A billion business chats a day just got an employee
On June 3, 2026, at its Conversations conference in London, Meta switched its Business Agent from pilot to global availability across WhatsApp, Messenger and Instagram. TechCrunch reports that more than one million businesses were already using it after pilots in India, Mexico and Brazil, and that people now have over a billion active business conversations every day across Meta's three apps. Meta's own announcement describes what the agent does inside those chats: it answers customer questions, recommends products, books appointments, qualifies leads, and hands the conversation to a human when a configurable threshold is crossed. The enterprise platform version connects to hundreds of external systems, Shopify and Zendesk among them.
For the more than 200 million small businesses on WhatsApp, this is not a distant platform story. It is a change in what a customer expects when they message you at 11pm.
What it costs, in plain numbers
Since August 1, 2026 the agent is metered at roughly $2.00 per million tokens. CNBC calculates that a typical customer conversation in the US lands around 4 to 5 cents. Small businesses get access through the WhatsApp Business Premium tier rather than raw token billing. The context matters: Meta's paid messaging business was already running at a $2 billion annual rate as of December, so this is not an experiment Meta will quietly drop. Mark Zuckerberg went further in July, predicting that billions of people will have personal AI agents within five years. Your customers' agents will be messaging your business agent sooner than that sounds.
Where the ready-made agent is genuinely enough
A default agent trained on your catalog and FAQ covers a surprising share of real conversations:
- Repetitive questions: opening hours, stock, shipping, pricing that lives in a catalog.
- Appointment booking and simple scheduling.
- First-pass lead qualification, so a human only sees conversations worth their time.
If most of your chat volume is this shape, the built-in agent at cents per conversation is hard to argue with. Turning it on and writing good escalation rules is a weekend project, not a transformation program.
When you need an agent that is actually yours
The ready-made option runs out of road in predictable places:
- Your data lives elsewhere. If answers depend on an ERP, a booking engine, or an internal database, you need an agent wired into those systems, not one guessing from a catalog.
- Brand voice is the product. A generic tone is fine for logistics questions and wrong for a brand people buy partly because of how it speaks.
- Escalation is business logic. Which conversations reach a human, in what order, with what context attached: that is policy you should own, not a platform default.
- Volume changes the math. Token pricing is cheap at low volume; at scale, the cost curve and the data you surrender both deserve a real calculation.
This build-or-adopt line is exactly where we spend most of our client conversations at Clodron: the honest answer is often "adopt for support, build for sales."
What to switch on this month
Start with an inventory, not a tool. List your top twenty inbound questions and mark which need live data, which need judgment, and which are pure repetition. Hand the repetition to the platform agent and set the handoff threshold conservatively. For the questions that need your systems or your voice, scope a custom agent with clear escalation rules and data ownership on your side. And read your chat transcripts weekly at first: the agent's mistakes will tell you more about your business than its successes.