Avoid surprise messaging costs

An agent at a b2b sales desk replies to a lead's question about pricing four days after the lead last wrote — not realising the free 24-hour reply window closed on day two — and the message silently fails to send at the API level, with no obvious error telling the agent what actually went wrong or why the customer never got a reply.

The conversation window and billing guard tracks that 24-hour customer service window, and the separate 72-hour free-entry-point window, in real time for every contact, and warns the agent in the inbox the moment a reply would need a paid template because the free window has closed. Before any broadcast goes out, it also estimates the total conversation cost by category and recipient country, so there's a number to check before committing to the send.

The customer service window opens for twenty-four hours from a customer's last message and allows free-form replies inside it; the free-entry-point window runs for seventy-two hours from certain qualifying entry points, like clicking a Click-to-WhatsApp ad, and allows a different kind of message. These two clocks are genuinely independent — a customer service window can close while a free-entry-point window from an earlier ad click is still open, or vice versa — and treating them as one window is the single most common source of billing confusion at volume.

When the customer service window closes and an agent tries to send a plain, non-template message anyway, the send doesn't get rejected with a helpful error — it fails silently at the API level. This is precisely why the guard has to block or redirect the attempt inside the inbox before the send is even tried, rather than relying on Meta's response to surface the problem after the fact; by the time a failure is visible, the customer has already not received a reply and doesn't know why.

Cost estimation before a broadcast depends on getting the country-code-to-rate mapping right, since Meta prices conversations differently by country and category (marketing, utility, authentication, service). Getting that mapping wrong for even a subset of a large segment understates the projected cost of the whole campaign, defeating the purpose of estimating it at all — the estimate is only as trustworthy as the rate table behind it, and that table needs to stay current as Meta adjusts its pricing over time.