We build AI voice receptionists for elevator companies, which means we have a financial incentive to tell you they can do everything. They cannot. Here is the list we would want if we were the ones buying, split into what holds up on real calls today and what still belongs to humans.
What they genuinely do well
- Answer instantly, always. Two rings, at 2 PM or 2 AM, on Christmas. No hold music, no "your call is important to us." This is the core of the value and it is not close.
- Follow a written procedure exactly. An entrapment call runs your protocol: confirm whether anyone is in the car, capture building, car, and floor, connect the on-call tech. The agent never skips a step because it is tired or flustered, which is more than can be said for most of us at 3 AM.
- Collect details without forgetting any. Building address, unit count, access instructions, contact on site. Collected once, never asked twice, spelled correctly in the ticket.
- Book real appointments. Connected to a calendar, the agent reads actual availability and books the visit during the call, then texts the caller a confirmation before they hang up.
- Log everything. Every call becomes a transcript and a log entry. Offices that have never had complete call records suddenly have them.
- Recognize repeat callers. A property manager who called last week gets treated like one, not asked to introduce herself again.
- Hand off to a human. The most underrated skill. A good agent knows what it does not handle and transfers to your on-call person with the context attached.
This is not exotic anymore. Salesforce's service research, as compiled by Vention, has 66% of service organizations running AI agents in 2026, up from 39% a year earlier. And the reason speed matters is old, peer-reviewed news: a Harvard Business Review study found companies that respond within five minutes are 21 times more likely to qualify a lead than those that wait thirty.
What they cannot do, and you should not let them try
- Diagnose machinery. An agent can capture "the car is stuck between 3 and 4 with a grinding noise." It cannot tell you whether that is a brake, a governor, or a bad roller guide, and any vendor who implies otherwise is selling you a mechanic shortage with extra steps.
- Make judgment calls about safety. Whether a car is safe to run until morning is a decision for your licensed people. The agent's job is to get the facts to them fast, not to freelance.
- Give code or compliance answers. Elevator code questions go to humans. Full stop. We configure agents to say so.
- Know your business without being taught. An agent is exactly as good as its knowledge base. Generic bots with your logo bolted on fail on the first contract-customer question. The build work is the product.
- Handle every acoustic situation. Machine rooms are loud, cell coverage in hoistways is bad, and a shouting caller in a lobby full of echo will occasionally beat the transcription. A well-built agent asks to confirm rather than guessing, but the failure mode exists and you should know it.
- Run unsupervised forever. Voice agents need monitoring, call review, and tuning, especially in the first weeks. Anyone offering set-and-forget is offering rot-and-embarrass.
The rule of thumb
If the task is collect, confirm, schedule, or notify, a voice agent does it well today, arguably better than a tired human. If the task is diagnose, decide, or negotiate, it belongs to your people, and the agent's only job is to hand it over cleanly.
An elevator office's phone traffic is mostly the first category. That is why this works: not because AI replaced your dispatcher's judgment, but because it took the eighty percent of the phone that never needed judgment in the first place.
If you want to test the boundary yourself, call our demo line at +1 (937) 530-6394 and push on it. Ask it something it should not answer. Watching where it hands off tells you more than any capability list, including this one.
Hear it instead of reading about it.
The demo line is a real AI receptionist. No signup, just a phone call.