Healthcare Technology

AI-Powered Scheduling and Route Optimization

A statewide behavioral health network scheduled thousands of in-home and clinic-based therapy sessions by phone and email. Practitioners sat at 65% utilization while a fifth of booked sessions were missed. We built the platform that matches practitioner to patient, plans the route between visits, and predicts the no-shows before they happen.

Statewide provider network — 5,000+ practitioners, 16,000+ patients, 100,000+ monthly sessions

95%

Practitioner utilization

85%

Better first-attempt matching

100K+

Monthly sessions

No-show rate, from 20%

The challenge

Coordination by phone and email: Scheduling consumed four hours of every scheduler’s day, and none of that effort accumulated into anything reusable.
A fifth of sessions missed: The no-show rate ran at 20%, against a network whose costs are committed the moment a practitioner is dispatched.
Practitioners idle at 65%: Capacity existed but could not be matched to demand, so a third of available clinical time went unused.
Two hours a day in transit: Routes were built by hand, so practitioners crossed paths constantly and averaged two hours of travel between appointments.
A 48-hour lag before confirmation: Families waited two days to learn whether a request had been accepted, with no visibility into practitioner availability in the meantime.
Matching by availability, not fit: Roughly 30% of sessions were assigned to a practitioner who was free rather than one whose skills best fitted the patient.

How we approached it

1Foundation
2Intelligence
3Automation

Nine months, in three phases of three. Each phase depended on the one before it: the models in phase two had nothing to learn from until the skill matrix and rules engine existed.

Foundation

Build the model of the problem before attempting to optimize it.

Practitioner skill matrix covering the whole network
Patient locations and clinical requirements mapped
Scheduling rules engine encoding the constraints that cannot be broken
First-pass optimization algorithms

Intelligence

Replace the rules-only pass with models that learn from outcomes.

ML models for practitioner-patient matching
Route optimization across the day’s appointments
No-show prediction
Real-time availability tracking

Automation

Move the work out to the people doing it.

Parent self-service portal for requests and rescheduling
Automated rescheduling within clinical parameters
Practitioner mobile apps with navigation and check-in
Predictive analytics dashboards

How it was built

Used by

Parent portal
Practitioner mobile app
Admin portal

Matching engine, route optimizer and prediction engine over a shared scheduling model

Connects to

Mapping and routing service
Geospatial database
Analytics dashboards

Three surfaces onto one scheduling core. The optimization runs centrally; the apps read and write the same schedule rather than holding their own.

Matching engine: Scores every available practitioner against a patient on skills, history and constraints, rather than taking the first who is free.
Route optimizer: Plans the sequence of visits with geospatial queries and travel-time buffers, so a day’s appointments form a route rather than a list.
Prediction engine: Flags the sessions likely to be missed early enough to act on, reaching 95% accuracy on no-show prediction.
Real-time availability: An in-memory layer holds current practitioner state, so the schedule reflects the day as it actually is.

Technology stack

Python optimization stackConstraint solverMachine learning modelsGraphQL API layerGeospatial databaseIn-memory cacheCross-platform mobile

What it delivered

Scheduling

Utilization 65% → 95%: Clinical capacity matched to demand instead of sitting idle.
No-shows 20% → 5%: A 75% reduction, through prediction and parent-side rescheduling.
Scheduler time 4 hours → 30 minutes: Per scheduler, per day.
Travel 2 hours → 45 minutes: Average time between appointments.
Same-day scheduling 0% → 40%: A capability the manual process could not offer at all.

Operational

Optimal skills matching 70% → 95%: Patients matched on clinical fit rather than availability.
Confirmation 48 hours → 15 minutes: Families learn the outcome of a request the same hour.
40% less travel distance: Total distance driven across the network.
Double-bookings eliminated: Conflicts resolved at the point of scheduling rather than on the day.

People

Parent satisfaction 68% → 94%: Measured on the network’s own satisfaction scoring.
Practitioner satisfaction 72% → 91%: Against a baseline where routing was a named source of burnout.
35% less turnover: Practitioner retention improved as routes became workable.
50% more capacity: Growth the network could absorb without adding practitioners.

While others deliver consulting, we deliver working solutions.

Production-ready in weeks. Talk directly to the senior architects who will do the work.

Start Your Project See Enterprise Results