Why I Built Cheryl

I work two jobs. Most of my week, I run my own clinic. A few days, I see patients in clinical trials at a large academic hospital.

Two worlds. Two systems. One shared frustration: the EHR.

The Small Clinic Problem

When I opened my practice, I discovered that most EHRs are designed for large hospital systems. They want you to click through 15 screens to document a 10-minute visit. The software costs $300+/month, the training takes weeks, and your front desk still has to answer every phone call manually.

Small clinics get the worst deal: enterprise-grade complexity at enterprise-grade prices, without enterprise-grade staff to run it.

So I started building my own tools. A simple system that let me document the way I actually think—in plain language, not someone else's dropdown menus. Scrappy, but it worked.

The Epic Reality

Then I walked into the hospital.

The EHR there was Epic—state-of-the-art, billion-dollar infrastructure.

Every 15-minute patient encounter required 30 minutes of documentation afterward.

"It Was Better When We Just Scribbled"

One day, an older physician said something that stuck with me:

"Honestly? It was faster when we just scribbled on paper. The EHR is great for billing and lawyers. For patients? A step backward."

But I kept thinking: handwritten notes were fast because they were natural. Doctors wrote the way they thought—abbreviations, arrows, sketches in the margins. No dropdown menus. No mandatory fields. Just language, flowing from brain to page.

Then it hit me.

Scribbling Is Just Chatting

What's the difference between scribbling on paper and texting a friend? Nothing. Both are natural language. Both are unstructured. Both are fast.

When I chat with an AI, I don't fill out forms. I just say what I'm thinking. It asks follow-ups. It organizes my messy thoughts into something coherent.

Why couldn't an EHR work the same way?

So I Built One

I started with my own clinic. Instead of clicking through forms, I type to my AI assistant like I'm texting a colleague:

Seconds later: a complete, structured clinical note. No templates. No clicking. Just a conversation.

Then I realized: this doesn't need to replace your EHR. It just needs to sit next to it. A Chrome extension. Open the sidebar, paste your notes, get a chart back, copy it in. Works with any EHR. No integration. No IT.

The result? Documentation went from 30 minutes to 3. I finish notes before patients leave. I look at them instead of my screen. And I go home on time.

Then I Looked at the Money

One day I pulled my billing data. I was coding almost every visit as 99213. But when I actually reviewed my documentation? Half of those visits had enough complexity for 99214. That's $44 more per visit. Five visits a day, 260 days a year—I was leaving $57,000 on the table.

Not because I was lazy. Because coding is exhausting when you're already drowning in charting. You pick the safe code, the fast code, the one you always pick. And you move on.

So I built a billing coach into Cheryl. While you're writing your note, she reads it and says: "You documented 2 chronic conditions and Rx management. This qualifies as 99214, not 99213. That's $44 more." One click to accept. Done.

Then She Started Learning

I thought the billing coach was enough. Then I noticed something weird.

I kept correcting Cheryl on the same thing — "this is a 5-region CMT, use 98941, not 98940." First time, fine. Second time, mildly annoying. By the fourth time, I was ready to throw my laptop.

So I wrote her a rule: three corrections, then you promote it. If the same clinic corrects Cheryl the same way three times, she stops making that mistake. Forever. The fourth visit, she asks about regions automatically. The fifth visit, she just applies the right code.

But here's the part that surprised me: rules decay. If I stop reinforcing a correction for 60 days, it fades. Because billing evolves — NCCI edits change, payers update rules, my own clinic's habits drift. Static software rots. Cheryl stays fresh by forgetting on purpose.

I also realized something bigger: per clinic, not per user. When my front desk corrected Cheryl, my locum doctor got the fix the next day. One brain, one clinic. That alone saved us a week of onboarding every new provider.

Then I Counted the Tabs

I was halfway through my third patient on a Monday when I noticed something: I had six browser tabs open.

My EHR. Availity (eligibility). Office Ally (claim submission). My clearinghouse inbox (ERA). A payer portal I'd opened to check a pre-auth. And Cheryl.

Six tabs. For ONE patient. I was the thing carrying the patient between them — copying a name here, pasting an ID there, re-typing the same diagnosis three times.

So the next thing I’m building into Cheryl is the tab problem. Not just the EHR — the whole trail. When you open Availity, Cheryl will read the eligibility response and carry it to the patient’s note. When you submit in Office Ally, she’ll pull the submission ID. When the ERA arrives, she’ll capture the payment posting automatically — so you finally know which claims got paid, which got denied, and why.

Shipping in Phase 1 alongside the multi-patient queue. Today Cheryl lives in the EHR tab; the portal bridge is the next thing on the build list.

Then I Looked at My Own Data

This is the one I didn’t see coming when I started.

Cheryl silently logs every visit, every code, every correction, every denial — aggregated, no patient names, no diagnoses, nothing identifying. Just: what codes am I using, when do I undercode, which payer denies me most. The capture layer is already running for early members so the baseline builds up. The dashboard that surfaces it is the next thing after the portal bridge.

The pictures I want early members to see, once enough data is in:

“$3,240 recoverable this month.” Undercoded 99213→99214 on 18 visits. Cheryl flagged each one in the moment; you rejected the upgrade. The dashboard will show you when — maybe Friday afternoons when you’re tired. Fatigue with a dollar amount on it.

“Monday afternoons run 2.1× longer than Tuesdays.” Not your front desk — your visit mix. You were scheduling it that way.

“Your 99214 ratio is top 15% for family medicine.” Peer benchmarks without ever shipping patient data out.

Metrics capture is live today. The dashboard UI ships after the portal bridge. Founding members get it first.

None of it is patient data. All of it is your data — how you bill, how you schedule, how you compare to other clinics using Cheryl. HIPAA-safe by design, not by promise.

What EHRs Got Wrong

Electronic health records were designed for compliance, billing, and interoperability. Noble goals. But they separated three things that should never have been separate: charting, billing, and knowing how your clinic actually runs.

By the time a billing coder sees your note, the context is gone. By the time you look at last quarter's numbers, it's too late to do anything about them. By the time your EHR can "talk to" the clearinghouse, you've already copy-pasted the claim by hand.

The future isn't fancier forms or smarter billers.