Insight
What AI calling agents can and cannot do
AI calling agents handle real phone conversations in the caller’s language. They are good at some jobs, bad at others, and the difference matters more than the demo suggests.
Voice agents demo extremely well, which is exactly the problem. A demo is a call that goes as expected. The value is in what happens on the ones that do not.
What they are genuinely good at
Answering every call at peak instead of losing the overflow. This is the one that pays for itself: most businesses do not lose calls because their people are bad, they lose them because eleven rang at once.
Qualifying a lead the moment it arrives rather than the next morning. Speed to first contact moves conversion more than almost anything else in a sales process, and an agent does not sleep.
Booking, confirming, and chasing appointments. Bounded task, clear success condition, no judgment required.
Speaking the caller’s language. An agent can hold the conversation in English, Arabic, Urdu, or Hindi, and switch mid-call when the caller does — which in some markets is simply how people talk.
What they are bad at
Anything where being wrong is expensive and the rule is unclear. If you cannot write down how the decision gets made, an agent should not be making it.
Being the last line. An agent that cannot hand off gracefully is worse than a voicemail, because it wastes the caller’s time before failing them. Knowing when to transfer — on request, on frustration, on anything outside its remit — is most of the design work, not an afterthought.
Say that it is an AI
We build agents that disclose it when asked, and we recommend disclosing up front. It is the honest choice, and in a growing number of jurisdictions it is the legal one. It also costs you very little: people mind being deceived far more than they mind talking to a machine that is useful.