At Spare Connect Live, the Pinellas Suncoast Transit Authority (PSTA) shared what it learned after eight months of running AI Voice in production.
Why PSTA Turned to AI Voice
PSTA was facing rising call volumes, longer wait times, and riders who expected faster support. The team turned to AI Voice to handle routine tasks so staff could spend their time on the calls that need a human.
Before launching, they asked four questions:
Could a voice agent reliably complete a full reservation end-to-end?
Status checks, modifications, cancellations — the everyday asks.
Free up reservationists to focus on the more complex calls.
Plug into the platforms already running the operation.
A Phased Rollout Built on the Right Foundations
PSTA rolled it out in phases. Access services came first, then MOD services, then AI Chat. Getting it right took more than the technology itself. The team had to sort out call routing, authentication, testing, and escalation before riders ever picked up the phone.
- Phase 01
Access services
Paratransit calls first — highest-touch, highest-value.
- Phase 02
MOD services
Microtransit added to the same voice layer.
- Phase 03
AI Chat
Extending the same logic to chat channels.
The foundations that had to come first
- Call routing. Deciding what reaches AI Voice and what gets escalated.
- Authentication. Confirming riders securely without adding friction.
- Testing. Running real scenarios before any rider ever called in.
- Escalation. Clean handoffs from AI to a human when complexity warranted it.
What AI Voice Handles Today
AI Voice now handles common rider requests like trip booking, status checks, modifications, cancellations, and transfers to customer service. The live demo showed a rider completing a full booking through a natural conversation, with saved profile and location details making repeat trips faster.
The Results After Eight Months
After eight months, PSTA reported the following:
Calls handled
Rider actions completed
Calls supported per month
Rider outcomes delivered daily
Change in key contact-center metrics with AI Voice in production
Covering the overnight gap
The agency also noted that AI Voice provided valuable overnight support by covering the overnight gap. It answered roughly 2,000 calls between 7 PM and 7 AM over a two-month stretch when reservation staff were off the clock.
Staff hours vs. the overnight window AI Voice now covers
~2,000 overnight calls answered over a two-month stretch.
Lessons From Production
AI works best alongside human teams. About half of all calls are now fully resolved by AI Voice, while the more complex ones still go to staff. The split is roughly 50/50, and that balance is the point — the system handles the routine volume so staff have the time and focus for the conversations that actually need a person.
What the team had to tune after launch
- Recurring trips. Managing repeat bookings without creating duplicate or conflicting reservations.
- Local language and pronunciation. Getting place names, street names, and rider names right.
- Post-launch tuning. Treating launch as the start, not the finish — the system kept getting better with use.
The takeaway
Adopting AI is an operational change, not just a tech install. PSTA's advice for agencies weighing AI Voice was simple:
Pick one service, get it right, then expand.
Operations, IT, and customer service together from day one.
Track hold times, abandonment, sentiment — and tune from there.