Why you have a call center, and what you're actually paying for
Before we talk money, let's agree on what you're actually paying for. A business phone line has exactly three jobs, and each one moves revenue in its own way.
- Sales. New customers and new revenue: growth for the business.
- Working the existing base. Upsells and repeat sales to people who already bought once. The most underrated of the three: the base is already paid for by past marketing, and every euro you pull from it raises lifetime value.
- Service. Retention: keeping customers satisfied, keeping service quality high, and keeping the base from drifting to competitors.
All three jobs share one thing in common: revenue happens in a single moment, when an agent is talking to a customer.
Conversation is the only thing that makes money. Every other minute on the line, the business pays for, but none of it generates revenue.
Cost per minute: the metric no CEO has in their head
Ask an owner what their call center costs, and they answer right away. Ask how many agents they have, and they answer that too. Ask the third question, how much does one minute of an agent actually talking to a customer cost, and there's almost always a pause.
That's not a criticism. It's the norm nearly everywhere, for a simple reason: nobody tracks this metric, because there's nowhere to pull it from.
We'll say it plainly: we didn't arrive at this metric right away either. We've run live agent teams for years, and only recently sat down and worked out what a minute of actual customer conversation costs us.
In-house call center or outsourcing: two different sets of headaches
Running your own call center means owning the full set of operational headaches, and each one costs money and nerves. You're constantly hiring, because turnover in this industry runs around 50% of headcount a year. You train people, and a trained agent leaves a few months later, taking that experience with them. You manage schedules, cover sick leave and vacation, and account for plain human factors: mood, fatigue, burnout. On top of all that: quality control and operational monitoring, without which the line drifts out of control.
Outsourcing takes these headaches off your plate and hands them to a vendor. It sounds like relief, and in part it is. But you pay roughly one and a half times more than it would cost to do the same work in-house, simply because for the vendor this is a business, and their margin is built into the price.
One problem neither model solves: capacity at peaks. When customers all call at once, you can't add agents instantly. You can't hire people in an hour, and an outsourcer's agents are already spread across other clients.
What actually makes up the cost of a call center agent
- Take-home pay: EUR 680 a month. This is what the agent actually receives. The math often gets mistaken for the final number, but it's only the starting point.
- Plus payroll tax and employer contributions: typically a 30-45% markup on top of take-home pay across most of Europe. At roughly 35%, EUR 680 becomes about EUR 918.
- Plus management overhead, about 20%: supervisors, quality control, training, and support sit above every agent, and that layer adds roughly a fifth on top. Total real cost: about EUR 1,100 a month.
So a 'EUR 680 salary' turns into about EUR 1,100 a month in real cost per agent.
Where an agent's time actually goes: half the minutes aren't conversation
An agent on the line and an agent in conversation are two different things, and the gap between them is enormous.
- Dialing through filters. Reaching a customer isn't simple anymore. Carriers run spam filters, and business calls increasingly get flagged and blocked. The agent sits and waits for the call to connect.
- Talking to answering machines. When a call does connect, the agent often lands on a smart answering system, now AI-powered too. The result is a robot talking to a robot, and the business still pays for it.
- Talking to the customer. The only part that makes money: selling, retaining, servicing. Everything else exists to make this possible.
- Wrap-up. Every call ends with data entry into the CRM. Necessary, but it has nothing to do with talking to the customer.
- Technical downtime. Waiting for the next call, switching between systems, small gaps between contacts: all of it is paid time on the line.
- Lunch and rest. People get tired. They need breaks and lunch, or call quality drops by the end of a shift. Normal, but it doesn't make the business any money.
Add it all up, and only about 50% of paid time actually goes to talking with customers.
You pay for 100% of the time on the line. Only half of it makes you money.
The real number: a minute of conversation costs about EUR 0.22
Here's the full math: take EUR 680 in take-home pay, multiply by roughly 1.35 to add payroll tax and employer contributions (a markup of 30-45% is typical across Europe), then by 1.2 for management and quality control overhead. That lands at about EUR 1,100 a month.
An agent works about 170 hours a month, roughly 10,000 minutes on the line.
If anyone bothers to calculate a per-minute cost at all, this is usually the number they land on. But that's the cost of presence on the line. The cost of conversation is a different number entirely.
