01 the business
Primary care network
12 practices, 70 clinicians
Portal messages have become a second workload. Clinicians answer refill questions, test preparation queries and booking changes into the evening. Nurses triage the inbox by hand.
02 before
The problem
Patient portal messages have become a second workload for clinicians and nurses. Many are routine, such as refill questions, appointment changes or how to prepare for a test, but each one still needs reading and a careful reply. Inboxes spill into evenings.
- Clinicians answer messages after hours.
- Admin requests land in clinical inboxes.
- Worrying messages can sit behind routine ones.
03 the build
What the AI does
- Step 1: Reads new portal messages and sorts them: administrative, medication, results questions, symptoms or other.
- Step 2: Sends administrative messages, such as booking changes, to the front desk queue.
- Step 3: Drafts a reply for clinical messages using the patient's record and your clinic's approved wording, in plain language.
- Step 4: Highlights any message that mentions worrying symptoms so a clinician sees it first.
- Step 5: Places the draft in the clinician's inbox for review; nothing is sent automatically.
paused: waiting for a person to approve
How it was put together
- Sorted messages by type and sent admin requests to the front desk queue.
- Drafted replies to clinical messages from the record and approved wording.
- Highlighted messages mentioning worrying symptoms so clinicians see them first.
- Placed drafts in clinicians' inboxes; nothing sends automatically.
04 human in the loop
Where a person stays in charge
A clinician or nurse reads, edits and sends every clinical reply. Messages describing new or worsening symptoms are always handled by a clinician, never by a draft alone.
- A clinician or nurse reads, edits and sends every clinical reply.
- New or worsening symptoms are always handled by a clinician.
- Patients are told replies may be drafted with AI help and reviewed by their care team.
05 rollout
From kickoff to live in 8–12 weeks
- Weeks 1–3Agree categories, approved wording and safety rules with clinical leads.
- Weeks 4–7Pilot in two practices, measuring draft use and clinician experience.
- Weeks 8–12Extend across the network if measures hold.
06 after
What changes
Clinicians start from a sensible draft instead of a blank box, and routine admin messages go straight to the right team. Studies so far show less mental load rather than big time savings, so this is about easing burnout.
Assume 6,000 portal messages a month, a quarter of them administrative and moved straight to the front desk.
6,000 × 25% = 1,500 messages a month that no longer reach a clinician's inbox.
What to measure
- Share of drafts used, edited or discarded.
- Clinician task load and burnout scores.
- Time to reply for flagged messages.
- Admin messages routed away from clinicians.
07 evidence
What the research says
In a pilot with 162 clinicians at Stanford, AI drafts were used for 20% of replies; task load scores fell, but time spent on replies did not change.
43.2% of US physicians reported at least one symptom of burnout in 2024, down from 48.2% in 2023 and 53% in 2022.
Observed physicians spent 27% of their office day on direct time with patients and 49.2% on the electronic record and desk work, plus 1 to 2 hours of after-hours work each night.
08 lessons
Mistakes to avoid
- Expecting big time savings. Evidence so far shows lighter mental load more than saved minutes.
- Auto-sending. Every clinical reply needs a person.
- Hiding AI use from patients.
09 questions
Common questions
- Does it save time?
- A Stanford pilot with 162 clinicians found drafts were used for 20% of replies and task load fell, but time spent did not change. The biggest win is often routing admin messages away from clinicians.
- Is it safe for symptom questions?
- Messages about new or worsening symptoms are highlighted and always handled by a clinician, never by a draft alone.