Voicebot and PBX: How Phone System Integration Works

by Crafter.ai · July 13, 2026 · 9 min read

Voicebot PBX integration is far simpler than most companies expect: nothing needs to be replaced, the existing infrastructure stays untouched, and the technologies involved are the same standard, thoroughly proven ones that businesses already rely on every day for their telephony. A voicebot brings to the phone an assistant that speaks and understands natural language: it answers calls, understands the request, handles or routes it and — when needed — hands the conversation over to a person.

Yet whenever a company evaluates a voice AI agent, the first question from the IT department is almost always the same: "how does it connect to our phone system? Do we have to change anything?". The short answer is: no. The voicebot sits alongside the existing PBX through two standard paths — a dedicated phone number or a SIP connection — and the company decides which calls to hand over to it, and when.

In this guide we look at how a phone voicebot is built, what technically happens during a call and how it integrates with the company PBX in a straightforward, low-impact way, including a concrete example based on an Alcatel PBX. The underlying message is simple: there is nothing esoteric about it.

Table of Contents

• What a voicebot does on the phone
• The four building blocks of a voicebot
• Classic pipeline or speech-to-speech models
• What happens during a phone call
• How the voicebot connects to the company PBX
• Reliability, security and privacy
• How to get started: the information you need
• A concrete example: the Alcatel PBX
• Quick glossary
• Conclusions
• FAQ

What a voicebot does on the phone

Unlike a menu-based answering system — the classic IVR with its "press 1 for…" — a voicebot understands free-form sentences, keeps the context of the conversation and knows how to act: it can query business systems, give precise answers and transfer the call to the right person. It is the same gap that separates an intelligent voice assistant from an evolved answering machine, which we explored in our article on the difference between voicebots and chatbots.

Here are some concrete examples of what a voice assistant connected to the company telephony can do:

• welcome and route calls to the right office or person;
• manage appointments and bookings;
• answer questions about order status, shipments, opening hours and general information;
• cover peak times and out-of-hours slots without letting phones ring unanswered;
• make outbound calls: reminders, confirmations, callbacks.

All of this happens alongside the existing PBX, without replacing it. This is exactly what makes the voicebot a practical lever for call center automation: the company keeps its own telephony and simply adds one more resource.

The four building blocks of a voicebot

A voicebot is born from the collaboration of four elements, each with a clearly defined role built on consolidated standards. Understanding who does what makes it clear how modular — and free of magic — the architecture really is.

Every element is modular and replaceable: this keeps the solution solid over time and free from technological lock-in. The knowledge base, in particular, is what allows the bot to answer only with verified company information — a mechanism we covered in depth in our guide to Retrieval-Augmented Generation.

Classic pipeline or speech-to-speech models

The voice engine can work in two ways, and the choice depends on the goal.

The classic pipeline (STT → LLM → TTS) consists of three distinct steps: speech is transcribed into text by speech recognition, the text is understood and processed by the language model, and the textual answer is converted back into speech by voice synthesis. It is a proven and highly controllable approach, which also offers a wide range of synthetic voices to choose from.

Realtime models (speech-to-speech) remove the text step altogether: the audio stream of the call is sent in streaming — typically over a websocket — directly to the model, which in turn responds with an audio stream. They can offer lower latency, but that advantage is theoretical: it depends on the contract with the model provider and on network load at the time of the call. Keep in mind that voice selection is limited to the model's own voices, while the classic pipeline offers many more — in both cases, though, very natural sounding.

What happens during a phone call

When a call comes in, the voice engine manages a real-time dialogue: it listens, understands, answers and, if needed, queries business systems or transfers to a person. Everything happens in a fraction of a second, with the naturalness of a conversation — the caller can even interrupt the bot, and the bot notices.

The flow of a call handled by a phone voicebot can be summarised like this:

1. the caller dials the number and the network — or the PBX — routes the call to the voicebot;
2. the bot answers with its welcome greeting and listens to the request in natural language;
3. at every conversation turn the voice engine understands the request and works out the answer;
4. if real data is needed (an order status, an appointment), the bot queries the business system and uses the result in its answer;
5. if a person is needed, the bot transfers the call to the right extension or operator, through the PBX.

The same mechanism applies to outbound calls: the platform initiates the call to the recipient, then the dialogue proceeds in exactly the same way.

How the voicebot connects to the company PBX

The voicebot neither touches nor replaces the PBX: it sits alongside it as an additional resource, and the company decides which calls to hand over and when. There are two ways to connect them, both standard and well established; they often coexist.

Path A: a dedicated phone number

The voicebot has its own number. The PBX forwards to it the calls it should handle: a branch of the menu, evening hours, overload moments, or a number published directly. On the PBX side, a single forwarding rule is all it takes, and when an operator is needed the bot transfers the call back to the company.

It is the fastest path to activate and the one with the lightest footprint: ideal for starting right away or for a first pilot phase.

Path B: a SIP connection

The voicebot connects to the PBX as a SIP trunk, meaning the two systems "talk" to each other over VoIP. Routing stays inside the company telephony, so even transfers to internal extensions happen without going out over the public network.

It is the most integrated path: maximum control over quality and security, immediate internal transfers. Suited to significant call volumes and structured use.