Now divide that same EUR 1,100 by conversation minutes only, the 50% that actually make money. That's about 5,000 minutes. The cost of one minute of real conversation with a customer jumps to about EUR 0.22. Twice as much.
The number no CEO has in their head
It looks like EUR 0.11 a minute. In reality it's about EUR 0.22, twice as much, because half the time you're paying for, the agent isn't talking to a customer at all.
How many AI agents it takes to cover the peaks
If conversation is the only thing that makes money, it's logical to pay only for conversation. There's no reason to pay for silence, ringing, or answering machines.
Customers don't call on a schedule. They call at night, early morning, during lunch, in a spike right after an ad runs, whenever it suits them. The shift on duty at that moment might not be staffed at all: a live call center runs on a schedule.
One live agent talks to customers for about 5,000 minutes a month. So if you're short, say, 10 agents to cover daytime peaks, that's 10 x 5,000 = 50,000 minutes a month in AI agent terms.
Live agents work fixed hours and take the complex conversations that need a human touch.
When call volume spikes beyond the scheduled team's capacity, AI agents pick up the overflow instantly, no hiring or training required.
Outside business hours, AI agents keep answering, so a call never just hits an empty schedule.
Live agents and AI agents combine into a single hybrid line, each handling what it does best.
Per minute of real conversation, an AI agent runs at roughly a third of the cost of a live agent, about EUR 0.07 rather than EUR 0.22. The reason is simple: you stop paying for everything that never made you money in the first place.
What an AI agent can do that's hard for a live call center
- Holds the peaks. An AI agent scales instantly. When everyone calls at once, it absorbs the load and you don't lose calls. No hiring, no training for the spike.
- Replaces the old phone menu. Customers sit through thirty to sixty seconds of press one, press two, press three, and some just hang up. An AI agent greets them in natural speech, asks what they need, and routes them to the right agent itself.
- Works around the clock. Nights, weekends, holidays are ordinary hours for an AI agent. A customer's call never runs into a shift schedule.
- Gives management real quality control. A manual QA team typically manages to review 3-5% of calls, the rest is a blind spot. Our speech analytics product, Locator, listens to 100% of conversations and shows you what's actually happening in them.
- Helps train live agents. Our sales and service academy includes an AI-based training simulator: agents rehearse conversations while we help build the knowledge base and standards specific to your customers.
How we work: partnership, not boxed technology
Most players in this space sell you technology. The pitch is simple: we built the AI agents, take them, configure them, and figure out the rest yourself. It sounds convenient. In practice, it just hands the problem back to the client.
A business should focus on its business. It shouldn't have to become an AI configuration shop. Working out how to train an agent, on which scenarios, what it should say, and why it's saying the wrong thing: that's a separate competency most businesses don't have and shouldn't need.
We work differently. We don't sell a service with a fuzzy outcome. We become a partner invested in your business actually growing. We take ownership of the result: we get into your business, gather context, train the AI agents, and only put them on live calls once they hold a conversation with your customers at the level of a live agent, sometimes better.
Technology here is a tool we use. It's not the product we sell.
Where to start: run the tests before the treatment
It's tempting to just switch an AI agent on right away. Don't. Putting an untrained AI agent in front of your customers is the same mistake as putting an untrained new hire on the phones: unprofessional, and bad for the business.
For an agent to talk to customers correctly, it has to be trained, and that starts with gathering context.
So the first step is a business check-up, the equivalent of running tests before treatment. We connect Locator and look together at what's actually happening on your line.
Context comes in three layers, and there's usually a gap between them:
- What your customers actually say: what they're calling about, what's bothering them, what they want.
- What agents are supposed to say: the task as your marketing and your standards define it.
- What agents actually say: and this is where the real gap hides, the one you can't see without analytics.
A business check-up is like switching the lights on in a dark room. For the first time, you're looking at 100% of calls instead of a 3% sample.
You immediately see whether agents' actual speech matches what they're supposed to say. You immediately see how much time really goes to conversation. And once you know your real costs and your real conversation time, you get your actual cost per minute, based on your business's own data. The calculator was only ever a first estimate.
Only once the tests are back and the picture is clear do we gather full context, train the AI agents, and put them on live calls. That's the path to a hybrid model that actually works.