AI voice agents and callbots: a business guide

An AI voice agent, often called a callbot, is a system that answers or places phone calls and holds a conversation in natural language. It turns speech into text, a large language model (LLM) decides what to say or which action to take, and the reply goes back down the line as synthetic speech. Unlike a phone menu, a callbot understands free-form sentences and can query your systems to actually resolve the call.

This guide covers how they work, when they beat an IVR, what to measure, and which EU and Spanish rules apply, including Spain’s new customer service law (Ley 10/2025).

How an AI voice agent works

A callbot is an AI agent with a demanding channel: the phone. It has five building blocks:

1. Telephony. The call arrives on a phone number through a VoIP provider, usually a SIP trunk or a cloud voice platform. This layer streams the audio in real time to the rest of the system.
2. Speech-to-text. A speech recognition model transcribes what the caller says while they speak. It has to detect when the caller has finished and cope with background noise, accents and half-finished sentences.
3. An LLM with tools. The language model interprets the request and decides the next step. When it needs data, it calls a tool: check the calendar, look up an order, book an appointment or log an incident. Tools are what connect the callbot to your systems.
4. Text-to-speech. The reply is converted into synthetic speech and played back. Callers must be able to interrupt the callbot, just as they would interrupt a person.
5. Handoff to a human. When the callbot can’t resolve the call, spots a sensitive case or the caller asks for it, it transfers the call to a human agent along with the context: who is calling, what they want and what they have already said.

The system must also log every call: transcript, actions taken and transfers.

The hardest problem is latency: on the phone, a long pause feels awkward. Every block adds time, so the whole pipeline has to stream.

Callbot vs IVR vs human agent

The practical takeaway: a callbot doesn’t replace your team. It filters and resolves the repetitive part. Complex calls, complaints and callers who ask for a person should reach a human. In Spain, for many companies, that is now a legal requirement (more on this below).

Use cases for a virtual phone assistant

A callbot works well when the call has a clear goal and the answer depends on data that already lives in one of your systems:

• Appointments and bookings. Check availability, book, reschedule or cancel. This is the most straightforward case because the action is specific and easy to verify.
• FAQs. Opening hours, locations, required paperwork, service terms. If the information lives in documents, the callbot can answer with RAG over your company documents instead of making things up.
• Order, shipment or booking status. The caller gives a reference number, the callbot queries the system and answers.
• After-hours triage. Capture the reason for the call, rate its urgency, log the incident and alert the on-call team if needed. The next morning, every case is already in the system.
• Outbound reminders. Call to remind someone of an appointment and let them confirm, reschedule or cancel on the same call. Specific rules apply to automated calls; see the legal section.

Don’t start with complex complaints, decisions about people or conversations where a mistake has serious consequences.

What we have built at LCApps

At LCApps we have built callbots that answer phone calls: they turn speech into text, decide with an LLM connected to the business’s data and reply with synthetic speech. Every call is logged, and callers hear a data protection notice at the start.

We also run an asynchronous audio transcription service that turns voice recordings into text and notifies the requesting system when it is done, which is useful for analysing past calls.

We design and run these projects in line with the GDPR and the EU AI Act.

What to measure

Without metrics, you won’t know whether the callbot helps callers or just gets in their way. Track these from day one:

There is no universal “good” containment rate; it depends on the use case. Measure a baseline before the pilot and compare against your own numbers. High containment with lots of abandoned calls is not a success.

Legal requirements for callbots in the EU and Spain

A callbot talks to people, processes personal data and runs over the phone network, so several sets of rules apply at once. For the full picture on the AI Act, see our guide to EU AI Act obligations for businesses.

Telling callers they are talking to AI (AI Act, Article 50)

Article 50(1) of Regulation (EU) 2024/1689 requires providers to design AI systems that interact with people so that those people are informed they are dealing with an AI system, unless that is obvious. Article 50(5) adds that this must be done clearly and at the latest at the first interaction.

Good synthetic voices can sound human, so don’t assume it’s obvious: say it at the start of the call. According to the European Commission, the transparency rules apply from August 2026. The amendment known as the “AI Omnibus”, in force since 27 July 2026, pushed back the high-risk obligations but not this disclosure duty.

Privacy notice at the start of the call (GDPR, Article 13)

Callers share their voice, their phone number and whatever they say. Article 13 of the GDPR requires you to inform them when you collect that data. You can’t read out a full privacy policy on a call, but Spain’s data protection act (LOPDGDD, Article 11) allows a layered approach: give the basic information (who the controller is, the purpose, and that the caller can exercise their rights) and point to where the rest can be found, such as a web address.

