Policy-Bound Clinical Documentation
AI drafts the clinical note the moment a visit ends — but nothing is final until a clinician signs it, and every edit they make becomes the audit trail.
The challenge
220 clinicians across 14 clinics were typing or dictating every note by hand after each visit. That workflow carried costs beyond the clinician's calendar:
- 9,800 visits a week, each requiring a manually written or dictated note
- Documentation routinely finished after hours, extending the clinical day
- Note structure and completeness varying clinic to clinic with no shared template
- Some notes not finalized for days, creating compliance and continuity-of-care risk
- No governance model in place for how AI assistance, if introduced, would be audited
At 11.2 minutes per visit across 9,800 weekly visits, documentation alone consumed roughly 1,829 clinician-hours a week network-wide — time spent charting rather than seeing the next patient.
How it works
A draft that never signs itself
The system was built around one non-negotiable policy: AI produces a draft, and only a clinician's signature makes it a record.
- 01
Ambient audio captures the visit conversation with patient consent, alongside structured EHR fields — vitals, orders, medications
- 02
The AI drafts a note in a specialty-specific template immediately after the visit ends
- 03
The draft is explicitly policy-bound to never auto-submit — a clinician must review and sign before it enters the chart
- 04
Every edit the clinician makes to the draft is captured at the field level, not just as a final accept or reject
- 05
That edit log — not a separate compliance report — is the primary key of the audit system
- 06
Aggregated override patterns by specialty feed quarterly retraining of the drafting model
What we built
Key capabilities
Consent-based ambient capture
Visit audio and structured EHR data merge into a draft automatically, with patient consent captured at the point of visit.
A draft-never-final policy, enforced
No AI-drafted note reaches the chart without a clinician's signature — the system has no path that bypasses review.
Overrides as the audit backbone
Every clinician edit is logged at the field level and treated as the record's primary key, not a fallback metric to explain away.
Specialty templates that keep improving
Override patterns per specialty retrain the drafting model quarterly, so templates converge toward how each specialty actually documents.
Before vs after
What changed at the point of care
- Documentation time per visit
- 11.2 min → 3.4 min
- Clinician-hours on documentation weekly
- 1,829 → 555
- Notes finalized same day
- ~68% → 97%
- Clinician override rate tracked as audit key
- Not tracked → 61%
Business impact
What it changed
1,274 clinician-hours returned weekly
1,829 hours of documentation time fell to 555 once drafting moved to the AI — hours clinicians get back for patients or for leaving on time.
Override is the audit key, not an edge case
The 61% edit rate isn't a defect to be driven to zero — it's the record of exactly where and why a clinician's judgment overrode the draft.
Accountability stayed with the clinician
Because nothing enters the chart unsigned, every note in the system still traces to the clinician who stands behind it.
Technology stack
“The system was never built to replace the clinician's judgment — it was built to make every place that judgment overrides the machine fully traceable.”
Keep reading