How much does it cost to build an AI agent in India?

Guide

Key facts

Meta charges per delivered message, not per conversation, since 1 July 2025. (Meta)

How much does an AI agent cost in India?

It has two parts: a one-time build fee and a monthly running cost.

One-time build

  • Paid once, at the start
  • Set by scope
  • Fixed quote after a discovery call

Monthly running cost

  • Paid every month
  • Set by volume
  • AI model, voice, WhatsApp, hosting, support

What drives the one-time build cost?

Scope. More channels, systems and rules mean more work to build and test.

  1. Scope

    One task, or many steps and agents?

  2. Integrations

    Each CRM or sheet needs setup.

    Details

    Old systems with no API, or messy data, add work.

  3. Channels and languages

    Voice and Hindi need more work.

  4. Approvals

    Human approval, logs, consent, opt-out.

  5. Testing

    Price or medical answers need more.

What drives build cost. Each step adds work

What are the monthly running costs of an AI agent?

Usage (AI model, voice, WhatsApp) grows with volume. Hosting and support are mostly fixed.

Cost itemHow it is chargedUnit price used in this post
AI model (LLM)Per million tokensClaude Haiku 4.5: US$1 input / US$5 output (Anthropic). Larger models such as Claude Sonnet: US$3 / US$15 (Anthropic pricing)
Voice callsPer minuteAbout US$0.13 for a typical setup; range US$0.07–0.31 (HereFlow AI)
WhatsApp messagesPer delivered messageIndia: marketing ₹0.8631, utility ₹0.115, plus 18% GST. 1,000 free service messages per number per month from 1 Oct 2026, then charged (ChatMitra, Meta)
WhatsApp provider feeMonthly planOften ₹999–9,999 a month (ChatMitra)
HostingMonthlyOften US$20–100 for a small agent (assumption)
Maintenance and supportMonthly planPart of our fixed quote

Examples use US$1 = ₹88 (assumption). Prices change often.

How do channel and model choices change the bill?

A lot. Voice costs far more per contact than WhatsApp, and a larger AI model costs about 3 times more than a small one.

What does a full monthly bill look like?

For a mid-size lead agent, usage can be about ₹25,000 a month, before support.

How can you lower the cost of an AI agent?

Most savings come from scope and channel choice.

What should you not cut?

Do not cut maintenance, testing, human approval or compliance. Under India's DPDP Act 2023, penalties go up to ₹250 crore (PIB).

How do you get a reliable cost estimate?

Bring your monthly volume, channels and systems. Then a vendor can quote, not guess.

  1. List your facts

    Volume, channels, systems, languages, approvals.

  2. Discovery call

    Walk through the task.

  3. Written scope

    What the agent does and does not do.

  4. Fixed quote

    Split into build, monthly usage and support.

  5. Compare vendors

    Same split, like with like.

How to get a fixed quote for an AI agent

Frequently asked questions

How much does it cost to build an AI agent in India?

It depends on scope: channels, integrations, languages, approval steps, data cleanup and testing. A one-channel agent costs less to build than a voice and multi-channel agent. We share a fixed quote after a short discovery call.

What is the monthly cost of running an AI agent?

It depends on usage: AI model tokens, voice minutes, WhatsApp messages and hosting, plus support. In our worked example for 1,500 leads a month, usage costs about ₹24,600. Support is part of our fixed quote.

How much does WhatsApp Business API cost in India?

Meta charges per delivered message. Reported India rates for 2026 are about ₹0.86 per marketing message and about ₹0.115 per utility or authentication message, plus 18% GST and your provider's fee. Check Meta's current rate card.

What does an AI voice agent cost per minute?

Platform rates usually fall between about US$0.07 and US$0.31 per minute. A typical setup is reported at about US$0.13 per minute, including the voice, the AI model and phone line. Telephony changes the final number.

How can I lower the cost of an AI agent?

Start with one channel and one task. Use a small AI model for simple steps and a bigger one only for hard steps. Use WhatsApp before voice calls. Keep a clear scope so the build does not grow.