AI receptionist launch checklist

• Billing plan matches what you signed up for (no surprise overage rate)
• Time zone set correctly (wrong TZ breaks after-hours, breaks scheduling, breaks reporting)
• Business hours configured including lunch / holidays / early-close days
• Greeting replaced with brand-voice version (no default vendor text remains)
• Recording disclosure phrase reviewed by counsel for your state's consent law
• Call recording retention set per your data policy (most teams: 90 days minimum)
• Team members added with right roles (admin / editor / viewer)
• Notification rules set for escalations, missed calls, integration errors

• All 4–7 top-level branches built and named clearly
• Greeting → branch decision is fast (<3 seconds of AI talking before user can respond)
• Qualification questions narrowed to 3–5 critical fields
• FAQ branch handles top-20 customer questions without escalating
• Escalation triggers configured: explicit request, sentiment, topics, VIP, off-script, emergencies
• Always-on 'speak to a person' fallback tested in EACH branch
• Booking branch tested if booking is in scope
• After-hours flow tested (emergency vs routine paths)
• Spanish-language path tested if multilingual is in scope

• CRM integration tested: net-new caller → new Contact with custom fields populated
• Dedup tested: existing caller (by phone) → existing Contact updated, NOT duplicated
• Call recording + transcript URLs attached to the Contact activity
• Calendar integration tested: booking lands in the right calendar with right duration
• Service-type to duration mapping verified for each service type
• Conflict prevention verified (simulate concurrent booking, verify only one wins)
• API failure tested (disable CRM key, verify fallback queue catches the lead)
• Reschedule + cancellation flow tested

• Number provisioned (new) OR forwarding configured (existing) OR port scheduled (later)
• Inbound caller ID is preserved (test with multiple carriers if forwarding)
• Voicemail fallback configured (what happens if the AI itself is down)
• SMS/text capability configured if you use that channel
• Toll-free vs local number choice made deliberately (cost + caller perception)

• Internal team trained on what the AI handles + when to expect escalations
• Escalation routing tested end-to-end (where does an escalated call actually land?)
• Reporting dashboard reviewed — you know where to look for containment rate, booking rate, escalations
• Runbook documented for common issues (integration error, mis-escalation, wrong booking time)
• First-week tuning calendar blocked (~30 min/day to listen to recordings)
• Stakeholders notified of cut-over date (sales team, customer success, anyone who answers escalations)

If something is meaningfully broken in the first 24-48 hours, you need to be able to revert without a panic. Have this lined up:

• Forwarding can be flipped off in <5 minutes (no port = no panic)
• Vendor has a 'pause' or 'disable' mode that stops new calls from going to the AI
• A live person is on standby to answer the line during the cut-over window
• You know who to call at the vendor's support for production issues
• Customer-facing apology language drafted for the worst case (cut-over hits a snag, customers get a weird experience)