Provisioning Phone Numbers for AI Calling: Twilio + VAPI Setup Guide
A practical guide to phone number provisioning, A2P registration, and area code strategy for AI calling deployments.
Haroon Mohamed
AI Automation & Lead Generation
Why phone number setup matters
The phone number you call from affects:
- Answer rate (local numbers answer at 2-3x the rate of toll-free)
- Spam labeling (poor reputation = "Spam Likely" labels = no answers)
- Compliance (A2P 10DLC required for US business calling)
- Cost (toll-free vs. local pricing varies)
A poorly-provisioned number tanks your AI calling deployment before any call connects. Let's set it up right.
Step 1: Choose your number type
Local number (recommended for outbound)
A 10-digit number with a local area code (e.g., 415-XXX-XXXX for San Francisco).
Pros:
- Highest answer rate (people trust local numbers)
- Lower cost (~$1.15/month at Twilio)
- Less likely to be marked spam initially
Cons:
- Limited to one geographic area perception
- May not match your business location
When to use: Outbound calling to consumers in a specific region.
Toll-free number
800, 833, 844, 855, 866, 877, 888.
Pros:
- Recognizable as business numbers
- No geographic implication
Cons:
- Lower answer rate (often associated with telemarketers)
- More likely to be marked spam
- Slightly higher cost
When to use: Inbound only, or when you specifically want a "business" perception.
Long Code SMS-only number
Some platforms offer SMS-only long codes. Cheaper but cannot make voice calls.
When to use: SMS-only campaigns.
Step 2: Register for A2P 10DLC (US business calling)
A2P 10DLC = Application-to-Person 10-Digit Long Code messaging registration.
Required by US carriers since 2021 for business SMS/voice calling. Without it:
- Carriers throttle or block your messages
- Calls get "Spam Likely" labels
- Compliance violations
Registration process
-
Brand registration: identify your business
- Legal entity name
- EIN (US tax ID)
- Address, phone
- Approval typically 1-7 days
- Cost: $4-$44 one-time depending on type
-
Campaign registration: describe what you'll send
- Campaign type (Standard, Sole Proprietor, Public Service, etc.)
- Use cases (Marketing, 2FA, Mixed, etc.)
- Sample messages
- Opt-in/opt-out flow
- Cost: $1-$15/month per campaign
-
Phone number assignment: link numbers to the campaign
- Numbers can be assigned in batches
- Up to 49 phone numbers per Standard campaign
Brand types
- Sole Proprietor: for individuals, ~50 daily message limit
- Standard (Low Volume): for small businesses, up to 6,000 messages/day
- Standard (High Volume): for growth, up to 200,000 messages/day
- Charity / Public Service: different categories
For most automation deployments: Standard Low Volume.
Step 3: Provision through Twilio
Buy the number
Twilio Console → Phone Numbers → Buy a Number.
Filters:
- Country: US
- Region/state: choose based on your audience
- Capabilities: Voice, SMS (both)
- Type: Local (typically)
Pricing: $1.15-$3.00/month per number depending on type.
Configure for AI calling
After purchase:
- Navigate to the number's configuration
- Voice → Configure with: SIP trunking or webhook
- If using VAPI: VAPI handles voice routing; configure number to point to VAPI's SIP endpoint
- SMS → Configure with: webhook to your messaging tool (GHL, your custom endpoint)
VAPI's setup wizard walks you through this if you're using their platform.
Step 4: A2P registration in Twilio
Twilio Console → Messaging → Regulatory Compliance → A2P 10DLC.
-
Register your brand:
- Brand type
- Legal info
- Wait for approval
-
Register a campaign:
- Campaign use case
- Sample messages
- Phone numbers to link
- Wait for carrier approval (5-15 days)
-
Verify approval status
Until campaign is approved and numbers linked, expect SMS deliverability issues.
Step 5: STIR/SHAKEN attestation
STIR/SHAKEN is the FCC's caller ID authentication framework. Calls without proper attestation can be marked as spam.
Twilio handles STIR/SHAKEN automatically for verified numbers, but it requires:
- Verified business identity
- Proper number provisioning (not transferred from sketchy sources)
- A2P registration
For production AI calling: ensure your account is in good standing with verified caller ID.
Step 6: CNAM registration (caller ID name)
When you call someone, their phone may show "John from XYZ Company." This is CNAM (Caller ID Name).
Without CNAM registration: shows as just the number, or worse, "Unknown" or "Spam."
Twilio CNAM service: ~$0.30/month per number.
