An AI voicebot as the first line: calling back every enquiry while it's still warm

When someone fills in a form on a website or under an advert, their interest has a short half-life. They have usually contacted two or three companies at once, and the first one to call back often wins. Yet in many companies the callback waits for a free moment in someone's day, and enquiries that arrive in the evening wait until the next morning.

One of our clients receives a steady stream of enquiries, most of them through online forms. Its team could not call everyone back quickly enough, especially after office hours. So we built an AI voicebot that calls every new enquiry back immediately, while the person is still interested. It is one of several voicebots we have built, and a good example of what we mean by AI in processes rather than in products: no new app for anyone to learn, just a step in an existing process that now happens on time.

The problem was time, not enquiries

The client did not lack enquiries. It lacked hours. A large part of the team's day went on dialling, reaching voicemail, trying again and repeating the same first sentences.

The first call is highly repetitive: confirm the enquiry, check the contact details, answer basic questions about the offer and say what happens next. That makes it a good fit for a voicebot, provided the bot does exactly that and nothing more.

How the callback works

1. A person fills in an online form.
2. An automation passes the enquiry to the CRM the team works in, together with the context: the offer the person responded to and its language.
3. The new record creates a task for the voicebot, which calls straight away, within the agreed calling hours.
4. The bot introduces itself as an AI assistant calling on behalf of the company, says the call is recorded, and talks in the language of the offer. If the person prefers another language, it switches.
5. It confirms the details, answers questions about the offer, and says who will be in touch and when.
6. A summary of the call goes back to the record in the CRM, and a team member gets a task to follow up, with the context already in front of them.

The bot does not assess anyone and does not decide anything. Every enquiry it handles goes on to a person. It takes over the first, repetitive contact, so that the team spends its time on conversations that need a human.

GDPR and the AI Act from day one

A voicebot talks to people and processes their personal data, so the legal side was part of the project, not an add-on. We treated it as part of the project and reviewed every point with the client before regular operation:

• People know they are talking to AI. The bot says so at the start of every call. It is a matter of trust, and since August 2026 also a transparency obligation under the AI Act.
• A separate privacy notice for the bot.
• A clear legal basis. The call is a step the person asked for by sending the enquiry (Art. 6(1)(b) GDPR); the recording rests on legitimate interest (Art. 6(1)(f)). If the person objects to the recording, the bot ends the call and a team member takes over.
• No profiling, no automated decisions. The bot confirms details and answers questions; every decision stays with people.
• The CRM is the record. The summary and the answers live in the CRM, under the client's retention rules. On the voice platform, recordings and transcripts are kept only for a short operational window and then deleted.

From test calls to daily operation

We rolled the bot out in steps: first test calls with the client's team, some of whom did their best to throw it off topic; then a short evening pilot on real enquiries; then regular operation outside the team's working hours. A few things mattered more than we expected:

• The bot is only as good as its context. Some forms sent incomplete details to the CRM. We fixed the forms, not the bot.
• Fewer questions, not more. An early version asked one question too many, which for some offers made no sense. We removed it and left it to the team.
• Short calls. After a few minutes the bot sums up and closes the conversation.
• Graceful exits. "Call me later", "I am no longer interested" and a missing surname in the form must be handled as calmly as the main path.

One of several voicebots

Callbacks are only one place where a voice agent helps. We also build multilingual first-line support agents that answer routine questions in the caller's language, route urgent cases on a separate path and open a ticket in the CRM for everything else. Two lessons repeat: keep the knowledge base small and clean, and do not assume the language from a country code; in multilingual countries the bot has to ask.

The telephony side draws on years of experience, going back to our VoIP work for mobile platforms.

The effect

Every enquiry now gets a call back straight away, evenings and weekends included, not only when someone finds a free moment. And the first contact is consistent: the same information and the same care with personal data, every time.

If you have a process where the speed of first contact matters, read more about AI process automation or book a conversation with us.