The worst outages are the quiet ones

When Meta rejects the credentials, nothing sends: no replies, no broadcasts, no automations. You are told the moment it is detected, with the reason Meta gave.

Meta issues these tokens for around sixty days and they do not renew themselves. A scheduled check asks Meta when yours lapses and warns you inside the last week — the difference between a planned reconnection and an outage.

Meta can flag or restrict a number while your credentials still work — every login check passes and sends fail anyway. You are told that it happened and that reconnecting is not the fix, because that is the first thing anyone tries.

Meta lowers your rating when recent messages draw blocks or reports, and lowers your reach with it. YELLOW is the point where changing what you send still helps; RED means sending is already restricted. Both are reported.

Meta limits how many people an unverified business can start a conversation with each day, whatever your contact count. Nothing fails when you hit it — campaigns simply reach fewer people — so it is the kind of ceiling that goes unnoticed for months. The verification state is checked on a schedule and reported once a week until it clears.

A page access grant can be revoked, or the admin who granted it can lose the role, and inbound DMs simply stop. Nothing errors, because nothing tries — the message is dropped before anything runs. The connection is checked hourly and you are told which page stopped and what Meta said about it, because an expired token and a lost page role read the same from the outside and are not fixed the same way.

If your provider rejects its key or errors, the chatbot answers nobody. You are told which provider failed and whether the key itself was refused — on WhatsApp and on the website widget separately.

Separate from the AI failing, and easier to miss. A chatbot switched on with no usable API key answers nobody, and every screen still shows it enabled — so the first sign would be a customer mentioning they were ignored. You are told which provider has no key and what to add.

Credentials are checked when saved, but a revoked app password or a rotated provider credential passes that check and fails later. A scheduled check catches it and says so, and says WhatsApp is unaffected.

A campaign where some recipients failed is otherwise indistinguishable from one that worked. You are told how many did not receive it, and the likely reason.

Separate from a campaign failing, and quieter. Contacts who unsubscribed, a second copy of an address already on the list, and rows with no email address are all skipped before sending — correctly — but the campaign then records only the number it actually sent to, so a send aimed at 500 that reached 300 looks identical to one aimed at 300. You are told the original size, how many were skipped and which of those three reasons applied. When it skips everyone, you are told that nothing failed and there was simply nobody left.

Your credentials can be perfectly valid while no app is subscribed to receive messages for your WhatsApp account — the exact state where sending still works and replies never arrive. That combination is the easiest of all of these to miss, because every test you would think to run passes. The subscription is checked and you are told to reconnect the number.

WhatsApp accepts a message and can still report minutes later that it was not delivered — the number is not on WhatsApp, the 24-hour window has closed, the template is paused. Each of those needs a different fix, so you are told which one it was, with the wording WhatsApp used, linked to the conversation. Campaign sends are excluded here because they are reported together as one result rather than one alert each.

Scheduled campaigns, delayed automation steps and queued social posts are each carried out by a background job on a timer. If one of those stops doing its work, nothing fails and nothing errors — the queue simply stops moving, and the first sign is a campaign that never went out. Anything still waiting more than half an hour after its due time is reported. The alert says which queue and how many, and does not guess at a cause: a job that has stopped and a job that runs and fails look the same from here.

Holding more contacts than your plan allows does not delete anything, and your inbox, broadcasts and existing contacts all keep working — what stops is importing more, which surfaces as a CSV import being refused for no visible reason. You are told how far over you are and on which plan.

WhatsApp pauses a template when its quality rating drops and disables one that keeps being reported, and it does neither of those things loudly. Your template list is checked against WhatsApp every hour rather than only when you press sync, so a template that is no longer sendable stops being offered instead of failing a whole campaign. You are told which templates and which word WhatsApp used, because paused lifts by itself when the rating recovers and disabled does not.

The only failure here that is nobody’s fault but yours. A conversation somebody started and no one answered does not error, does not appear in any failure count, and looks exactly like a quiet inbox. Waiting time is measured from the customer’s own last message — not from the last activity on the thread, which a campaign would otherwise reset — and once anyone has been waiting more than four hours you get one alert naming the longest five, linked to the unread filter.