Dedicated number or SIP trunk: the comparison

In both cases the company remains in charge of the rules: when to send calls to the bot (schedules, menus, overload) and who to transfer to when a person is needed.

Reliability, security and privacy

A phone channel is a critical channel, and integrating a voice assistant must meet the same reliability standards as professional telephony. There are five fixed points:

• Always a plan B: if for any reason the bot is unreachable, the call goes back to the PBX and to the usual handling. No risk of "dead" lines.
• Protected communications: on the SIP connection, signalling and audio encryption and restricted access are used — the standard practices of professional telephony.
• Data in Europe: the platform — and with it all the data collected for statistics — is hosted in a datacenter within the European Union.
• Zero Data Retention (optional): a mode can be enabled in which conversation content is not stored, for maximum GDPR compliance.
• Data handled with care: conversations are personal data and are managed in compliance with the GDPR, with clear purposes, notice to callers, defined retention periods and recording only if agreed.

How to get started: the information you need

Designing a tailored integration takes just a few pieces of information, gathered with the company's technical contact in a short initial conversation:

• the PBX (brand, model and version) and whether it is on-premise or in the cloud;
• the preferred path: dedicated number or SIP connection;
• when calls should reach the bot and who to transfer them to;
• any systems to integrate (CRM, ERP) for the answers;
• guidelines on recording and consent, where applicable.

With these elements a concrete plan and activation timeline are defined. The technical connection work is handled by the platform provider, in coordination with the company's IT team.

A concrete example: the Alcatel PBX

To make it concrete how straightforward the integration is, let's take an Alcatel PBX as an example: a manufacturer that has been operating for many years with solid, reliable solutions, whose PBX systems are fully compatible with IP telephony. It is precisely this maturity that makes the integration a well-known, surprise-free journey: both paths described above are supported.

With Path A (immediate), a rule is set up that forwards the desired calls to the voicebot's number — a branch of the menu, evening hours, peak moments. It is extremely quick and does not modify the existing PBX configuration. With Path B (integrated), a dedicated SIP connection to the voicebot is set up, with the related routing rules; transfers to internal extensions stay inside the PBX.

In practice, nothing extraordinary is required: the operations are the usual configuration tasks — a forwarding rule or a SIP connection — carried out together with the company's IT team. To spell out the exact steps, knowing the PBX model and version is enough: from there, a precise plan is prepared and shared with whoever manages it.

The same reasoning applies, with minimal differences, to PBX systems from all the other major manufacturers: the two paths — dedicated number or SIP connection — are the industry standard.

Quick glossary

• PBX (Private Branch Exchange): the company phone system that manages calls and extensions.
• SIP / SIP connection: the standard way to make two phone systems talk to each other over the Internet or an IP network.
• VoIP: telephony that travels over the data network instead of traditional lines.
• IVR: the automated phone menu ("press 1 for…").
• Speech-to-speech (STS): voice models that receive and return an audio stream directly, without going through text.
• Transfer (handover): the moment the call passes from the bot to a person.
• Zero Data Retention: a mode in which conversation content is not stored.
• CRM / ERP: the business systems holding the data (customers, orders, appointments) the bot queries to answer.

Conclusions

Integrating a voicebot with the company phone system requires no revolution: a dedicated number with a forwarding rule to start quickly, or a SIP trunk for a structured integration with immediate internal transfers. The architecture is modular — phone line, voice engine, platform, business systems — every element is replaceable, and the company always stays in charge of the routing rules. With the fallback to the PBX, encrypted communications, data hosted in Europe and optional Zero Data Retention, reliability and GDPR compliance are covered too.

If you want to see how a voice agent can answer your company's calls, explore the Crafter.ai voicebot solutions or request a personalised demo: the details of your PBX are all it takes to define a concrete activation plan.

FAQ: voicebot and PBX

How does a voicebot connect to the company PBX? In two standard ways: with a dedicated phone number the PBX forwards calls to, or with a SIP connection (SIP trunk) that lets the voicebot and the PBX talk over VoIP. The two paths often coexist.

Do I have to replace my PBX to use a voicebot? No. The voicebot sits alongside the existing PBX without touching or replacing it: in the simplest case, a single forwarding rule on the PBX is enough.

What happens if the voicebot is unreachable? The call goes back to the PBX and to the usual handling. There is always a plan B, so there is no risk of dead lines.

What is the difference between a dedicated number and a SIP connection? The dedicated number is the fastest, lightest path, ideal for a pilot: only a forwarding rule is needed. The SIP connection is the most integrated path: transfers to extensions stay inside the company telephony and it suits high call volumes.

Is a phone voicebot GDPR compliant? Yes, as long as the platform treats conversations as personal data: notice to callers, defined retention periods, data hosted in a datacenter within the European Union and, if requested, a Zero Data Retention mode that does not store conversation content.

Can a voicebot transfer the call to a human operator? Yes. Transfer rules are defined on the platform: when a person is needed, the bot hands the call over to the right extension or operator while the PBX keeps its usual call handling.

Related articles

• Voice AI Agents: The Complete Guide
• Chatbot and IVR Comparison
• Technology in Call Center Automation
• Voicebots and Chatbots