How to Connect Your Business Number to an AI Receptionist

What you are actually connecting

Callers do not care about your vendor. They care that the digits on the website still work. The connection job is: take an inbound call that already belongs to you, deliver it to software that can answer media, and still have a path if that software is unreachable. Everything else — greetings, CRM, calendars — is downstream of a working voice path.

There are two honest architectures. In the first, the number stays with your current operator and a forward (or a PBX dial-plan) places a second call to a number or SIP URI the vendor gave you. In the second, the number is hosted on a trunk you control (or the vendor hosts it) and inbound SIP is the native event. Mixing them without a diagram is how loops start.

Call forwarding: how the call actually routes

Alice dials your published number, B. B’s switch accepts the INVITE or the SS7 setup. Instead of ringing a handset that belongs to B, the switch originates a new leg to destination C (the vendor number or URI) and bridges A–C. Alice’s phone still shows that she called B. Whether C sees Alice’s number, B’s number, or a generic withheld identity is a carrier policy, not a marketing choice. You only know which you got after a test call from a mobile that is not in your office group.

The forward consumes a channel on B. On many GSM and analogue lines that means one conversation at a time. A second caller hitting B while C is still bridged may hear busy, carrier voicemail, or a “please wait” if the operator sells a queue — you cannot assume a queue. Landlines behind a small PBX may allow more if the PRI or SIP towards the operator has spare channels.

Audio quality is the worst of both legs. If C is an international number, you may also pay two termination charges and pick up extra delay. Some operators refuse to forward to off-net or overseas destinations without a ticket. If the vendor hands you a local DID in the same country as B, forwarding is more likely to succeed and cheaper to run.

SIP and BYOC: how the call actually routes

SIP is signalling. Your number is a DID on a trunk. When Alice calls B, the carrier sends an INVITE to the session border controller (SBC) you nominated — yours, your PBX vendor’s, or the receptionist vendor’s. That INVITE already contains the called number and, if the carrier is well behaved, the calling number. The receptionist answers that INVITE directly. There is no second PSTN hop unless you later transfer to a mobile.

BYOC means you keep the commercial relationship with the carrier. You point the trunk at the vendor’s SBC (IP allow-list, TLS or IPSec if they require it, codecs you both support, usually G.711 for PSTN comfort). You own number porting, CLI presentation, and the emergency-services story for that number. The vendor owns answering the INVITE. If you already have a 3CX, Teams Direct Routing, or a hosted PBX, the “connection” may be a dial-plan: inbound DID → send to vendor SIP URI, with a failover destination if the OPTIONS ping fails.

SIP is how you get overlapping calls, cleaner caller ID, and a failover you can actually program (route to a hunt group if the vendor SBC does not reply). It is also how you get a weekend outage because a certificate expired or an IP allow-list still has last year’s SBC. Someone in the business has to own that box.

Headers decide callbacks. From, P-Asserted-Identity, and Diversion are not interchangeable. If the carrier puts Alice in PAI and a generic number in From, a naive CRM mapping stores the generic number. Ask the vendor which header they persist, then prove it with a withheld-ID call and a normal mobile call. Codec mismatch (one side insisting on a wideband codec the PSTN cannot carry) shows up as one-way audio or robotic speech after a transfer to a GSM mobile. Agree G.711 on the PSTN-facing leg unless both sides have tested otherwise.

Conditional versus unconditional forwarding

Unconditional forwarding (often advertised as “divert all”) sends every inbound call to C immediately. The desk phone on B never rings. Use it when the desk should not compete, or when the number is a dedicated inbound line. It is simple to reason about and easy to forget you enabled.

Conditional forwarding is a family. No-reply: B rings a handset for N seconds, then C. Busy: if B is already in a call, go to C. Unreachable: if the handset is off-net, go to C. These are the right tools when a person at the desk should win when they are present. They are also how you accidentally double-answer: the desk picks up at second eight while C has already been invited because N was six. Align the ring timer with the vendor’s answer timeout.

Time-of-day forwarding lives on the PBX more often than on the SIM. “After 18:00 send to software” is a schedule, not a GSM star code. If your only control is the mobile, you will be changing diversions by hand, which will fail on the first public holiday.

What you need in hand before you start

• The public number in E.164, and confirmation it is the one on the website, Google listing, and invoices — not a second “office mobile”.
• Carrier account access or a named person who can raise a ticket. Forwarding PIN, if the operator uses one.
• The destination the vendor expects: a DID, a SIP URI, and the IP addresses or FQDN of their SBC.
• A fallback number that a human will actually answer, plus the voicemail box that is checked.
• A second SIM or a colleague’s mobile on another network, for outside-in tests.
• Written cancel instructions: star codes, portal screenshots, or the PBX login that disables the route.

