharper

Plain-language glossary · for non-clinical readers

Prior Authorization (PA)
Before covering certain expensive drugs, the insurer requires proof the patient genuinely qualifies. cheryl writes that proof.
GLP-1 drugs
A blockbuster class for weight loss and type-2 diabetes — Wegovy, Zepbound, Mounjaro, Ozempic.
BMI
Body-mass index — a common payer threshold for weight-management drugs.
A1C
A 3-month blood-sugar control marker — central to diabetes approvals.
Step therapy
The insurer requires trying cheaper drugs first before approving a pricier one.
IRO
Independent Review Organization — a neutral medical panel that rules on a denial after internal appeals fail. Like appealing to a neutral court.
Medicare Bridge
A federal Medicare GLP-1 access pathway (opens 2026-07-01) with its own rules and appeal chain.

See it run on a real chart.

🧩 Add to Chrome