Skip to main content
November 2026 · Vobiz Team AI cold calling puts a voice agent on your outbound prospect list — it dials, pitches, handles objections, qualifies leads, and logs outcomes automatically, at a scale no human sales team can match. A team of 10 human dialers making 50 calls each per day produces 500 dials — an AI dialer on the same infrastructure can run 2,000–10,000 dials in the same window, with consistent messaging and zero fatigue.
Key takeaways
  • AI cold calling = automated outbound dialing + conversational AI agent handling the call
  • The AI uses ASR (speech recognition), LLM (understanding + response), and TTS (voice output)
  • Answering machine detection (AMD) skips unanswered calls and reduces wasted spend
  • In India, TRAI compliance is mandatory: 140-series numbers, DLT registration, NDNC filtering
  • Vobiz provides the telephony layer — connect Retell AI, Vapi, or a custom agent for the AI layer

​
AI Cold Calling vs Traditional Cold Calling

The economic case is clearest for high-volume, structured conversations — intro calls, appointment setting, survey collection, payment reminders — where the script is predictable and the goal is qualification or booking, not complex negotiation.

​
How an AI Dialer Works

Step 1 — Dial: The campaign manager dials numbers from the list at the configured concurrency (e.g., 50 simultaneous dials). The telephony provider (Vobiz) handles the PSTN connection. Step 2 — AMD: Within 1–3 seconds of answer, the system detects whether a human or machine answered. Machine detection uses audio pattern analysis — voicemail greetings have different spectral profiles than live human responses. If machine detected, the system either hangs up or plays a pre-recorded voicemail. Step 3 — AI conversation: If human, the AI voice agent takes over. It plays its intro, listens to the response, and runs the conversation using:
  • ASR (Deepgram, Whisper): Transcribes caller speech in real time
  • LLM (GPT-4o, Gemini): Understands intent, generates response
  • TTS (ElevenLabs, OpenAI TTS): Converts response to natural speech
Step 4 — Qualification and handoff: The agent qualifies the prospect based on predefined criteria. If qualified and interested, it transfers the call to a human sales rep. If not interested, it logs the outcome and ends the call. Step 5 — CRM update: Call outcome, duration, transcript, and any captured data (email, budget, timeline) are pushed to your CRM via webhook.

​
Writing an Effective AI Cold Calling Script

The AI agent’s behaviour is controlled by a system prompt. For cold calling, the prompt structure:
The LLM interpolates the script with live conversation context — it doesn’t just play recordings.

​
AMD: Answering Machine Detection

AMD is one of the most economically important features in outbound AI calling. Without it, the AI agent runs a full conversation with a voicemail greeting, consuming LLM credits and telephony minutes without any value. Vobiz includes AMD via the machineDetection parameter on the Make Call API:
When a machine is detected, Vobiz calls your machineDetectionUrl webhook with AnsweredBy: machine_start. Your server can respond with XML to play a voicemail (<Speak> or <Play>) or simply <Hangup>. See the answering machine detection guide for full implementation details.

​
TRAI Compliance for AI Cold Calling in India

AI cold calling is regulated in India. Non-compliance attracts penalties from TRAI and carriers blocking your numbers. 140-series numbers: All B2C outbound voice calls must originate from a 140-series number. Standard 10-digit mobile numbers cannot be used for commercial calling. Vobiz issues 140-series numbers with self-serve provisioning. DLT registration: The principal entity (your company) and the purpose of the call must be registered on the Distributed Ledger Technology (DLT) platform operated by TRAI-authorized entities. See the DLT registration guide. NDNC/DND filtering: Numbers registered on the National Do Not Call registry cannot be called for promotional purposes. Vobiz provides NDNC filtering on outbound campaigns. Always scrub your contact list against NDNC before dialing. Calling hours: TRAI restricts promotional calls to 9 AM – 9 PM. AI cold calling systems must respect this window — configure your campaign start/end times accordingly. Opt-out handling: If a prospect says “remove me from your list” or “don’t call again”, the AI agent must log this and your system must add the number to your DNC list immediately.

​
Building an AI Cold Calling System on Vobiz

Architecture:
Step 1: Create a campaign on Vobiz In Vobiz Campaign Manager, create a new campaign, upload your contact CSV, select your 140-series caller ID, configure concurrency (start with 10–20 simultaneous dials), and set calling hours (9 AM – 9 PM). Step 2: Configure AMD Enable machine detection in your campaign settings or via the API. Set a voicemail action (drop voicemail audio / hang up / schedule callback). Step 3: Connect an AI agent Configure an answer webhook URL pointing to your AI agent. The agent can be:
  • Retell AI: Create a campaign in Retell, configure the system prompt, and point Vobiz to Retell’s SIP endpoint
  • Vapi: Use Vapi’s outbound calling API triggered by Vobiz webhooks
  • Custom: Build a Python/Node WebSocket server that connects to Vobiz’s real-time audio stream
Step 4: Set up CRM logging Configure a webhook on call completion that sends the call outcome, transcript, and any captured data to your CRM. Vobiz delivers post-call webhooks with full call metadata. Step 5: Monitor and iterate Track connected call rate (answered / total dials), conversation rate (lasted > 30 seconds / connected), and qualification rate (qualified / conversation). Iterate on the script prompt based on where conversations drop off.

​
Cost Model for AI Cold Calling

For comparison, a full-time human SDR in India costs ₹30,000–60,000/month in salary, benefits, and infrastructure. That buys 10,000–60,000 AI connected calls at the same monthly budget.

​
Next Steps