In one line
Evaluated conversational AI platforms against a self-built bot to choose the foundation for voice AI. The PoC included an out-of-office bot that picks up calls after office hours, notes the caller's intent and schedules a callback; the callback notes are then routed to the right customer support agent.
Context & problem
All calls after office hours are redirected to the out-of-office bot. The bot introduces itself, tells callers they're calling outside office hours, notes their intent and schedules a callback date and time. All of this, including transcript, summary and caller information, is presented to the subject matter expert. The voice media stream (on Twilio) is redirected to a voice agent running on the GPT Realtime model in Azure AI Foundry.
What I did
- Telephony / voice integration with all evaluated platforms and the self-built bot
- Compared Cognigy, ElevenLabs and other commercial CAI platforms with a self-built bot on Twilio (built by a colleague)
- Built structured scoring matrices and a test-case register
- Worked with Compliance and IT Security to get the platforms and solutions assessed and approved: EU AI Act compliance, GDPR approval, internal risk and vendor assessments
Result / impact
We chose a self-built solution because of compliance and security concerns, existing in-house expertise and implementation speed – and took it through Compliance and IT Security approval (EU AI Act, GDPR, internal risk and vendor assessments). Approved by the Board, Compliance and IT Security since early 2026.
Try it: call the main Swiss Life (Schweiz) AG line after hours.
Details
- Standards & methods: structured evaluation with scoring matrix and test-case register
- Testing: voice latency and response-speed testing, evaluation and scoring