On Tuesday morning I got an email. A friendly, precise, complete answer to the question I had sent on Saturday evening. I needed it on Sunday at seven in the morning.
My wife and I spent last weekend in the Julian Alps. We slept at the Erjavčeva koča hut on the Vršič pass, climbed the Hanza route up Mala Mojstrovka on Saturday (the larches aren’t yellow yet) and planned a gentler Sunday: below Jalovec to the shelter under Špiček and down into Trenta. A free hikers’ bus was supposed to take us back up to Vršič; the municipalities of Triglav National Park run it in summer to keep cars off the roads. The season was ending and I didn’t know whether the summer timetable still applied. I needed one piece of information: does the 17:12 bus from hairpin 49 in Trenta still go up to Vršič?
On Saturday evening I called Nomago, which I found listed online as the operator. Nobody answered. I sent an email. Then I went to bed, and on Sunday I walked down to Trenta without an answer.
I now tell this story in sales meetings where I present voice and email agents. Contact centre managers usually have two objections. “Our customers won’t want to talk to a robot.” And then, more quietly: “What happens to my team?”
The first objection is too broad. The second touches something real, so I’ll come to it separately.
When I ask whether the 17:12 bus is running, I care more about getting a reliable answer in time than about who gives it. The same goes for a Thursday appointment or a parcel in transit. For questions like these, an empathetic human two days late is not a better experience than a machine on time. It’s a worse one, because friendliness without an answer is still a delay, just with a nicer tone.
The contact centre managers I sit down with don’t have bad teams. They have good people who know the answers. Those people work eight hours. The customer also calls at nine on a Saturday night. The limit isn’t knowledge. The limit is the clock.
A metallic voice would have been enough, one that said it was an automated system and then: “Sunday, 27 September, Trenta, hairpin 49, 17:12, the bus is running.” No emotion, no courtesies, and a date, so I’d know the answer wasn’t from July. Or an automatic email with the same sentence three minutes after mine. I’d have known I was talking to a machine, and I’d have got what I came for: confirmation.
There’s a condition that columns about AI like to skip. My call went to Nomago, the bus was Arriva’s, and the timetable was changing with the end of the season. A machine that confidently read me the July timetable on Saturday night would have been worse than silence, because it would have left me standing on a hairpin bend on Sunday. Availability alone solves nothing. The agent has to be connected to the data that is valid right now. And someone in the company has to answer for that data being accurate and current. That is harder work than choosing a voice. And it’s work nobody can do without the people in the contact centre.
At FrodX, we spent ten months with Kinetara listening to what happens when the phone rings in a clinic. Across four dental and medical institutions on nine sites, the voice agent answered 3,453 calls. In 1,836 of them the patient spoke; nearly half the calls ended without the caller saying a word. All the numbers below come from those 1,836 conversations; the methodology is in our analysis Klici, ki jih nihče ni slišal (in Slovenian).
Three findings. 27.7% of conversations took place outside working hours; for urgent cases involving pain, swelling or bleeding, the share is 42.7%. Monday between 7 and 11 a.m. brings 3.8 times more calls per hour than the average of all other hours. And 24 patients out of 1,836 asked for a live person during the conversation. 1.3%.
We didn’t measure satisfaction. Those who never spoke aren’t in this number. It says only that, of the people who started talking to the machine, 98.7% did not explicitly ask for a human. A patient with a toothache at 6:52 on a Monday morning isn’t choosing between a human and a machine. They’re choosing between an answer and an unanswered phone.
The second objection is more serious because it’s personal, and I won’t smooth it over. Part of the work will change, and part of it may disappear. Where the job was repeating the same sentence, the team may get smaller; no slogan about new roles changes that.
What remains is work someone has to take responsibility for: listening to transcripts, fixing the answer that confused a caller, deciding where the machine stops and the human starts, and making sure the agent has the right timetable. In our analysis, the voice agent is only the sixth measure. Before it come five organisational ones, from measuring missed calls to a separate line for urgent cases; they don’t require a voice agent, but they do require time and work. If those five solve your problem, you don’t need an agent. If they don’t, the question isn’t whether one more friendly person on an evening shift would solve it, but whether that kind of organisation makes sense and can last.
On Sunday at 17:12 the bus came. The driver, from Serbia, spent the whole ride up to Vršič telling me how he describes Slovenian nightlife to his friends back home: like theirs during Covid, except nobody has announced it. So much for cultural differences. No robot will give me that conversation, and I wouldn’t want one to try.
But I got on the bus because it came at 17:12, not because the driver was friendly. The conversation was the reward; the answer was the precondition.
So the question for contact centre managers isn’t whether customers will accept a robot. The question is how many of your customers are waiting for Tuesday right now.
P.S. You can hear what such a conversation sounds like for yourself. Trial conversations are available for seven industries: dental, medical, car service, service businesses, retail, leisure and utilities.