Managed AI voice agent: what it is and what sets it apart

If you’re looking at an AI voice agent for your business, you’ll have noticed that almost every website explains the same thing: what it is, how it works, how many calls it handles. None of them tells you what really decides whether it works or not: who sets it up, who corrects it when it gets something wrong, and what happens the day you change a price, a schedule, or a service. That’s the difference between an agent you build yourself and a managed one. And it’s not a sales nuance: they’re two different products with two very different workloads for you.

What an AI voice agent actually is

It’s a system that picks up the phone, understands what the caller says in natural language, and responds or acts accordingly: gives some information, collects details, books an appointment, screens the reason for the call, or passes it to the right person.

The difference from the voice menu of old is that it doesn’t make anyone press 1, 2, or 3. The customer says “I’m calling about my car’s service” and the agent knows where that goes. The difference from an answering machine is that the agent resolves things, rather than taking a message for someone to listen to tomorrow.

That said, it’s worth lowering expectations where appropriate. A voice agent doesn’t replace your sales team or negotiate a contract. It does the repetitive and structured well: opening hours, addresses, order status, availability, data collection, first-level qualification. When the conversation goes beyond that, what matters is that it knows how to hand over the call quickly, without the customer having to repeat everything.

The two ways to have one

1. The self-service model: they give you the platform, the work is yours

You sign up for a tool, log into a panel, and build the agent: you write the scripts, define what it should answer in each case, connect the calendar or CRM, record or generate the voices, test, and publish.

It’s the model of most international platforms and it has a real advantage: total control and a low license cost. The problem shows up after launch, when the agent starts receiving real calls.

Someone at your company has to listen to conversations, spot where it fails, rewrite the script, and test it again. And that’s not a one-week project: it’s a recurring task. If that person doesn’t exist or has ten other things on their plate, the agent stays frozen at the launch-day version and loses accuracy as your business changes.

2. The managed model: you provide the knowledge, the provider does the work

Here you explain how your company works and the provider builds, tests, tunes, and maintains the agent. You review and approve. When you change something (a schedule, a rate, a new service), you ask for it and it gets applied.

The trade-off is that the monthly fee is higher than a bare license, because you’re paying for people’s hours as well as technology. In return, you don’t need to hire or train anyone to maintain it.

The comparison, no frills

 

That last point is honest and worth keeping in mind: with the managed model, you depend on someone else’s response times. That’s why the question to ask before signing isn’t “how much does it cost?”, but “how long do you take to apply a change, and how many are included?”.

The cost that doesn’t show up in the price table

When you compare quotes, don’t compare fees. Compare the total of what it’ll cost you to have the agent working well six months from now. With the self-service model, there are four line items that almost never appear on the provider’s website:

• Setup hours. Someone at your company will spend days writing and testing scripts. Those hours have a cost, even if they don’t arrive as an invoice.
• Monthly maintenance. Reviewing conversations and correcting them is ongoing work, not a first-month extra.
• Integrations. Connecting the agent to your CRM or appointment software usually requires someone technical. Sometimes it’s simple; sometimes it isn’t at all.
• The cost of mistakes. A poorly tuned agent gives wrong information to real customers. That bill arrives in the form of a lost customer, and nobody records it.

Seven questions that reveal what you’re being sold

Take them to any sales meeting. The answers will tell you in two minutes which model you’re in:

1. Who writes the agent’s scripts, you or me? If the answer starts with “it’s really easy, with our editor…”, it’s self-service.
2. When the agent answers something wrong, who catches it? If nobody systematically reviews conversations, your customer will catch it.
3. How long do you take to apply a change, and how many are included in the fee? A “as many as you need” with no fine print is worth more than a discount.
4. What happens when the AI doesn’t know how to answer? There should be a clear route to a person, with the conversation context already collected.
5. Does it integrate with what I already use, and who connects it? The fact that the integration exists doesn’t mean they’ll be the ones setting it up.
6. Can I listen to and read every conversation? Without access to transcripts, you’re blind to your own inbound channel.
7. Is the legal notice that an AI is answering set up by default? It’s mandatory to inform the person on the line. If they leave it to you, the responsibility is yours too.

Who each model suits

Self-service makes sense if you have someone with the time and judgment to maintain it (a technical or operations profile with real hours available), if your operations are very stable, and if you want total control over the detail.

Managed makes sense if your team is already stretched, if your catalog or schedules change often, if nobody at the company wants to be “the bot person,” or if the phone is a sales channel and you can’t afford for it to answer badly for three weeks until someone notices.

For a Spanish SME of between five and fifty people, the second situation is the usual one. Not for lack of technical ability, but because nobody has those hours free.

Where this fits with the PBX you already have

A voice agent doesn’t replace your virtual PBX: it sits in front of it. The PBX keeps handling extensions, forwarding, recordings, and statistics. The agent decides which calls need to reach a person and which ones get resolved on their own.

That’s why the scenario where it shows most is the one you already know: peak hours, when four calls come in at once and there’s one person to handle them. There the agent doesn’t take work off your plate, it takes missed calls off your plate, which is different and more profitable.

And if your main inbound channel is written messaging and not calls, then what you need isn’t this, but an AI agent for chat channels. Automating the wrong channel is the most expensive way to solve nothing.

How to know if you need one before spending a single euro

Look at three data points from your PBX over the last month: how many calls were missed, which time slot they cluster in, and how many of the answered ones were repeat questions that didn’t need anyone with judgment.

If missed calls cluster in specific time slots and a good share of answered calls are repetitive, a voice agent will give you hours and customers back. If they’re spread out at random and each one is its own world, your problem isn’t automation: it’s organization or staffing, and no agent will fix that.

If you don’t have that data on hand, start there. At Glofera we do that analysis for free: we look at how your company’s calls come in today and tell you which part makes sense to automate and which doesn’t. No obligation, and no selling you an agent if you don’t need one.

Frequently asked questions

Does an AI voice agent sound like a robot?

Today’s synthetic voices are hard to tell apart from a person in a short conversation. Even so, European regulations require you to disclose that an automated system is answering, so the customer will know because you tell them, not because it sounds bad.

How long does it take to be up and running?

It depends on how many cases it needs to cover. An agent that screens and informs can be launched in a few days. One that books appointments and looks up data in your CRM needs more, because there are integrations involved and they have to be tested with real calls before opening it to the public.

What happens if the customer asks to speak to a person?

It should pass the call immediately with the context already collected, so whoever picks up doesn’t start from scratch. If a provider doesn’t show you how that handover works, it’s the clearest red flag there is.

Can I start small?

It’s the recommended way. Start with the stretch that hurts most (after hours, or peak hours) and expand once you’ve seen real conversations. Automating everything at once multiplies the points of failure just when you have the least judgment to spot them.

What if my company gets few calls?

Then you probably don’t need one yet. With low volume, good forwarding and well-configured schedules solve almost the same thing for far less money. Having someone tell you this is part of the service too.