Voice Agents: When Your Brand Picks Up the Phone

A few weeks ago I called one of our own voice agents. Not to test it, but because I actually needed an answer. Within seconds I had forgotten there was software on the other end. A moment later I had what I needed, delivered politely, with no hold music. And then I sat there, phone in hand, thinking: that was a real conversation. With a machine. On the phone.

I have spent half my career giving brands a voice. As Germany's first official social media manager at Mediengruppe RTL, I was responsible for more than a hundred profiles with over twenty million fans, all of it in writing. Today, at BE BRAVE and ORION AI, I build systems where the voice is no longer a metaphor. The agent actually speaks, in real time, with interruptions and dialects on the other end. That changes the rules.

What real-time speech can do today

Real time has never let go of me. Back at RTL, I helped realize the first professional Facebook livestream of a Klitschko boxing match with Vidpresso. But a livestream is a one-way street with a chat next to it. A phone call is dialogue, and that is exactly where the technology has now arrived. The delay between question and answer sits within a normal conversational pause. You can interrupt the agent, and it stops talking and responds to your objection.

It understands half sentences, casual speech and people who think while they talk. That sounds trivial, but it is exactly the difference between a phone menu and a conversation. We run agents like this in production on the phone, as part of more than thirty AI products we operate live. The most important lesson from that is unspectacular: the agent does not need to be a genius. It needs to be reliable, stay polite and know when to hand over.

Use cases that actually pay off

The most interesting deployments are rarely the spectacular ones. They are the calls that either go unanswered today or pull a team away from more important work. Spend one week counting which calls actually come in and how many of them are routine, and the candidates almost pick themselves. A few patterns have proven themselves in our projects:

• After-hours availability: the agent picks up when nobody is in the office anymore, and no caller ends up stuck on hold.
• First-level triage: what is it about, how urgent, who needs to handle it? The team starts the day with sorted requests instead of a full voicemail box.
• Appointments: booking, moving, confirming. Unspectacular, but a huge share of call volume in many businesses.
• Standard questions: opening hours, prices, order status. The same answers every time, which no human should have to give for the hundredth time.
• Peak load: when everyone calls at once, the agent catches the overflow instead of a busy signal driving customers away.

Brand voice: the phone is your most intimate channel

At Phantasialand, where I managed digital communication, I learned how much tonality carries a brand. Around two million visitors a year, and every reply online had to feel like the park itself: warm, a little playful, never condescending. On the phone this matters twice as much. A voice transmits warmth or coldness faster than any text. If your voice agent sounds like a government office, you have misunderstood the channel.

That is why every voice project of ours starts with language, not technology. Which words does the brand use, which ones never? How formal is it? What does a no sound like that still feels friendly? From this we build guardrails the agent must not cross. The voice itself gets chosen the way radio stations used to cast a presenter: it has to fit the brand, not the current tech trend.

Limits: where a human has to take over

A voice agent does not replace people, it is a filter in front of them. Emotional situations do not belong in its hands: the furious complaint, the bereavement, the complex edge case with a long history. Good systems recognize those moments and hand over cleanly to a human, including a summary, so nobody has to tell the whole story twice. The worst mistake is an agent that insists on solving everything itself.

There are also hard limits on content. The agent must never promise anything the company cannot keep, and never invent an answer when it does not know one. You do not get there through hope, but through tight connections to real data and clear rules about what sits outside its playing field. An honest 'let me check that with a colleague and call you back' beats any confident piece of misinformation.

Acceptance and transparency: say what is going on

My position is clear: the agent introduces itself as an AI in its very first sentence. Not sheepishly, but as a matter of course. In the EU, the AI Act already requires that people are told when they are talking to a machine, and in Switzerland that expectation is growing too. Honestly, you should not need a law for this. If you deceive your customers, you gamble away exactly the trust a brand wants to build on the phone.

The surprising part from practice: acceptance is higher than many people fear. Callers do not hang up because an AI picks up. They hang up when they do not get help. An agent that answers immediately, understands the request and delivers within a few minutes beats any hold loop with piano music. Transparency plus usefulness, that is the whole formula for acceptance.

Start with one single reason people call

If you are considering this: do not look for the big prestige project, look for the one reason people call that costs your team the most time today. Build an agent for that single case, one that says honestly what it is, speaks your brand's language and hands over to humans when in doubt. Measure for four weeks what it solves and where it fails, and then decide on the next step.

And if you would like to hear what such a conversation sounds like before you commit, feel free to get in touch. I will show you live what works today, and just as openly what does not work yet. Until then, one question to take with you: which call in your company would be the first one you would trust to a machine with a clear conscience?