What an AI voice agent is
An AI voice agent is a system that holds a phone conversation: it listens to the person on the line, understands what they want, answers from your knowledge base and starts the next step itself after the call. It is also often called a voice assistant for phone calls. Unlike an automated menu where callers press keys, you talk to it naturally, and the person on the line can interrupt it just as they would a human.
We build voice agents on CallVista, our platform for call centres, which runs in production. You choose the agent's name, voice, speaking style and greeting. The agent works in Croatian, English, German, Italian and almost any other language, and we test it on your own calls before launch.
What the agent does on the phone
Inbound calls
The agent picks up calls to the number you choose, answers questions about products, prices and terms, collects the details of the enquiry and notes what the caller wants. When the same person calls again, the agent knows what was discussed before, because CallVista builds up the client profile across all calls.
Outbound calls
The agent calls contacts from your campaign or those who asked for a callback, presents the offer and answers questions. It calls within the hours you set or at a time the person arranged themselves, recognises voicemail and calls again later, and when someone says they do not want to be called again, it flags their number and does not dial it again.
Knowledge base
You upload price lists, terms, instructions and FAQs as PDF, Word or CSV documents, and the agent answers from them during the conversation. Before go-live, we check the answers against real questions from your customers and fill gaps in the knowledge base wherever the agent has no reliable answer.
When a person takes over the conversation
When the caller asks for a person or the question goes beyond the scope the agent was given, the agent records the request and the reason, politely ends its part of the conversation, and the call is flagged on the dashboard for someone to take over. Your employee sees the summary and the transcript, so they get back in touch with the full context and do not question the caller all over again. This is what handover to a live agent looks like in CallVista.
After the call: summary, tags, quote and callback
- A summary of every conversation, with the caller's intent and the outcome of the call.
- A sentiment score and topic tags, which you use to filter calls and see what people ask for most often.
- A quote by email: when the person shows interest, the agent identifies the requested products from the conversation, and right after the call the system sends a quote with details from the knowledge base on its own. A link in the email takes the customer straight to the purchase.
- A callback: when the person asks to be called later, the agent confirms the day and time, and the call is made automatically at the agreed time.
- A client profile that is updated with every conversation.
You see all of this on the dashboard. If the summaries also need to reach your CRM, we build that connection as part of the project. Read more on our AI integration with CRM and ERP systems page.
AI for call centres: what the agent takes on and what stays with people
AI for a call centre does not mean a call centre without people. The agent takes on the calls that repeat: questions about products, prices and terms, collecting enquiries, reminders and callbacks. Your team stays in charge of complaints, negotiations and sensitive conversations, and the summaries show them what happened before they take over. How the same approach works beyond the phone, in email, the CRM and the ERP, is described on our AI agents for companies page.
Telling people about AI and call recording
Since 2 August 2026, Article 50 of the Artificial Intelligence Act (EU AI Act) has required that a person be informed, at the latest at the first interaction, that they are talking to an AI system, unless this is obvious. That is why, in CallVista, the agent says in its greeting that this is an automated call. You can set the wording of that statement word for word, and when asked directly, the agent always answers honestly that it is an automated assistant.
Call recording can be switched on or off. If calls are recorded, the person must hear this at the start of the conversation, and you need a legal basis for recording under the GDPR, a notice on the processing of personal data and a set retention period for recordings and transcripts. We agree the wording of the notice with you and your data protection officer or lawyer. There is more on our EU AI Act compliance and AI security and governance pages.
Consent for automated sales calls
The Croatian Electronic Communications Act allows automated calling systems without human intervention to be used for direct marketing and sales only with the end user's prior consent. That is why we launch outbound sales campaigns only to contacts for whom you have such consent, and before launch we check with you where the consent for each list comes from. A callback the customer asked for themselves is recorded by the agent with the agreed time, and anyone who says they do not want to be called again receives no more calls.
What to measure
From day one, we track these measures:
- the number of inbound and outbound calls;
- average call duration;
- the share of calls resolved without a handover, and the reasons for handovers;
- conversion rate: quotes sent and responses to them;
- callbacks made compared with callbacks scheduled;
- the distribution of sentiment and the most frequent call tags;
- a quality score for each conversation.
How we introduce a voice agent
- A free first call: which calls the agent takes on, inbound, outbound or both.
- Scope, cost estimate and a phased plan.
- The knowledge base and the agent's name, voice and greeting, with the notice about AI and recording.
- A first working version in 4–8 weeks and test calls with your team.
- Go-live, with conversations monitored and the knowledge base extended.
- Maintenance and new types of call, with progress tracked in the client portal.
What the price depends on
- the number and length of calls, which determine telephony and language model costs;
- the types of call and the scenarios the agent handles;
- the size of the knowledge base;
- connections to your CRM, calendar or email;
- requirements for recording and data retention.
What you need to prepare
- documents for the knowledge base: price lists, terms and FAQs in PDF, Word or CSV format;
- a decision on the number the agent answers or calls from;
- a list of contacts with consent, if you plan outbound sales calls;
- a person or team to take over the calls the agent hands over;
- the wording of the notice about AI and recording, agreed with your lawyer.