AI receptionist: no call goes unanswered.

An AI receptionist takes incoming calls when no one is available: during the lunch break, at the end of the day, at peak times, or outside business hours. It relies on three building blocks: speech recognition, which turns speech into text; an AI model, which understands the request and picks the answer; and speech synthesis, which replies in a natural voice.

What it does well: greeting callers, identifying why they’re calling, answering common questions with your information, booking a slot, creating a CRM record, and alerting the right person.

What it shouldn’t do: negotiate a price, handle a sensitive complaint alone, make up an answer it wasn’t given, or pass itself off as a person. In those cases, it transfers the call or schedules a callback, with a summary of the conversation. The goal is simple: no caller hangs up without an answer, and your team keeps the conversations that call for judgment.

An interactive voice response system (IVR) offers a keypad menu: “press 1 for sales.” It routes the call but handles nothing. If no one picks up at the end of the queue, the call ends up in voicemail.

An AI voice agent has a conversation. Callers explain what they need in their own words, and the agent asks the right questions and acts: it checks the calendar, offers a slot, creates the record, sends the confirmation. There’s no menu tree to memorize.

• The IVR sorts; the AI voice agent handles.
• An IVR assumes someone is available at the end of the queue; the agent takes over when no one is.
• An IVR keeps nothing; the agent writes up every call.

The two can coexist: your phone system stays in place, and the agent only takes unanswered calls, or calls during the hours you choose.

A voice agent that can’t see anything about your business can only take messages. Connected to your tools, it becomes genuinely useful.

• The calendar (Google Calendar, Outlook): it sees each person’s open slots and books directly, following your rules on duration and lead time.
• The CRM: it recognizes an existing customer, finds their contact person, and files the call summary in their record.
• The ERP: it can give the status of an order or a delivery without tying up the order desk.
• Email and chat: the call summary goes to the right person by email or Microsoft Teams.

These connections go through your software’s APIs: that’s what API integration is for. On the phone side, you keep your number: forwarding unanswered calls, or calls during certain hours, is usually enough.

And when requests come in writing rather than by phone, the same knowledge can power an AI chatbot for your website.

Disclosing AI. Since August 2, 2026, Article 50 of the EU AI Act requires telling callers they’re talking to an AI, clearly and at the latest at the first exchange. In practice, the agent introduces itself as an AI assistant in its very first sentence. It’s also a matter of trust: a caller who finds out afterward that they were talking to a machine feels deceived.

Recording calls. A recording contains personal data: the GDPR requires informing the caller, limiting how long it’s kept, and protecting the data. Most of the time, a summarized transcript is enough, without keeping the audio.

Cold calling. Outbound calls to consumers are strictly regulated. The voice agents I build take incoming calls and call back people who asked for it. I don’t build automated cold calling.

For a sensitive use case (health, banking data), you still need a lawyer’s advice.

Like a subscription AI phone answering service, an AI voice agent is priced in three parts. Know them before you compare offers.

• Setup: the greeting, call reasons, transfer rules, connections to the calendar and CRM, testing. At DigitalCX, it’s priced once and for all, in writing, before any configuration starts.
• Usage: telephony and the voice and AI models are billed per minute by their providers. These accounts are in your name, with no markup from me: you pay for what you use.
• Support: adjustments, new call reasons, quality checks on answers. It’s monthly and optional.

Watch out for packages that hide real usage or charge for every new rule. Always ask what happens if your call volume doubles, and who owns the number.

Ready-to-use voice agents are available by subscription. They fit when the need is standard: taking a message, giving business hours, booking in a common calendar app.

Having your voice agent custom-built makes sense when it needs to:

• check your CRM or ERP to recognize a customer or give an order’s status;
• follow rules specific to your business (urgency of a service call, service area, type of customer);
• transfer calls based on precise criteria, to the right person at the right time;
• stay entirely in your name, so you can change its settings or your provider without losing everything.

Either way, ask the same questions: does the agent say it’s an AI? What does it do when it doesn’t know? Where does call data go? Can you shut it off in one click? If the answers are vague, move on.