AI Phone Agent
AI that handles phone calls at scale
About Sanis
Sanis ran a marketing campaign that needed to reach people by phone, at a scale manual dialling can't cover. A room of callers working a list one number at a time is slow, costly, and capped by how many people you can put on the phones.
Objectives
- Reach the campaign's contacts by phone at a scale a human team can't dial.
- Place outbound calls automatically and hold a natural conversation.
- Keep the agent on the campaign's message across a real back-and-forth.
- Guardrail it so it stays coherent rather than drifting off-script.
The Challenge
Sanis had people to reach by phone for a marketing push. Done by hand, that's a room of callers working a list one number at a time: slow, costly, and capped by how many people you can put on the phones.
The Solution
An LLM is connected to a telephony layer with carefully engineered prompts, so the agent places a call, speaks naturally, and stays on the campaign's message. The work is in the prompt design and the guardrails, keeping it coherent across a real conversation rather than a rigid phone-tree.
An LLM drives the conversation over a telephony layer that places the call and carries speech both ways, with a retrieval step grounding the agent in the campaign's specifics and prompt-level guardrails keeping it on message across a real back-and-forth. The engineering isn't the model, it's holding a coherent conversation on a live call where every pause is heard.
A campaign is only as big as the calls you can make
Sanis had people to reach by phone for a marketing push. Done by hand, that's a room of callers working a list one number at a time: slow, costly, and capped by how many people you can put on the phones.
A voice that holds the script
An LLM drives the conversation over a telephony layer with carefully engineered prompts, so the agent places a call, speaks naturally, and stays on the campaign's message. The work is in the prompt design and the guardrails, keeping it coherent across a real back-and-forth rather than a rigid phone-tree.
- Outbound calls placed at machine scale
- Natural conversation, kept on the campaign script
- Prompt design & guardrails over a rigid IVR tree
Technologies
Conclusion
The agent ran a phone campaign for Sanis, holding real conversations at a volume no human team could dial. Outreach that would have taken a room of callers ran at machine scale instead.
Have a project in mind? Let's talk