For local-area campaigns, CNAM display significantly improves answer rates.
Step 7: Avoid spam labeling
Even with proper registration, numbers can get flagged as spam by carriers' algorithms.
What triggers spam labels
- High call volume from a single number quickly
- High disconnect rates
- Reports from recipients
- Patterns that match known robocalling
How to avoid
- Spread volume across numbers. Don't run 5,000 calls/day from 1 number. Spread across 10-20 numbers, 250-500 calls each.
- Rotate numbers for SMS. Same principle. Better to send 100 SMS from 5 numbers than 500 from 1.
- Maintain low complaint rate. If recipients report your number, carriers flag faster. Quality of outreach matters.
- Warm up new numbers. Don't max-out a brand-new number on day 1. Start at 50 calls/day and ramp up over 2 weeks.
Number management strategy
For AI calling deployments doing significant volume:
Number pool
10-20 phone numbers, all registered, all warmed up.
Each number used for ~250 calls/day. If one gets flagged, retire it and replace.
Geo-matching
For consumer outbound, match the area code to the prospect's area code.
- Prospect in 415 → call from 415-XXX
- Prospect in 305 → call from 305-XXX
This dramatically improves answer rates (people answer "local" numbers).
Tools: VAPI supports number selection logic. Or build in Make.com with a lookup table.
Number rotation
Even within a region, rotate numbers daily/weekly. Reduces single-number burnout.
Pricing summary
For 10-number deployment for AI calling:
- Phone numbers: 10 × $1.15 = $11.50/month
- A2P brand registration: ~$44 one-time
- A2P campaign registration: ~$10/month
- CNAM: 10 × $0.30 = $3/month
- STIR/SHAKEN: included
- Twilio per-minute: ~$0.013/min outbound × ~10,000 minutes = $130/month
- VAPI platform fees: ~$0.05/min × 10,000 = $500/month
Total estimated: $700-$800/month for a deployment doing ~3,000 calls/day average.
Common pitfalls
1. Skipping A2P registration
"It's working without it." Until carriers crack down. Then your throughput drops by 80% overnight.
2. Buying numbers in bulk without warm-up
20 fresh numbers + day-1 high volume = many of those numbers will be flagged within a week.
3. Not monitoring spam reputation
Free tools like FreeCallerLookup.com or carrier reports show how your numbers are perceived. Check monthly.
4. Reusing flagged numbers
If a number is flagged as spam, retire it. Don't try to "rehab" it. Carriers' bad reputation is sticky.
5. Running outbound and inbound on same number
If you do both, prospects who call back hit your AI assistant. That's fine. Just configure inbound and outbound logic separately.
Inbound configuration
If you also want to receive calls:
In VAPI: assistant configuration → "Inbound" assistant.
Different prompts for inbound vs. outbound:
- Inbound prompt: "Thanks for calling [Company]. How can I help?"
- Outbound prompt: as designed in your outbound campaign
VAPI supports separate logic per direction.
International calling
If you call outside the US:
- Different carriers, different rules
- A2P 10DLC is US-only
- Each country has its own regulations (UK ICO, Canada CRTC, EU GDPR)
- Generally more expensive per minute
For most US-focused businesses: stick to US calling. International adds complexity that's rarely worth it for SMB.
Sources
Twilio documentation from twilio.com/docs (phone number provisioning, A2P 10DLC). FCC STIR/SHAKEN information from fcc.gov/call-authentication. CTIA A2P 10DLC guidelines from ctia.org. Pricing as of April 2026 from each provider's pricing pages.
Need help setting up a phone number infrastructure for AI calling? Let's talk — typical setup is 2-3 days plus 1-2 weeks for A2P approval.
Need This Built?
Ready to implement this for your business?
Everything in this article reflects real systems I've built and operated. Let's talk about yours.
Haroon Mohamed
Full-stack automation, AI, and lead generation specialist. 2+ years running 13+ concurrent client campaigns using GoHighLevel, multiple AI voice providers, Zapier, APIs, and custom data pipelines. Founder of HMX Zone.
Related articles
How to Train Your AI Caller for a Specific Vertical: Solar, Real Estate, HVAC
Most AI calling deployments start with a generic prompt: "qualify this lead and book an appointment." Generic prompts produce generic conversations. They miss: - Industry-specific qualifications - Co…
AI Voice for Real Estate Lead Follow-Up: What Works in the First 5 Minutes
National Association of Realtors data is clear: ~50% of buyers and sellers go with the first agent who responds. Most real estate teams call leads within 10-30 minutes. By then, the lead has already …