Weekends · Before the pilot
Saturday mornings were the busiest hour, and nobody was on the phone.
A multi-site allied health practice had a familiar gap. The clinics opened at eight on weekdays. The phone, on Saturdays, did not.
Patients calling in on a weekend, most of them existing patients chasing a new appointment, went to voicemail. Roughly a third never called back. The Service Lead had been recovering some of them through Monday morning callbacks, but the slot had usually filled by then.
The pattern was easy to see in the booking system; harder to see was the quiet revenue walking out of the funnel every weekend.
Pilot · 21 days
A voice agent on the after-hours line, with one bright line: no clinical answers.
We did not staff the weekend. We answered the call that was easy to answer.
The voice agent picks up the after-hours line and tells a booking call from a clinical one. For a booking, it offers the next three available slots from the practice management system, confirms by SMS, and ends the call inside two minutes.
For anything that sounds clinical or unusual (pain levels, post-op questions, anything outside its scope), it takes a callback request and routes it to the duty practitioner’s queue for Monday morning. The agent never tries to give a clinical answer. That line was bright on purpose.
Twilio and Vapi carry the call, ElevenLabs gives the agent its voice, and Anthropic Claude recognises the intent and picks the slot. n8n runs the booking workflow and the SMS confirmation. Nexus, Raava’s orchestrator, coordinates the agent, the hand-off queue and the post-call transcript redaction.
Built in 21 days as a pilot, then nine weeks of iteration with the Service Lead and the duty-roster practitioners.
| Time | Caller | Sorted as | Outcome |
|---|---|---|---|
| 8:12 am | Saturday caller | Sorting | Picked up by the agent |
| 9:05 am | Second caller | Clinical | Callback for Monday |
Offered to the Saturday caller
- Mon 8:45 am
- Mon 10:00 am
- Tue 9:30 am
Monday, 8:00 am
Duty practitioner’s queue
Callback, second caller
A question about pain after surgery
No clinical answer given
A Saturday call comes in. The after-hours line rings with nobody at the desk, and the agent picks up.
The after-hours line rings with nobody at the desk, and the agent picks up.
The agent tells a booking call from a clinical one before it offers anything.
For a booking, the next three open slots from the practice management system.
The caller gets an SMS confirmation, and the call ends inside two minutes.
Pain, post-op questions or anything out of scope become a callback request in the duty practitioner’s queue.
Voice Twilio + VapiVoice synth ElevenLabsModel Anthropic ClaudeGlue n8nCoordination Nexus, Raava’s orchestratorPrivacy Post-call transcript redaction
Iteration · Nine weeks
Deciding when the agent steps out of the conversation.
The nine weeks after the pilot went on a single line: when does the agent step out of the conversation? We logged every call where the patient drifted from booking into clinical territory and watched the agent’s response in review.
Some were obvious: the agent should hand off. Some were tone calls (the patient was venting, not asking), and we tuned a hand-off phrase the practitioners liked the sound of when they heard it on Monday.
The agent never needed to be smarter at clinical questions; it needed to be quicker at recognising it was not the right voice for the call. Once the hand-off felt warm, the Service Lead stopped reviewing the transcripts daily and started reviewing them weekly.
~6%
settled hand-off rate after the nine weeks of iteration
- Window
- Nine weeks of iteration
Approximate, as the page gives it, and the rate it settled from is not on the record.
“The agent is not pretending to be a clinician. It is just picking up the phone on a Saturday morning. That is the bit our patients were missing.”
Post-pilot · 90-day rolling window
The Saturday roster felt it before the report ran.
Saturday mornings went to voicemail before the pilot. No one was on the phone. The agent now picks up most of the after-hours queue without escalating to the duty practitioner.
Numbers verified by the Service Lead against the practice-management booking exports. Anonymised under our standard case-study disclosure: vertical and outcome are real; practice name, locations, and PMS-specific identifiers are not.
92%
of after-hours calls answered by the agent without escalating to the duty practitioner
- Baseline
- 0%: before the pilot, after-hours calls went to voicemail
- Window
- A 90-day rolling window, post-pilot
- Checked by
- The Service Lead, against the practice-management booking exports
+28%
callback recovery on weekend bookings
- Baseline
- The prior quarter’s Monday-morning catch-up rate
- Checked by
- The Service Lead, against the practice-management booking exports
The period the callbacks were counted over is not on the record.
110 s
mean hand-off time to the duty practitioner’s queue, when a call moved outside the agent’s scope
- Checked by
- The Service Lead, against the practice-management booking exports
No hand-off time from before the pilot to set it against.
Curious whether your service has the same after-hours gap?
A free 45-minute audit. We look at your call volume, your booking system, and where the after-hours funnel is leaking. You get a written follow-up within 48 hours, whether we’d be a fit or not.
