At one of Kinetara’s clients, the call data showed a man who called 35 times in a row. Every two seconds, because nobody picked up. The switchboard counted 35 unanswered calls. In reality it was one thing somebody wanted done.
I don’t know what he wanted. I don’t know whether he ever got it. I know he tried 35 times to reach someone about one thing. In any other channel, that much effort would be visible. On the phone it disappears into a statistic where every call counts the same.
Since then I’ve looked at call logs differently. Fifty calls a day means business is good. The question is how many of them exist only because the previous one failed.
My thesis: the value of a voice agent is not measured by whether it answers. It is measured by how fast the customer gets the thing they called about, and how much work it took. On both sides.
Answering is the easy part.
Taking a request is not resolving it
Let me put it this way. On Saturday a patient realises he can’t come in on Monday at ten. He calls. Kinetara answers. What happens in the next two minutes decides everything.
The agent can log the request and prepare it for reception. On Monday morning the staff know what needs to be done, and the patient doesn’t call again in the morning rush. If the agent is connected to the booking system and has the authority, it moves the appointment itself. The patient hangs up with a new slot.
Both have value. Not the same value. In the second case the customer got everything done; in the first, only half. And the first case created work: someone has to pick up the request, deal with it by a certain hour and tell the patient it’s done. If nobody owns that, the request sits in an inbox. On Monday the patient calls again. And we’re back to 35 calls, except now a robot takes the first few.
So the question for a manager isn’t whether the agent can move an appointment. It can. The question is who takes over the request, by when, and how the customer finds out it’s done.
Where the work went
Igor Nestorović, CCXP, published a piece on 7 October called “The Work Didn’t Disappear. It Moved.” He describes the first month of an AI assistant in his own customer support: it resolved around 80% of the tickets it was eligible for. Then he asks the question no dashboard can answer: what was left for the team?
Not 20% of the same work. What remained were the harder cases, the exceptions, reviewing errors, maintaining information. The queue gets shorter. The work gets denser.
On the phone you see this fast. In our analysis of dental calls, roughly 60% of conversations were about appointments. When the agent takes those, reception is left with the patient in pain, the complicated patient and the price question only one person knows the answer to. Plus something that wasn’t there before: reviewing the summaries the agent prepared, and a price list that says one thing on the website and another at the front desk.
Work that was nowhere before
Now for the part Nestorović didn’t write. In most of our deployments the agent answers outside working hours and when staff are busy. The call it takes would otherwise have gone unanswered. At one larger dental clinic we counted 138 calls an hour on a Monday between seven and eleven. The weekly average is 36. Same reception, same people. Nobody blames them for not picking up everything on a Monday. But those calls went somewhere. To a competitor. To another call at ten. Into 35 attempts.
Once the agent starts answering, the company hears for the first time what those people wanted. It expected relief. It got a list of tasks that existed nowhere before, because nobody had heard them.
Being reachable is not relief. Being reachable is an obligation.
Reception doesn’t get less work on Monday, it gets different work: tasks instead of silence. For the business that’s progress, because it’s finally reachable. For the organisation it’s a process that has to be in place before the agent answers for the first time. Not after.
Fourteen days for the agent, fifteen years for everything around it
At FrodX we’ve spent more than 15 years building the systems that hold a customer relationship together, from HubSpot to SAP Engagement Cloud. The whole time we’ve been answering the same three questions: what the customer needs right now, what the employee needs to know to handle it, and how to organise the work so the matter is actually resolved. Kinetara has more than 20,000 real calls behind it and it is the same work. It just starts on the phone.
That’s why most of the implementation time goes into things that aren’t AI. We write down the rules the experienced receptionist keeps in her head. We connect the agent to the booking system. We decide what it can resolve on its own and what it hands over. Where the agent has a clear task, the data and access to the system, it resolves the matter on its own: 100% of gas cylinder orders, around 90% of appointment confirmations (both for those tasks, in the cases we analysed, not for all calls). Where the service technician keeps the next three days’ schedule in his head, it can’t. We sort that out first.
Then we watch real calls: does the conversation start at all, where does it get stuck, does the customer get what they came for. Early on, 27% of callers hung up in the first fifteen seconds; that share keeps falling. The agent is up and running in fourteen days. Everything around it decides whether the man with 35 attempts will need just one next time.
If you get 50 calls a day
Don’t buy an agent because of that number. Spend a week looking at what’s inside it. Why people call. How many call a second time about the same thing. What the agent could resolve on its own, and what would land on your team afterwards. If the answer is “all of it, just by email now”, the agent solves nothing. It only moves the queue.
And measure differently. Not calls answered but matters resolved. Not response time but time to resolution. How many people called twice. How much work your team did after the conversation, not during it. That last number tells you whether you bought relief or just a friendlier intake for the same problem.
Somewhere in today’s call log is a person who called three times about one thing. I don’t know what they want. You could.
igor.pauletic@frodx.com
P.S. The numbers come from a conversation in which Gorazd Zakrajšek, after Kinetara’s first year in production, answers Andrej Bergant playing the sceptical buyer: who hung up, who asked for a human, and where it really got stuck. The full story after 20,000 calls is in episode 56 of the RASTezanja podcast.
