Transfer a Call to an AI Voice Agent From Your IVR
Learn how to route inbound calls to an AI voice agent (Groq + ElevenLabs) straight from your IVR, with a clean fallback to a human agent.
Omnifox's AI voice agent answers calls by actually talking in real time: it understands what the caller says, replies in a natural voice, and handles routine questions without a human picking up. The magic happens when you wire it into your IVR with the Transfer to AI voice agent node (powered by Groq + ElevenLabs). This guide walks the full flow node by node, when to use it, and how to pair it with a human handoff.
Why an AI voice agent inside the IVR
A classic IVR only routes by keypress: "press 1 for sales." An AI voice agent goes further because it holds a conversation. Use it to answer hours, order status, FAQs, qualify the caller, or capture details, all by voice and available 24/7. You build it in the IVR editor (under Channels → Calls), which is its own editor, separate from the chat workflow builder.
The flow node by node
Here is the path we will build:
- Entry node (call starts)
- Play audio → welcome greeting (multi-language TTS, barge-in on)
- Wait for DTMF digit → main menu
- Digit 1 → Transfer to AI voice agent
- Digit 2 → Transfer to team (human sales agents)
- No input / invalid digit → Play audio retry and loop back to the menu
Step 1 · Greeting with Play audio
The first node after entry is Play audio. Type the greeting text and Omnifox turns it into speech on publish (it pre-generates the TTS audio). Turn on barge-in so the caller can interrupt the greeting if they already know what they need. Example: "Hi, welcome to Omnifox. To talk to our smart assistant press 1; for sales press 2."
Step 2 · Menu with Wait for DTMF digit
Connect the greeting to a Wait for DTMF digit node. Each digit gets its own edge: 1 goes to the AI, 2 goes to the human team. Set a sensible timeout (around 5 seconds) and a "no input" edge that plays a retry prompt and returns to the same menu.
Step 3 · The star node: Transfer to AI voice agent
From the digit-1 edge, connect Transfer to AI voice agent. In its settings you pick which AI agent answers (the same one you define in the agents section) and pass it context. From there the AI takes over the audio: it listens through speech recognition, generates the reply, and speaks it back in a natural voice. The caller converses as if talking to a person.
Handy trick: place a Find contact node before this one using {{$call.from}} so the AI greets the caller by name and knows who it is talking to. If it finds the contact, the conversation starts personalized.
Step 4 · Human fallback
Never trap the caller with the AI. Configure the flow so that if the caller asks for a human, it exits to Transfer to team (round_robin or first_available) or Transfer to agent over SIP to a specific extension. The AI solves the simple stuff and escalates the complex without friction.
Pair it with Detect language
If you serve multiple languages, drop a Detect language node right after entry, before the greeting. It detects by phone number or by text and branches: each language leads to its own Play audio and its own Transfer to AI voice agent configured in that language. One IVR serves English, Spanish, and Portuguese without duplicating all the logic.
Common mistakes
- Not testing a real call. The editor's lint won't catch audio issues; place an actual call after publishing.
- Forgetting the human fallback. With no exit to Transfer to team, a frustrated caller has no escape.
- Menu too long. With barge-in on the greeting, keep options short; callers cut in as soon as they hear theirs.
- Not ending with Hang up. When the AI conversation finishes, route the flow to a Hang up node or a Voicemail for after-hours cases.
Useful variants
- Hours: add an Hours/Calendar node before the menu. In hours you offer AI + human; after hours, AI only or straight to Voicemail.
- Repeat callers: with Find contact and a Condition, send VIP clients straight to a human and everyone else to the AI.
- Analytics: insert an Analytics event node on the digit-1 edge to measure how many calls the AI resolves.
In Omnifox the IVR editor is visual: you drag Transfer to AI voice agent, wire it to the menu, and publishing generates the audio. The AI uses Groq to understand and ElevenLabs to speak, so it sounds natural in every language.
Want your IVR to answer with AI voice and escalate to humans only when needed? Head to Omnifox and build your first voice flow today.
Comentarios (0)
Todavía no hay comentarios. Sé el primero en compartir tu opinión.
Dejá un comentario
Tu email nunca se publica. Los comentarios se moderan antes de aparecer.