Two years ago our clinicians at RSLNT Wellness were averaging 90 minutes of charting a night. Some of them were charting on Sunday afternoons. The chart was the second job; the patient was the first.
Today we're under 20 minutes a night across the whole team. Most clinicians sign their last note before they leave the room. This post is what changed — and what we tried that didn't work.
What didn't work
We tried four things first. None of them moved the number much.
- Better templates. We rebuilt our DAP template three times. Saved a few minutes per note. Didn't move the needle.
- Pre-charting. We blocked time at the start of the day to prep. Worked for 2 weeks. Got eaten by patient calls.
- Voice-to-text dictation. Dragon and Apple dictation both bombed on clinical language. Fixing the transcription took as long as typing.
- Outsourcing scribes. Real human scribes worked but were expensive and added a HIPAA risk surface we didn't love.
What actually worked
Three changes, in this order, did 90% of the work.
1. The template hard-anchored to the appointment type.
Most EHRs make you pick the template after you start the note. We made the appointment type select the template. New intake → intake template, follow-up → follow-up template, med-management → med-management template. Zero clicks. The clinician opens the chart and the right form is already loaded.
2. AI drafts the note from the session — in our template.
We trained an AI scribe on our actual templates and our actual sessions (BAA-covered, never used to train upstream models). It listens (opt-in), drafts the note in our template, suggests CPT and ICD codes, and links to the treatment plan. The clinician reads, fixes the parts that matter, signs. The 'parts that matter' are usually 2 to 3 lines per note.
3. The 'sign before you leave' rule.
Once the AI draft made it possible to sign in 5 minutes, we made it the default expectation. Sign before the next patient walks in. If you can't, the chart goes in a 30-min after-lunch block — never after 5pm. The rule made the tool sticky. Without the rule, even with the AI, people slipped back.
What we measured
We tracked three numbers weekly for the first 90 days.
- Median time-to-sign per note (target: under 6 minutes)
- Notes signed after 7pm (target: under 5%)
- Clinician-reported 'chart-related stress' on a 0–10 scale (target: under 3)
By week 12, median time-to-sign was 4.2 minutes, after-7pm notes were 3% (and trending down), and stress score was 2.1 (down from 7.4 at baseline). The team renegotiated the boundary themselves — they wanted the 'sign before next patient' rule kept even when we offered to relax it.
What we'd do differently
Two regrets, both about communication.
First — we didn't introduce the AI scribe loud enough. Some clinicians thought we were monitoring them. We weren't (the AI assist is opt-in per session and logs separately from the note), but we should have spent a full team meeting on what the audit trail shows and doesn't show before turning it on. Trust the team — explain the wiring.
Second — we should have invited patients into the conversation earlier. We now disclose the AI assist in the intake (opt-in, can be turned off any session, and they can ask the clinician to chart by hand). Almost nobody opts out. But the option matters.
What this looks like for your practice
If you're charting after dinner today, here's the smallest possible test you can run this week:
- Pick one clinician + one appointment type.
- Hard-anchor the right template to that appointment type — no template-picker step.
- Turn on AI drafting for that appointment type only.
- Add the 'sign before next patient' rule for that one appointment type.
- Measure time-to-sign daily for two weeks.
If it works — and in our experience it works for ~85% of clinicians who try it — roll it out one appointment type at a time. Don't try to convert everything at once. The behavior change matters more than the tool change.
We rebuilt our practice around this. The clinical work didn't change. The work we used to do at 9pm did.