Full case study · Client
Unicall · Sistema Unimed Paraná
Júlia in Unicall’s voice service
AI voice agent · Autonomous service on the Unimed Paraná IVR system
Contact center specialized in supplementary health · 850K beneficiaries · 18 cooperatives
- Segment
- Healthcare / Contact Center
- Case study context
- HealthcareIn productionVoice / IVRContact Center
Source and scope
Case study published by Starya. The text, indicators and testimonials below belong to the scope described. They are not independent validation or a guarantee of results in another operation.
Published title
Júlia
Key Results
- AHT reduction per call
- −75%
- Calls fully resolved by AI
- 12%
- Beneficiaries served
- 850K
- Unimed cooperatives integrated
- 18
The Challenge
Unicall's voice channel received a high volume of calls that went directly to human agents. Without an automation layer, beneficiaries waited in queues and had to repeat the context of their issue. The average handling time (AHT) reached 8 minutes per call, with peaks of up to 30 minutes in more complex situations. This directly impacted the beneficiary experience and raised the operational cost per contact.
The Solution
StaryaAI developed Júlia, an AI voice agent with natural speech and resolution capabilities, built on the proprietary NebulaOS architecture. The solution was deployed on the Unimed Paraná IVR system, integrating ElevenLabs for voice synthesis, Google Gemini as the AI engine, and Oracle Cloud as infrastructure.
- Automation of two priority flows: invoice delivery and authorization status inquiry
- Beneficiary authentication by name, ID number, and health card before any transaction
- Transfer to human agent with full interaction context when needed
Integrations
- ElevenLabs (voice synthesis)
- Google Gemini
- Oracle Cloud
- Invoice API
- Authorization API
- Identity Validation API
- IVR / Telephony
Understanding the operation
Júlia resolves in 2 minutes what took 8 with human agents.
The real scalability gain lies in expanding automated flows: each new flow implemented proportionally reduces overflow volume and cost per contact.
The Results
In the first month of operation, 12% of calls were fully handled and resolved by Júlia, with no human intervention. AHT dropped from 8 minutes in the traditional model to 2 minutes in automated service, eliminating queues for covered flows.
Intelligent routing now directs each call to the correct queue from the initial triage, reducing agent rework and eliminating unnecessary transfers between departments. When overflow occurs, the human operator already receives the full interaction context, including beneficiary identification and data collected during the flow.
The project was delivered in six months, with go-live phases interspersed by adjustment and collaborative learning cycles with the client. All calls are recorded and converted into dashboards with categorization, enabling continuous operational observability and identification of improvement opportunities.
Governance and controls
- Overflow with context: agent receives the complete interaction history
- Pre-filtered routing to the correct queue, without cascading transfers
- All calls recorded and categorized in operational dashboards
- Authentication by name, ID, and health card before any transaction
- Scope limits defined by design: complex topics go straight to humans
- Real-time KPI monitoring: AHT, retention, overflow, and abandonment
What they say about the project
In the first days of operation, Júlia already showed impressive results: 12% of all calls were fully handled by the virtual assistant, with an average resolution time of just 2 minutes, while traditional human service takes an average of 8 minutes. These numbers reinforce the potential of the technology and the success of voice integration, but also reveal new opportunities for continuous learning and optimization. The more Júlia interacts, the smarter and more efficient she becomes.
We innovatively delivered the voice-based service center to handle day-to-day demands with agility and kept the human element for the moments that matter most: complex cases that require close, welcoming, and personalized contact.
The core of the solution is our proprietary NebulaOS architecture. We built an orchestration platform that allows us to combine the best voice technologies, like ElevenLabs, and AI, with Google Gemini, on a robust Oracle Cloud infrastructure. This translates into a scalable and resilient operation that simplifies complexity and delivers security and agility as fundamental values to our client and their beneficiaries.
Technology and operational partners
- ElevenLabs
- Oracle
- Yuni Digital
- dbm contact center
- Unicall Solutions
- Unimed do Estado do Paraná
Next steps
The next steps describe plans recorded in the case study; they do not confirm subsequent delivery.
New flows
Expansion to reimbursement, contract verification, and income reports.
Expanded pre-filter
Júlia acting as intelligent triage for cancellation and suggestion topics.
Resolution scalability
Increasing total resolution rate based on accumulated operational data.
What changes in your context?
Use this case study to discuss the process, data, responsibilities and evaluation criteria for your operation.