A typical opening line covers both duties: “Hi, this is [company]’s virtual assistant. This call is handled by an AI system and is logged to manage your request. Our privacy notice is at [website]. If you’d rather speak to a person, just say so at any time.”

Other GDPR points to check:

• Recordings and transcripts. Decide why you keep them, for how long and who can access them. Keep only what you need.
• Vendors. Speech-to-text, LLM and text-to-speech providers are processors: you need an Article 28 GDPR contract and you need to know where they process data.

Outbound and marketing calls (Spain’s Telecoms Act, Article 66)

Article 66(1) of Spain’s General Telecommunications Act (Ley 11/2022) grants two rights that directly affect a callbot placing calls:

• (a) Not to receive automated calls without human intervention for marketing purposes without prior consent.
• (b) Not to receive unwanted marketing calls, unless there is prior consent or another lawful basis under Article 6(1) GDPR.

In practice, an outbound marketing callbot needs the person’s prior consent. A reminder about an appointment the customer booked is not a marketing call, but don’t turn it into a sales pitch halfway through.

Spain’s customer service law (Ley 10/2025)

Law 10/2025 of 26 December regulating customer service (Official State Gazette no. 312, 27 December 2025) has been passed and has been in force since 28 December 2025. Companies within its scope have twelve months to adapt, until 28 December 2026 (sole transitional provision).

It covers every company providing basic services of general interest (water, gas, electricity, transport, postal services, electronic communications and financial services). In other sectors, it covers companies that exceed at least one of these thresholds: 250 employees, turnover above €50 million or a balance sheet above €43 million (Article 2). It applies to companies established in Spain or in another country when they provide those services.

For callbots, the key provision is Article 8:

• It bans automated answering systems or similar tools as the only channel for customer service.
• Automated answering systems and chatbots must let customers request personal attention at any point in the interaction, from the very start. The law defines personal attention as a human operator answering in real time.
• 95% of those requests must be answered, on average, in under three minutes.
• Companies may not end a call because of a long wait.

The law does not ban AI. Its preamble explicitly allows AI systems that make communication easier. What it requires is that the callbot is a door to a person, not a wall. Article 12 also states that when a complaint or incident is raised by phone and the customer gives express consent, the company must record the call and keep the recording or its transcript until the resolution is notified.

A five-step pilot plan

1. Pick a narrow use case. One type of call with volume and a clear action: booking appointments, checking a status or logging after-hours incidents.
2. Measure the baseline. Monthly calls, average duration, missed calls and staff hours spent. Without this you have nothing to compare against.
3. Design the script and the limits. Opening message (AI disclosure, privacy notice, option to reach a person), which tools the callbot can use, which cases always go to a human and which data it must never ask for.
4. Integrate and test. Connect the callbot to your calendar, CRM or relevant system. If your systems lack an API, that’s a systems integration project in its own right. Test with real calls from your team: accents, noise, interruptions and out-of-scope questions.
5. Launch on a share of traffic and review. Listen to or read a sample of calls every week, fix what fails and decide, based on data, whether to scale up, adjust or stop.

If the callbot needs several integrations, the cost behaves like any other software project; we break it down in custom software development cost.

Frequently asked questions

Can a callbot replace my phone support team?

Not entirely, and in Spain many companies are not allowed to: Ley 10/2025 bans automated systems as the only customer service channel. The callbot handles repetitive calls and gathers context; your team handles the cases that need a person.

Do I have to tell callers they are talking to an AI?

Yes. Article 50 of the AI Act requires informing people that they are interacting with an AI system, at the latest at the first interaction. With a natural-sounding synthetic voice it isn’t obvious, so say it in the opening message.

Can I use a callbot for sales calls?

Only with the person’s prior consent. In Spain, Article 66(1)(a) of the Telecoms Act grants the right not to receive automated marketing calls without human intervention unless the person has consented beforehand.

What about recordings and transcripts?

They are personal data. Define the purpose, retention period and who can access them, and mention them at the start of the call.

Conclusion

An AI voice agent pays off when it resolves specific calls using real data and hands over to a person as soon as needed. Disclose the AI and the data processing at the start, limit what it can do, track containment and transfers, and check whether Ley 10/2025 applies to you before 28 December 2026.

At LCApps we build callbots, multichannel chatbots and transcription services connected to the systems you already use, designed in line with the GDPR and the EU AI Act. Tell us which process you want to improve and we will help you design a realistic pilot.