What they genuinely do well
Four things, all structural. They answer instantly, at any hour, which no staffed desk can promise. They answer every call at once, so a busy Friday produces no queue and no busy tone. They are consistent: the price quoted at 2am matches the one at 2pm, because both come from the same knowledge base. And they document everything: summary, transcript and booking for every call.
For the calls that dominate most small-business lines, bookings, prices, opening hours, directions, messages, this is the entire job. Software resolves those calls end to end, and the alternative for many businesses was never a receptionist: it was voicemail, which most callers refuse to use.
Where they honestly fall short
Judgment, first. An AI receptionist follows the rules and knowledge it is given; it does not read between lines, sense what a distressed caller is not saying, or improvise well outside its brief. Conversations that need those things need a person, which is why transfer rules exist and why some businesses rightly pay for human answering.
Speech recognition, second. It is good and improving, but strong accents, noisy lines and unusual names still cause mistakes. A caller who dislikes automated answers, third: a minority will hang up on any AI, however natural. If your clientele skews that way, weigh it seriously.
And configuration, last. An AI receptionist with a thin knowledge base gives thin answers. The businesses that get the most from Kalvo are the ones that spend twenty minutes telling it what a real receptionist would know.
How the risks are managed in practice
Three mechanisms carry most of the weight. Transfer rules put a person on the line when your chosen words or situations occur, so emergencies never sit with software. Approval mode holds outgoing replies for your one-tap sign-off, so the AI drafts but you decide. And transcripts mean nothing is a black box: every call can be audited afterwards, which is more than a phone answered by a rushed human offers.
The honest position is that an AI receptionist is not a replacement for human judgment. It is a replacement for the unanswered call, and it should be judged against that baseline: transcripts against voicemail, bookings against hang-ups. Our comparison with human answering services draws the line in detail.