Meta Business Agent

For many businesses the Business Agent is the shortest path to an AI that answers on WhatsApp around the clock: no model to choose, no retrieval pipeline to build, and Meta's own guardrails and measurement. The trade-off is control. You configure it rather than engineer it, it is limited to eligible verticals and countries, and its cost scales with tokens rather than messages. Knowing exactly what it does — and where a custom AI agent fits better — is now a standard decision for anyone planning WhatsApp support or sales.

1. 1

   Check eligibility

   For a WhatsApp Business Platform number (connected directly or through a partner), the business must be in an authorised country, meet trust and verification requirements and not be in an excluded vertical, and the number must not already run another authorised AI agent. Meta's eligibility check is the authoritative answer.

2. 2

   Give it knowledge

   Add business information, FAQs, website pages and files. Curated, current sources matter more than volume — the agent answers from what it is given.

3. 3

   Set instructions and tone

   Write the priorities, brand voice and boundaries: what to recommend, what never to promise, and which topics always go to a person.

4. 4

   Connect actions and handoff

   Link your own APIs and webhooks for bookings, order status or payments, and define the events that pass control to your app or team inbox with the conversation history.

5. 5

   Test, launch and watch the meter

   Send test conversations, review answers before and after going live, and track tokens per message — Meta bills every agent message by tokens consumed.

• Assuming it is free

  The launch announcement spoke of getting started for free, but since 1 August 2026 Meta has billed every agent message on WhatsApp per token — including inside the 24-hour window and the free entry point window. Budget by tokens per message, not by message count.

• Feeding it everything

  Long, stale or contradictory sources raise both cost and error rates. Give it a curated knowledge base with current policies and remove outdated pages.

• No plan for the handoff

  The agent should not argue a refund dispute. Define handoff triggers and make sure a staffed inbox picks up those chats quickly, with the history attached.

• Discovering ineligibility late

  Businesses in excluded verticals such as finance or health cannot use it, and a number already running another AI agent integration must switch. Check eligibility before planning a launch.

• Start with the ten questions customers ask most and make sure each has a clean answer in the sources.
• Review a sample of agent conversations every week for wrong answers and missed handoffs.
• Compare its per-message token cost with a custom agent's model cost plus the ₹0.145 service charge for your own replies.
• Keep templates for proactive messages: agent messages are non-template replies sent only inside an open 24-hour window, so campaigns still need approved templates.