Comparison

AI assistant or answering service: how to choose

Both pick up when you cannot. They do not cost the same, do not work the same hours and do not fail in the same ways.

Both answer when you cannot. From there, everything differs: the cost, the hours, what happens when five people call at the same time, and the way each of them gets things wrong.

We sell one of the two, which gives you a good reason to check what follows. Every figure links to the page we read it on, including the ones that do not suit us.

Replace, or split the calls?

“Should the answering service be replaced by an AI” is a badly framed question, and the industry data shows it. In a survey of 321 customer service leaders published in April 2026, Gartner reports that 85% are expanding the responsibilities of their human teams, and that only 31% have made or are planning AI-related job cuts on the front line.

“The real advantage comes from combining AI efficiency with human judgment, empathy and experience to deliver outcomes that technology alone cannot.”
Eric KellerSenior Director Analyst, Gartner, April 2026Source

What gets decided, then, is which calls go where. And the answer differs for a garage and for a law firm.

Hours: the difference you cannot make up

This is the most concrete point, and the least discussed. Telib publishes availability from Monday to Friday, 08:00 to 19:00. MEDES publishes 5 days a week, 08:00 to 19:00, which is 55 hours. Those are good ranges for a staffed service. It still remains that nobody picks up at 19:30, nor on a Sunday, nor on 1 August.

AI assistant168 h/week
MEDES (5 days, 08:00 to 19:00)55 h/week
Telib (08:00 to 19:00, 5 days)55 h/week
Hours covered per week. Ranges published by the providers in August 2026, compared with continuous coverage. Public pages of Telib and MEDES

Does that matter to you? It depends entirely on your line of business. For a fiduciary, very little: clients call during office hours. For a garage, a restaurant or a trade business, it is often the whole problem: people call you on the way home from work, on a Saturday, or when something breaks.

Price, at comparable volume

You have to be precise here, because the comparison is easy to rig. The rates recorded in August 2026: Telib runs from CHF 65 to CHF 649 a month incl. VAT, the entry package covering around fifty calls; Secrétaire 24 advertises from CHF 490 a month. Our plans run from CHF 99 to CHF 599 a month excl. VAT, for 500 to 3,000 minutes.

At very low volume the human option wins, and it should be said plainly: if you get two calls a day, CHF 65 a month for someone to genuinely pick up is a good deal. The gap widens with volume, because a human subscription is billed per call or per bundle of calls, whereas an AI assistant is billed per minute and does not cost more because three people call at the same time.

The comparison page published by MEDES in fact puts the alternatives at CHF 8 per call when you handle them yourself and CHF 6 per call with in-house staff. The pricing detail is in our article on costs.

The comparison, criterion by criterion

Comparison on the criteria that actually decide it. Ranges and rates recorded in August 2026.
Outsourced answering serviceAI assistant
HoursAround 55 h a week24 hours a day, all year
Simultaneous callsDepends on who is freeSeveral in parallel, no waiting
LanguagesDepends on who is on dutyDetected automatically
Complaint, disputeClearly betterTo be transferred to a person
Judgement, goodwill gestureCan decide, or play for timeDoes not decide for you
Knowing a regular customerRecognises, remembers, adaptsDepends on what you gave it
Cost at high volumeGrows with the number of callsGrows with the minutes, more slowly
Written record of every callDepends on the providerSummary by email, every time
Comparison on the criteria that actually decide it. Ranges and rates recorded in August 2026.

Where a human answering service is still better

None of the situations below is an edge case.

  • The complaint. An unhappy customer wants to be heard by someone who can do something about it. An AI that handles a complaint by giving the right information often makes things worse: information is not what the person is looking for. It is also the situation where Swiss law is strictest about recording, as we explain in our article on the revised FADP.
  • Judgement. Moving an appointment for a long-standing customer, making a goodwill gesture, deciding that a case is urgent when the grid says otherwise. An AI should not settle that, and we do not give it the power to.
  • The customer who does not want a machine. They exist, they are not wrong, and the right answer is to give them a person straight away, not to talk them round.
  • Complex triage. In a medical practice, telling what counts as an emergency takes a skill that does not fit into a script.

That lines up with what Gartner measures on the customer side: in a survey of 5,801 consumers in the United States in early 2025, 54% trust a person more than an AI for a product or service recommendation, against 32% the other way round. The figure is American, not Swiss.

Where AI clearly wins

  • Off-hours. Evenings, weekends, holiday periods. This is the main argument, and it is structural.
  • Peaks. A property management office on a Monday morning, a garage after the first frost: fifteen people call within the same hour. An AI assistant takes them all at the same time.
  • The repetitive calls. Opening hours, the address, the prices, availability. Often half the calls, and nobody enjoys handling them. We set out elsewhere what a voice assistant can do with a call, task by task.
  • The written record. A summary of every call, every time, without anyone having to write it.

The useful distinction is not human against machine, it is automating against augmenting, and the Stanford economist Erik Brynjolfsson puts it better than we do:

“The distributive effects of AI depend on whether it is primarily used to augment human labor or automate it. When AI augments human capabilities, enabling people to do things they never could before, then humans and machines are complements.”
Erik BrynjolfssonStanford Digital Economy Lab, Dædalus, 2022Source

The setup that works: both

In practice, the businesses that get the most out of it do not choose. They let the assistant take every call, handle what is simple, and pass the rest to a person, saying what it is about. The answering service does not disappear: it stops picking up to give the opening hours.

Concretely, it comes down to a few lines: the assistant always answers; it transfers immediately if the person asks or if the subject is a complaint; it sends a written summary in every case. The rest is tuning.

If you want to put a figure on what that gives at your business, our calculator works from your own numbers. If you would rather understand the market first, the comparison of providers and the buyer's guide cover the four possible options, including the ones that are neither.

Sources

  1. telib.ch, public rates and hours (recorded in August 2026)
  2. medes.ch, published availability (recorded in August 2026)
  3. medes.ch, comparison of medical phone answering rates
  4. secretaire24.ch (recorded in August 2026)
  5. Gartner, press release of 28 April 2026 on customer service teams
  6. Erik Brynjolfsson, “The Turing Trap”, Dædalus 151(2), 2022
  7. localsearch and Hochschule Luzern, “KMU Digital Pulse 2025”

Hear Voko answer a call

Fifteen minutes, your own questions, and a clear quote at the end.

See pricing