An ambient AI scribe listens to the visit and produces the chart note, so clinicians stop doing two hours of documentation after a nine-hour day.
Documentation is the largest uncompensated time cost in a small practice. Surveys across independent clinics consistently put charting at one to two hours per clinician per day, most of it after the last patient leaves — the "pajama time" that drives the burnout numbers. The work itself is highly structured: a conversation happens, and a note in a known template has to come out the other end. Ambient AI scribes (Freed, and the AI capabilities inside platforms like Weave for the front-office half) record the visit, generate a structured note, push it into the EHR, and suggest ICD-10, CPT, and E/M codes. This is one of the few AI workflows where the ROI arithmetic is uncontroversial: you are converting unpaid after-hours labor into either capacity or time back, and the tool costs less per month than one patient visit.
The scribe has to write somewhere. Browser-based EHRs are the easy case — the scribe pushes the note directly. If you run a legacy desktop system, verify copy-paste-into-template is acceptable to your clinicians before you buy, because that is the fallback.
Two things, and do not skip either. Sign a BAA with the vendor — Freed and comparable tools will provide one, and without it you have a HIPAA problem, not a workflow. Then add a one-line patient notice to your intake form and a verbal disclosure at the start of the visit. Vendors that do not retain recordings make this conversation much shorter.
Do not roll out practice-wide. Choose your highest-volume, most-templated visit type — hygiene recall, follow-up, med check — and one clinician who is willing to give feedback. Template consistency is what the AI learns from, so a repetitive visit type produces the fastest quality gains.
The first week of notes will be 80% right and formatted slightly wrong. Every vendor lets you customize output structure. Spend the two weeks correcting format rather than tolerating it — you are training the template once and using it thousands of times.
The coding assistant surfaces ICD-10, CPT, and E/M candidates. A credentialed human accepts or rejects each one. This is both the vendor's framing and the only defensible way to run it — an AI-suggested code that nobody reviewed is an audit finding waiting to happen.
Track one number: minutes between last patient out and last note signed. That is what you are buying. Practices that see this land typically report the after-hours block collapsing within a month.
Expand only after the pilot clinician says they would be annoyed if you took it away. Each additional clinician gets their own template pass — documentation style is personal and a shared template produces shared friction.
Use these templates as-is or customize for your business.
INTAKE FORM ADDITION: To improve the accuracy of your medical record, we use an AI documentation assistant that helps your clinician write visit notes. Audio is processed securely under a HIPAA business associate agreement and is not retained. You may decline at any time without affecting your care. VERBAL DISCLOSURE (start of visit): "Before we start — I use an AI assistant to help me write up my notes so I can focus on you instead of the screen. It's HIPAA-compliant and nothing is stored. Is that okay with you?"
1. Will you sign a BAA? (If no, stop here.) 2. Are patient audio recordings retained? For how long? Can retention be set to zero? 3. Is our data used to train your models? Can we opt out in writing? 4. Where is data processed and stored geographically? 5. Do you hold SOC 2 Type II? Send the report, not the badge. 6. What happens to our notes if we cancel? 7. Is there an audit log showing who accessed which note and when?
Track every correction so you fix the template once instead of every note:
Date | Visit type | What the AI got wrong | Template change made
-----|-----------|----------------------|--------------------
| | e.g. put allergies in HPI instead of its own section | Added explicit Allergies heading to template
After 10 entries, review for repeats. Anything appearing 3+ times is a template problem, not an AI problem.Get a new AI workflow every week. Prompts, tool stacks, and ROI math included.
Single agent with function-calling: one LLM with a defined toolbox (CRM, calendar, knowledge base) decides which tool to invoke at each turn. Easiest to debug; appropriate for most well-scoped business workflows.
Learn the agentic glossary →Where this workflow tends to break in production — and what to put in place before you ship it.
Note hallucinates a finding the clinician did not state
Mitigation: Clinician signs every note before it enters the record. Never enable auto-sign, no matter how good the accuracy looks in month three.
Ambient mic captures a second patient or a hallway conversation
Mitigation: Verify the vendor scopes recording to the active session; keep the app closed between patients rather than running continuously.
Coding suggestions drift toward upcoding
Mitigation: Audit a 10-note sample monthly against the actual documentation. Track accepted-vs-suggested code distribution over time.
Vendor changes data retention terms after signing
Mitigation: Put zero-retention in the contract, not the settings panel, and set a calendar reminder to re-check terms annually.
Skip this if your visits are short and already fully templated — a two-minute hygiene check does not have two minutes of documentation to save. Skip it if your EHR is a closed desktop system with no integration path and your clinicians will not accept copy-paste. Skip it if you cannot get a BAA from the vendor; there is no version of this workflow that is worth an unmanaged PHI exposure. And do not buy it to replace a human scribe you already trust — the economics are real but so is the quality gap on complex visits.
A phased approach to get this workflow running and delivering ROI.
Days 1–30
Foundation
Days 31–60
Optimization
Days 61–90
Scale
Four vendors, four incompatible pricing models, and one arithmetic trap that doubles your bill without anyone telling you.
The general-purpose voice agent and the trades-specific one cost about the same. One of them books more jobs. The difference is not model quality — it is everything around the model.
Three AI receptionists targeting the same SMB market but built for different niches. Here is an honest comparison of Goodcall, Rosie, and Smith.ai based on production deployments.
One practical AI workflow per week. No fluff.
Get the full guide with step-by-step setup, workflow templates, and copy-paste assets.