A voice agent for outbound, including cold, calls helps assess interest, identify needs, and arrange a meeting. It introduces itself as an AI assistant, speaks in short turns, allows interruptions, and ends the contact at the other person's request. We agree on the contact database, permission to call, timing, and number of attempts before launch.
For fast dialogue, we consider OpenAI Realtime; for standalone synthesis, ElevenLabs Flash, Cartesia Sonic, and Yandex SpeechKit. We connect phone calls through a SIP provider, for example with LiveKit or the ElevenAgents platform. We test the choice in your language and on a real phone line.
In the CRM, we save the confirmed call outcome and the next task. We assess not only the number of calls, but also the share of substantive conversations, agreed next steps, errors, and refusals of further contact. We measure the full delay before an audible response separately from synthesis speed.
Comparison of voice services, WhatsApp, and AI assistant architecture.