PickupWindow
The timing model behind every placed call
A per-doctor ML model that learns best-pickup days and hours from historical call outcomes — and refuses to schedule a call outside that window.
- ± 15 min Median accuracy after one year of data
- Per doctor Learning — not a global office-hours guess
- Quiet Hours and holidays honored by default
PickupWindow is the quiet reason MedReach sounds respectful on the phone. Every doctor has a rhythm — rounds at 8 am, clinic at 10, ward work after lunch, reading at 7 pm. A call at 8 am will never work; a call at 7:30 pm might work for one doctor and be intrusive for another. PickupWindow learns that rhythm, per doctor, across campaigns.
The model reads call outcomes: picked up, listened-and-declined, ignored, voicemail, busy. Over a few waves, a per-doctor distribution emerges — this doctor picks up between 14:00 and 16:00 on Tuesdays and Thursdays; that doctor picks up only before 9 am on weekdays; a third doctor only reliably answers on Fridays. The cadence planner consults the model before every scheduled call.
New doctors start with a segment prior — a cardiologist at a public hospital has a different default window than a private dermatologist. The prior is overwritten by that doctor's own data as soon as a few calls land. A year into a programme, the platform's guess of a doctor's pickup hour is within about fifteen minutes of their actual behaviour.
PickupWindow also returns the distance to the next window — if a campaign needs to call now, but this doctor's next window is thirty-six hours away, the planner can choose a WhatsApp touch instead. Timing is not just "when to call" — it's also "when to not call".
Everything PickupWindow handles for you
-
Per-doctor distribution
Days and hours learned individually — no global office-hours guess.
-
Segment priors for cold starts
New doctors start with a specialty-based prior, refined as signals arrive.
-
Distance to next window
Planner can choose a non-voice touch when the next good hour is too far away.
-
Outcome-driven learning
Picked-up, busy, voicemail and ignored all feed the model with weighted signals.
-
Tunable quiet hours
Operators can bound the model to business hours, holidays and jurisdictional quiet-hours.
The when of every scheduled call
PickupWindow is read by the cadence planner before every call. It reads outcome data from the recorder and dialer, and enriches DoctorGraph with the learned distribution.
- CadenceConductor consults best-pickup hours before every scheduled call.
- VoiceDialer emits call outcomes that feed back into the model.
- DoctorGraph stores the learned window as part of the doctor profile.
- AIFlow runs the refine-on-signal pipeline.
Wire PickupWindow into your product today
Book a consultation with our founders and we'll walk you through the whole microservice stack — not just this one — live on your domain.