Phone Forwarding for AI Agents: Getting Calls to Your Bot

Building the agent is the interesting part. Getting real calls into it is the part that eats a Tuesday. A business has a number on its truck, its website, its Google Business Profile, and its invoices going back years. That number lives inside a phone system somebody set up a while ago, and nobody wants to change it. Your job is to get calls from that number to your AI agent without breaking anything the business relies on.

There are only a few ways to do this, and picking the wrong one is where the pain comes from. Here is how I decide, and the traps I have hit doing it for real clients.

The four ways to get calls to an agent

1. Buy a new number and point it at the agent. Simplest possible setup. You provision a number from your voice provider, attach it to the agent, done. Works great for a new marketing channel: a number that goes on a landing page, a Google Ads campaign, or a specific ad set. Terrible if the business wants its existing number answered.

2. Forward the existing number to a new agent number. The business keeps its number and its current provider. Their phone system forwards calls to the number you provisioned for the agent. This is what I use most, because it is reversible in under a minute. If the agent misbehaves at 9am on a Monday, you turn off forwarding and calls go back to ringing the desk phones exactly as before. That rollback story is worth a lot when you are asking a business to trust a bot with its front door.

3. Port the existing number to your voice provider. You take ownership of the number and point it directly at the agent. Cleanest architecture, no double hop, no extra monthly line. But porting takes days to weeks, it is not instantly reversible, and if anything goes wrong the business has no phone service. I only port after an agent has been running well on forwarding for a while and the client explicitly wants to consolidate.

4. SIP trunk into the agent. If the business already runs a PBX, you can route a trunk or an extension to the agent. Powerful, and the right answer for larger setups, but it needs somebody who understands the PBX. See SIP trunking for the concept.

For most small businesses the answer is 2, then maybe 3 later. Start with forwarding.

The carrier forwarding trap

Here is the one that cost me real time and is not written down anywhere useful.

Business phone systems from carriers often have several features that all sound like they should forward a call, and only one of them actually does what you need. On one deployment I spent a while fighting a system where the obvious path was its auto-receptionist: the menu tree feature that answers calls and offers options. It looked like the right place to send calls onward to our AI number. It was not. The auto-receptionist would answer the call itself, and the call never reached the agent cleanly.

What worked was the group forwarding setting, one layer down in the account, not the auto-receptionist. Once calls were forwarded at the group level, they landed on the agent number intact and everything behaved.

The general lesson, which applies well beyond that one carrier: if a carrier-side feature answers the call before forwarding it, you have a problem. Anything that plays a greeting, offers a menu, or otherwise picks up the line has already taken control of the call. What you want is a feature that redirects the call without answering it. Look for settings named forwarding, hunt group, or find-me/follow-me. Be suspicious of anything named auto-attendant, auto-receptionist, IVR, or virtual receptionist.

Practical approach when you land in an unfamiliar carrier portal:

• List every feature that mentions forwarding, and try the least clever one first.
• If a feature offers to record a greeting, it answers the call. Skip it.
• Test after every single change with a call from your own cell. Do not batch three settings changes and then test.
• Write down what worked. You will be back in that portal in six months and you will not remember.

Which forwarding rule to use

Assuming your carrier supports the normal set, you have three choices:

Rule	Behavior	Good for
Always forward	Every call goes straight to the agent	AI-first setups where the agent is the front door
Forward on no answer	Rings the business first, agent picks up after N rings	Businesses that want humans to have first shot
Forward on busy	Agent catches overflow only	Shops with one line and real call volume

Most clients want "forward on no answer" at first, because it feels safer. It is a reasonable starting point but be honest about the tradeoff: the caller now sits through four or five rings before the agent answers, which is five seconds of dead time and a measurable share of hang-ups. Once a client sees a month of transcripts and trusts the agent, many of them move to always-forward.

If you use forward-on-no-answer, set the ring count deliberately. Four rings is roughly twenty seconds in most configurations. That is a long time to a caller who is comparing three contractors on Google.

The double-voicemail problem

The most common breakage in a forwarded setup: the caller ends up in a voicemail box instead of the agent, or the agent's call goes to the business's voicemail on transfer back.