Step-by-step setup sequence

Do not start on the published number. Worked sequence for a Dubai landline that currently rings a desk phone, moving to forwarding (the SIP version swaps steps 3–5 for trunk credentials).

1. Put the vendor on a spare DID or a test mobile you control. Complete greetings, recording notice, and a dummy calendar there. 2. Call that test destination from the outside SIM until answer time, language, and CRM write are boring. 3. Ask the carrier whether B can forward to the vendor DID; if they say no, stop and move to SIP or a local DID they will accept. 4. Enable forwarding on a quiet hour, not during a campaign. 5. Place one test call to B from the outside SIM. 6. Place a second call while the first is still up. 7. Force the vendor unreachable (they should have a “maintenance” or you firewall their IP in a controlled window) and confirm B still hits the fallback, not a loop. 8. Only then tell staff the public number has moved. 9. Keep the cancel code in the same drawer as the office keys.

For SIP: 1. Get the DID on a trunk you can change without a week-long port. 2. OPTIONS ping the vendor SBC from your side. 3. Send a test DID, not the public one. 4. Check CLI in both directions. 5. Add the failover dial-plan: if 503 or timeout, hunt mobiles. 6. Port or point the public DID last.

Pre-launch test plan

• Outside-in call from another operator: answer within the ring count you promised yourself. Time it. If it is slower than your carrier voicemail timer, voicemail will win in the field.
• Withheld caller ID: the flow still asks for a number and reads it back. Confirm the CRM stored digits, not “private”.
• Two overlapping inbound calls: document what the second caller hears, in writing, with a screenshot of the CDR if you have one.
• Transfer to a mobile that does not answer: timeout, then next hop, then voicemail you check. Leave a unique test phrase and retrieve it.
• Vendor unreachable: public number must not loop; it must ring fallback. Do this in a planned window with a colleague on the fallback phone.
• CLI callback: call the number the CRM stored; it must be Alice, not your own DID.
• Long call (several minutes) for audio drop-outs and one-way audio, including after a transfer.
• Cancel-and-restore: disable the path, confirm the desk rings, re-enable once. If you cannot undo, you are not ready.
• DTMF: if you collect a PIN or extension, test from a GSM keypad, not only from a softphone.

Keep a sheet with date, tester, network (which operator the outside SIM uses), and pass/fail. A test that only worked from the office Wi-Fi calling app is not a test of the public number.

Failure points that show up after you think you are done

Forwarding loops. B forwards to C; C’s “unavailable” destination is B again; the call circulates until a timer dies. You hear silence or a fast busy. Break it by making C’s failover a different number you own, never B.

Carrier restrictions. Prepaid or consumer mobiles with forwarding disabled; corporate plans that need an admin to enable divert; geographic numbers that cannot land on an international URI; operators that strip diversion headers so C never sees Alice. These are tickets, not settings inside the receptionist product.

Caller ID behaviour. Staff ring back the office because the CRM stored B. Sales think Alice withheld her number when the forward presented “private”. WhatsApp on the same mobile steals the ring. Fix identity at the trunk, and do not use a personal dual-SIM as a public DID.

Simultaneous call limits. The greeting is perfect and the second caller is in carrier voicemail you do not monitor. Raise capacity on the trunk or publish a second number for campaigns; do not blame the script.

Unreachable vendor. If their SBC is down and your only path is forward-to-them, callers get whatever the carrier does with an unanswered forward — often the operator mailbox. Program fallback independently of the vendor’s health.

Why the same number behaves differently by country and carrier

Star codes are not a standard you can trust across borders. The GSM codes many people remember work on some UAE and India mobiles and do nothing on others, or require a USSD menu, or are blocked on business plans. Always confirm on that SIM, on that operator, with a test that you then cancel.

Number type matters. A mobile, a geographic landline, a toll-free, and a SIP DID have different forwarding, CLI, and porting rules. Toll-free in particular often cannot be casually diverted without the toll-free provider in the ticket. Porting a number between operators can reset diversions you thought were permanent.

Cross-border hops add consent and quality issues. Sending a UAE number to a destination that answers in another country may be disallowed, expensive, or slow. Keeping media in-region is an operational choice as well as a data-handling one; products in this space often describe data handling as designed to align with UAE PDPL and India DPDP requirements, which does not replace a conversation with your own adviser about where audio sits.

This guide cannot configure your PBX for you, cannot promise a carrier will enable a divert, and cannot make a consumer mobile into a contact-centre trunk. If the number is politically the founder’s personal SIM, fix that identity first.

Frequently asked questions

Monarc Voice is in development and is built to sit on an existing business number once the forwarding or SIP path is agreed — it is not a carrier, and it is not launched yet.