> ## Documentation Index > Fetch the complete documentation index at: https://www.vobiz.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # What Is AI Phone Answering? How AI Answers Your Business Calls > AI phone answering uses a voice agent to pick up inbound calls, understand what the caller wants, and respond or route — 24/7, with no hold time. How it works, what it costs, and how to set it up. *November 9, 2026 · By [Piyush Sahoo](https://www.linkedin.com/in/piyush-s713/)* **AI phone answering is when a voice agent picks up inbound calls and handles them in real time — listening to what the caller says, understanding the intent, and responding naturally — without a human on the other end. It is the difference between a caller reaching voicemail at 11 pm and reaching an agent that can actually help them.** It is not an IVR. An IVR plays a menu and routes by keypad press — the caller still presses 1, 2, or 3. An AI answering system conducts a conversation: the caller says "I want to reschedule my appointment for Thursday" and the system understands, checks availability, and books it. > [!NOTE] > > **Key takeaways** > > * AI phone answering = inbound call + voice agent (STT + LLM + TTS). The caller speaks naturally; the AI understands and responds. > * It differs from IVR (keypad menus) and voicemail (record-and-retrieve). > * Works 24/7, handles unlimited concurrent calls, never puts a caller on hold. > * Transfers to a human agent when it cannot resolve, with full transcript context. > * For India: works on any inbound number; no special TRAI constraints for inbound-only deployments. ## What is AI phone answering? AI phone answering is the inbound equivalent of AI outbound calling. When a caller dials your number, an AI voice agent answers — in milliseconds, with no queue — and handles the conversation from greeting to resolution or handoff. The AI agent is powered by three components: 1. **Speech-to-text (STT):** converts the caller's voice to text in real time, typically with streaming output so the system starts processing before the caller finishes speaking 2. **Large language model (LLM):** processes the transcript, understands intent, generates a response based on your business context, instructions, and knowledge base 3. **Text-to-speech (TTS):** converts the response to natural-sounding audio and plays it back to the caller The whole loop — caller speaks, AI understands, AI responds — happens in under 1,500 ms for a natural conversation. ## AI phone answering vs. alternatives | | Voicemail | Traditional IVR | AI phone answering | | - | - | - | - | | **Caller experience** | Leave a message, wait for callback | Press 1, 2, 3 | Natural conversation | | **Resolution on call** | No | Partial (routing only) | Yes, for most queries | | **Available 24/7** | Yes | Yes | Yes | | **Concurrent calls** | Unlimited | Unlimited | Unlimited | | **Understands natural speech** | No | No | Yes | | **Can answer questions** | No | No | Yes | | **Can book / take action** | No | No (without integration) | Yes (with integrations) | | **Transfers with context** | N/A | Yes (blind transfer) | Yes (with transcript) | ## Common use cases **Receptionist replacement.** AI answers every call, identifies the caller's need, and either resolves it or routes to the right person. No more calls going to voicemail during lunch or after hours. **Appointment booking and management.** Caller says "I need to book a slot for Friday afternoon." AI checks availability, books, and sends a confirmation — no human needed. **FAQ handling.** Business hours, pricing, location, return policies, order status. AI resolves these without occupying a human agent. **After-hours coverage.** Human team works 9–6; AI handles everything outside those hours. Urgent calls get transferred to an on-call number; everything else gets logged and followed up the next day. **High-volume inbound campaigns.** Promotional or event-driven spikes where call volume exceeds team capacity. AI handles the overflow without queue times. **Lead capture.** AI answers calls from ad campaigns, qualifies the lead with a few questions, and either books a demo or routes to a sales agent in real time. **Multilingual support.** Route Hindi-speaking callers to an AI agent configured with a Hindi-language STT and TTS model, English speakers to a different configuration — all from the same phone number. ## How to set up AI phone answering The setup has three layers: get a phone number, connect it to an AI agent, configure what the agent does. **Step 1: Get an inbound number** For India, you need a DID number (direct inward dialing) from a provider like Vobiz. This can be a local number (e.g., +91 80 XXXX XXXX), a mobile-format number, or a toll-free 1800 number. **Step 2: Point the number at your AI agent** Using Vobiz, you configure the number to call a webhook URL when a call arrives. That webhook returns XML instructions — including a `` element that opens a WebSocket to your AI agent. ```xml When a call arrives on your number theme={null} wss://your-ai-agent.com/inbound-stream ``` **Step 3: Configure the AI agent** If you use a managed platform like Vapi, Retell AI, or ElevenLabs Conversational AI, you configure the agent through their dashboard — system prompt, knowledge base, transfer destinations, voice selection. The platform handles the STT/LLM/TTS pipeline. If you build a custom agent, your WebSocket server receives audio, runs it through your STT, calls your LLM, synthesizes the reply, and sends audio back via `playAudio` events. **Step 4: Handle transfers** When the AI decides to transfer to a human, it uses Vobiz's XML to dial a second number and bridge the calls: ```xml Transfer to a human agent with context theme={null} +919XXXXXXXXXX ``` The transcript of the AI conversation is passed to the human agent via your CRM integration or a webhook. ## Latency and voice quality A natural conversation requires end-to-end turn latency — from the caller finishing speaking to the AI starting its reply — under about 1,500 ms. The main contributors: * **STT processing:** 100–300 ms for streaming transcription * **LLM inference:** 200–500 ms for first token (streaming) * **TTS synthesis:** 100–300 ms for first audio chunk (streaming) * **Network:** 20–80 ms per hop Keep your AI pipeline in the same region as your SIP trunk endpoint. If your Vobiz WebSocket endpoint is in Mumbai and your STT/LLM runs in US-East, you add 150–200 ms of unnecessary round-trip on every exchange. ## Indian language support For India deployments, STT accuracy for Indian English and regional languages varies significantly by provider: | Language | Supported STT providers | | - | - | | English (Indian accent) | Deepgram, Google, AssemblyAI, Whisper | | Hindi | Sarvam AI, Google, Whisper (limited) | | Tamil, Telugu, Marathi, Bengali | Sarvam AI, Google (varying quality) | [Sarvam AI integrates natively with Vobiz](https://www.vobiz.ai/docs/integrations/sarvam.md) and is optimized for Indian languages. For English-primary deployments, Deepgram Nova-2 or Google STT provide reliable accuracy on Indian accents. ## Vobiz for inbound AI answering Vobiz provides: * **Inbound DID numbers** in India and 130+ countries * **Toll-free 1800 numbers** for India * **Bidirectional WebSocket audio streaming** via `` — your AI pipeline gets live audio and can inject synthesized replies * **Call recording** for QA and compliance * **Webhooks** for call events, status, and recordings For a no-code setup with Vapi or Retell AI, see the [Vapi dashboard integration](https://www.vobiz.ai/docs/integrations/vapi-dashboard.md) or [Retell AI dashboard integration](https://www.vobiz.ai/docs/integrations/retellai-dashboard.md). For a custom build, start with the [inbound AI voice agent guide](https://www.vobiz.ai/docs/guides/ai-voice-agent/inbound.md). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.