Both come from the same root. In a forwarded chain, more than one system in the path thinks it owns voicemail. If the business's carrier voicemail picks up after 25 seconds, and your forward-on-no-answer fires at 20 seconds, you are in a race. Sometimes the agent wins, sometimes voicemail does, and the client reports it as "sometimes it works".

The fix is to make one system own voicemail and disable it everywhere else. My default is to disable the carrier voicemail entirely once the agent is live, because a well-built agent takes better messages than a voicemail box and delivers them somewhere useful. If the client insists on keeping carrier voicemail, then set the forward timer well below the voicemail timer so the race is not close.

Same principle applies on the transfer leg. If your agent transfers to a human whose line rolls to voicemail, the caller lands in a personal voicemail box with no context. Handle that explicitly, which is covered in call transfer flows.

Caller ID survives, sometimes

When a call is forwarded, what your agent sees as the caller's number depends on the carrier. Usually the original caller's number comes through and the forwarding number appears as a diverted or redirecting party. Sometimes the carrier substitutes its own number, and every call arrives at your agent looking like it came from the business's own line.

This matters if you use caller ID for anything: personalizing the greeting, looking up an existing customer, or sending a follow-up text. Test it before you build logic on top of it. Place a real call from a phone whose number you know, and check what the agent actually received. Do not assume, and do not test this in a simulator, because simulators generally cannot inject caller ID at all. That gap is the subject of testing voice agents.

If caller ID does not survive forwarding, your options are to port the number (which removes the forwarding hop) or to have the agent ask for the number in conversation. Asking is not the end of the world. It also confirms the number is one the caller can actually be reached on, which a caller ID capture never does.

Voicemail-first setups for lead generation

Not every deployment needs the agent to answer live. For lead-gen sites, especially in home services, I sometimes run the agent as the voicemail rather than the receptionist. Calls ring the business, and if nobody picks up, the agent answers instead of a voicemail box. The caller gets a real conversation instead of a beep, and the business gets a structured lead with a name, a number, and a job description.

This is an easy first sell to a nervous client. Nothing about their existing call flow changes. The agent only ever handles calls that were already going to be missed. Pair it with missed-call text-back and you cover the callers who hang up before the agent answers too.

The rollout checklist

What I actually do before flipping forwarding on for a client:

1. Provision the agent number and confirm the agent answers it correctly on a call from my own cell.
2. Call every destination in the agent's transfer table from my own phone.
3. Set the forwarding rule in the carrier portal, one change at a time.
4. Call the business's published number from a phone that is not on the account. Confirm the agent answers, confirm what caller ID it received.
5. Check voicemail ownership. Disable duplicates.
6. Call again outside business hours and confirm the after-hours path.
7. Leave forwarding on for a low-traffic window first if you can. A Friday afternoon rollout is a bad idea; a Tuesday morning one gives you the day to watch it.

None of that is clever. All of it has caught something real.

FAQ

Should I forward my existing number or port it to the AI provider?

Forward first. Forwarding is reversible in a minute, which means a bad day for the agent is not a bad day for the business. Port later, once the agent has proven itself and the client wants to drop the extra line and clean up the routing.

Why do calls hit an auto-attendant instead of my AI agent?

Because the carrier feature you configured answers the call before forwarding it. Auto-receptionist and IVR features take control of the line. Look for a plain forwarding or group forwarding setting instead, usually one level deeper in the account settings than the feature that markets itself as call routing.

Will the AI agent see the original caller's phone number?

Usually, but not always. Some carriers replace the caller ID with the forwarding number. Test with a real call from a known phone before you build anything that depends on caller ID, and have the agent ask for a callback number as a backstop.

Does forwarding add cost or latency?

A small amount of both. You are paying for the forwarded leg on the carrier side plus the agent's usage on your provider side, and the extra hop adds a modest amount of connection time. Neither is usually significant at small-business volumes, and voice usage on these platforms tends to run at cents per minute.

Can I keep humans answering first and only use the agent for overflow?

Yes, that is forward-on-busy or forward-on-no-answer, and it is a common starting configuration. Just be aware the caller waits through the ring cycle before the agent picks up, and set the ring count low enough that they are still on the line when it does.