Most AI calling launches slip on paperwork rather than on technology. This is the operational checklist, written from deployments rather than from a statute. It is not legal advice, and anything with regulatory exposure should be reviewed by counsel in your market.
1. Carrier registration for messaging
In the United States, application to person messaging is registered with the carriers through brand and campaign registration. Until it is approved, texting is blocked. Rejections happen and they take time to resolve.
Practical sequencing: start the registration early, design the SMS follow up, and ship voice while it is pending. Voice is not gated by the same process.
2. A documented consent basis
For every list you intend to contact, write down: where the contact came from, when, what they agreed to, and where that record lives. If you cannot answer those four for a list, that list is not ready to dial.
Existing customers, inbound requests and purchased lists are different categories with different risk. Treat them differently rather than uniformly.
3. Disclosure
Decide what the agent says about being an agent, and say it the same way every time. Requirements vary by jurisdiction and are changing, so the defensible position is disclosure by default.
It is also the better product decision. Trust survives disclosure. It does not survive discovery.
4. Recording and retention
Who is notified that the call is recorded, in what words, and at what point. Where recordings are stored, who can access them, and how long they are kept. Consent to record varies by jurisdiction, including within the United States.
5. Calling windows
Set by local rule and local reality, per market, not by convenience. This overlaps with something that also improves results: calling in the window that actually connects produces better outcomes and fewer complaints.
6. An opt out that works
The most common failure on this list. Someone asks to stop, it is recorded in one system, and a different channel contacts them the following week.
Requirements: the opt out is honoured immediately, it propagates to every channel, and it is verifiable in the record. Test it end to end with a real contact before launch.
Two more worth having
A human path. Anyone who asks to speak to a person gets one, during hours when a person exists. An agent that cannot escalate produces complaints that are entirely avoidable.
An audit trail. For every contact: which list, which consent basis, what was said, what the outcome was. If you cannot reconstruct a contact six months later, you cannot defend it either.
Before you switch on
Run one final check that has nothing to do with law. Confirm somebody answers transfers during the calling window. An agent that qualifies and transfers into an unanswered line damages more than it earns, and no amount of compliance fixes that.