What is an AI receptionist — and how does it answer your phone?
A plain-language guide to AI voice agents: the technology that picks up a phone call in under a second, holds a natural conversation, and books a real appointment — and how it compares to the IVRs, answering services and call centers it replaces.
The definition
An AI receptionist (also called an AI voice agent or AI front desk) is a conversational AI that answers a business's phone line the way a good human receptionist would: it greets the caller by the business's name, understands free natural speech in the caller's language, answers questions from the business's own knowledge, books appointments into genuinely available slots, takes messages after hours, and transfers to a human when the conversation needs one. It answers every call on the first ring, 24×7, at any call volume.
How it works — the voice pipeline
Under the hood, three AI models run in a tight loop over a real-time telephony connection:
- Speech-to-text (STT) transcribes the caller as they speak. Naadham uses Deepgram's nova-3 multilingual model, with region-specific name boosting so Indian names are transcribed correctly instead of being anglicised.
- A large language model (LLM) is the brain: it reads the transcript, decides what to say, and calls tools — search the clinic's knowledge base, check the real appointment calendar, create a booking, or transfer the call. Naadham runs a function-calling loop on Groq-hosted open models with automatic fallback to Anthropic's Claude, so a provider outage never drops a call.
- Text-to-speech (TTS) speaks the reply in a natural, configurable voice (Naadham uses Cartesia). Voice activity detection (VAD) — Silero in Naadham's stack — handles turn-taking, so the agent knows when the caller finished speaking and never talks over them.
The call itself travels over SIP telephony (from carriers like FreJun in India or Twilio internationally) into a WebRTC media layer (LiveKit), where the agent worker joins the call like a participant. The same loop powers browser calls — which is why you can talk to Naadham's demo agent from this website with just a microphone — and the same "dialogue brain" answers WhatsApp, SMS and web chat, so the AI behaves identically on every channel.
AI receptionist vs IVR
| IVR ("press 1 for…") | AI receptionist | |
|---|---|---|
| Input | Keypad tones, rigid phrases | Free natural speech, any phrasing |
| Coverage | Only pre-built menu paths | Anything in the business's knowledge + live booking |
| Caller experience | Menus, waiting, repeats | Answered in <1s, conversational, one turn per question |
| Booking | Rare, clunky | Books real, conflict-free slots from the actual schedule |
| Records | Menu analytics | Full transcript + automatic quality score per call |
AI receptionist vs call center / answering service
| Call center / answering service | AI receptionist | |
|---|---|---|
| Answer time | Queue-dependent; busy hours = hold | Instant at any concurrency (scales automatically) |
| Hours | Shifts; nights/weekends cost extra | 24×7 included |
| Cost model | Per-seat or per-call retainers | Per connected minute (Naadham: ₹5–9.99/min prepaid) |
| Consistency | Varies by agent and day | Same persona, greeting and accuracy on every call |
| Complex cases | Handled by the agent | Transferred to your human staff, with context |
Why answer speed matters
Missed and abandoned calls are lost business: patients who hit voicemail call the next clinic. An AI receptionist removes the two places calls die — the unanswered ring and the hold queue — because it picks up in under a second and holds unlimited simultaneous conversations. After-hours calls, the ones a front desk never sees, become bookings instead of voicemails.
Security & compliance
- Data: Postgres with row-level security per tenant; secrets encrypted at rest; hashed session tokens and API keys; constant-time auth comparisons.
- Medical safety: the agent is hard-blocked from giving medical advice — every call's QA score carries a compliance flag for it.
- Outbound calling law: a consent registry records who agreed to be called; a public do-not-call list is honored on every dial path; calls happen only in lawful hours (India TRAI/TCCCPR: 10:00–19:00 IST; US TCPA: 08:00–21:00 local); the AI always discloses it is an AI on outbound calls; and clinics complete per-country KYC before any outbound calling. Details: the product guide.
What it costs (India)
Naadham is prepaid, pay-as-you-go, with no setup fee and no per-seat licences: ₹999 → 100 minutes (₹9.99/min), ₹3,999 → 500 (₹8/min), ₹14,999 → 2,500 (₹6/min), ₹49,999 → 10,000 (₹5/min). Every account starts with 15 free trial minutes — enough to configure your agent and hear it on a browser test call before spending anything. Full pricing.
How a clinic deploys one (about 3 minutes to the first conversation)
- Sign up at naadham.ai/login — instant, 15 free minutes.
- Customise with a form: greeting, tone, voice, languages, services, hours, FAQs, transfer number. Saves apply live.
- Take a browser test call — talk to your own configured agent with your mic, no phone number needed.
- Connect a number — usually assigned instantly and auto-wired; you just forward your existing clinic line to it.
- Upload your doctor schedule (.xlsx/.csv) and the AI books real, conflict-free slots.
Healthcare-specific detail, specialty by specialty: AI receptionist for clinics & hospitals. Evaluating platforms? How Naadham compares to Retell, Vapi, Bland and Avoca. Building your own integration? Developer API docs.
Frequently asked questions
What is an AI receptionist?
An AI receptionist is a conversational voice agent that answers a business's phone calls the way a human receptionist would: it greets the caller, understands natural speech, answers questions, books appointments and transfers to a human when needed — 24×7, with no hold time and no menus.
How does an AI voice agent actually work?
Three models run in a loop over a real-time telephony connection (SIP/WebRTC): speech-to-text transcribes the caller, a large language model decides what to say and which tools to use (look up FAQs, book a slot, transfer), and text-to-speech speaks the reply. Voice activity detection handles turn-taking so the agent knows when the caller has finished speaking.
What is the difference between an AI receptionist and an IVR?
An IVR makes callers navigate menus (press 1, press 2) and can only follow pre-built paths. An AI receptionist understands free natural speech, answers in full sentences, handles questions it wasn't explicitly scripted for, and completes real tasks like booking a specific appointment slot.
Can an AI receptionist replace a call center?
For front-desk workloads — answering, FAQs, appointment booking, reminders — yes, and it answers instantly at any call volume. For complex or sensitive conversations it transfers to a human, so most businesses run it as the always-on first layer with humans handling escalations.
How fast does an AI receptionist answer?
Naadham picks up in under a second — there is no ring queue and no hold music. Conversational responses stream back with sub-second turn latency in normal conditions.
How much does an AI receptionist cost in India?
Naadham is prepaid pay-as-you-go: ₹999 for 100 minutes (₹9.99/min) down to ₹5/minute on the largest pack, with 15 free trial minutes and no setup fee. You pay only for connected call minutes.
Is an AI receptionist secure?
Naadham stores data in Postgres with row-level security, encrypts stored secrets at rest, uses hashed session and API-key authentication, and never lets the AI give medical advice (a hard compliance flag). Outbound calling is gated by a consent registry, a do-not-call list, lawful calling windows and per-country KYC.
Does an AI receptionist disclose that it is an AI?
On every outbound call Naadham's agent discloses it is an AI in its first sentence, answers truthfully if asked whether it is human, and immediately honors requests to stop calling.
Hear it yourself — right now
Talk to the live demo agent from this browser, or sign up and talk to your own configured agent in ~3 minutes.
Try the live demo Start free — 15 minutes