A new-contact slot is not a message count
The default allowance is 50 new contacts per sending number in a rolling 24-hour period. A number or organization can have a different configured limit. Replies and follow-ups do not use another new-contact slot, and adding a name to Contacts does not consume one.
Rolling 24 hours means capacity does not necessarily reset at midnight in the dashboard timezone. Earlier outreach ages out of the window as time passes. The Analytics date picker changes the report, not the sending allowance.
Pacing is a separate check
First messages to new people go out about five minutes apart, varied a little each time, and never more than eight in an hour or three in ten minutes. Fifty new people take most of a working day rather than an hour: a number that reaches dozens of strangers an hour is what Apple stops trusting, and that is when its messages start going out as texts instead of iMessage. Replies and messages to people you already talk to are not paced this way.
Having new-contact capacity does not mean every burst of messages will be accepted. Per-line pacing, duplicate detection, content checks, and availability still apply. Read the returned reason to tell these cases apart.
For API rate-limit responses, honor Retry-After when provided and use a bounded retry policy. Do not repeatedly press Send or run multiple workers against the same failed request hoping one will bypass the limit.
What Auto routing changes
Auto can choose among eligible project numbers for a new contact and save that choice. It keeps existing conversations on their saved number. A full or unavailable assigned line is not a reason to silently move an established customer conversation.
Plan expected outreach around the capacity you actually have. If your normal workload does not fit, ask support to review the setup and keep the customer’s expected sending identity consistent.
Do not assume a rejected request is queued
Check the operation’s actual response. A rejection is not a promise to send the message later. A pending message or queued integration job is different and should be inspected before retrying.
If an automation schedules later attempts, make it stop when a customer replies or opts out. Keep retries linked to the original workflow execution so they do not become new duplicate outreach.