Is the AI front desk answering phone calls right now?

No. The chat concierge that answers website and message inquiries after hours is live. An AI voice receptionist that actually picks up the phone is on the roadmap, not in production, and it's described that way consistently.

How is it tested before it talks to a real customer?

Every agent has to pass a public, permanent test suite at the QA gate covering crisis handling, consent and TCPA behavior, prompt-injection resistance, common objections, and whether it actually captures the lead — before it goes live.

Who owns the system once it's set up?

The business does. The database, API keys, admin access, and the playbook all live in accounts the client owns. Cancellation is one email, and the system stays the client's afterward rather than reverting to a rented platform seat.

← Field notes

The missed-call problem: what an always-on front desk actually recovers

Q: What does the chat concierge actually do when someone messages after hours?

It answers the question, collects a name and a plain description of the job, and asks before offering a follow-up call or text — logging the consent language and timestamp. It routes anything urgent, disputed, or safety-related to a human instead of handling it.

Every call that rings out after 6pm is a lead deciding whether to try the next name on the list. AI front desk missed call recovery for a small business means something specific: the moment a call or message can't reach a human, something texts back, answers a chat, gets a name and a problem written down, and hands it to the owner with the urgent stuff flagged. It is not a phone system that talks. Today, for the Etzira team, that front desk runs as chat — live now. A voice version that actually answers the phone is on the roadmap, not in production, and it gets called that plainly every time it comes up.

The 9pm form and the two calls at once

Picture a normal Tuesday for an owner-operator. Two calls come in back to back at 2pm — one gets answered, one doesn't. A quote request lands through the website contact form at 9:14pm, after the crew's gone home and before anyone's checked email. A third caller tries once, doesn't leave a voicemail, and calls the next name in the search results instead.

None of that shows up as a missed-call notification in any dashboard. It shows up, if at all, three weeks later as "huh, revenue's soft this quarter." The owner's own math on this — what one missed job is worth, times how many calls actually go unanswered in a normal month — is the number that matters, not an industry average. That's the number an always-on front desk gets measured against.

What AI front desk missed call recovery actually catches

A live chat concierge sitting on the site and reachable after hours does three concrete things: it answers the 9pm question instead of leaving a form to rot in an inbox, it gets contact details and a plain description of the job before the visitor bounces to a competitor tab, and it asks before offering a follow-up call or text — logging the consent language and the timestamp rather than assuming silence means yes. STOP still opts a person out, no exceptions.

What stays human

It does not diagnose the actual problem, quote a firm price, or make a judgment call on anything that smells like a dispute, a safety issue, or a customer who's upset. Those get flagged and routed to a person, fast, rather than handled by a script pretending it has authority it doesn't. The point isn't to remove the owner from the conversation — it's to make sure the conversation is still happening when the owner isn't available to have it.

Chat is live. Voice is roadmap. Say that plainly.

This distinction matters enough to repeat without softening it: the chat concierge answering messages right now is a shipped, working system. An AI that picks up the phone and talks like a receptionist is not live anywhere in this stack yet — it's a roadmap item, and describing it as available before it is would be exactly the kind of overclaim a skeptical owner should walk away from. If a system claims a voice receptionist is running today, that's the tell to ask harder questions.

Before anything talks to a customer, it has to pass a test

Skepticism about AI answering customers is earned — a bad script can lose a customer faster than a missed call ever would. That's the reason for a public, permanent test suite at the QA gate: crisis handling, consent and TCPA behavior, prompt-injection attempts, common objections, and whether the thing actually captures the lead instead of just being pleasant. Every agent has to clear it before it goes live, and the gate stays open for anyone to look at, not just the client.

The skeptical case against AI customer contact isn't fringe, either — the curated research index collects the threads making that argument alongside the ones defending it. Reading the counterarguments before adopting anything is the right instinct, not a red flag.

Whose account is this actually running in

Everything gets installed in accounts the business owns: the database, the API keys, the admin login, the playbook that encodes how the front desk should talk. Cancel with one email, and it's still the owner's system afterward, not a rented seat in someone else's platform. The Etzira team's care retainer is optional; the vendor bills underneath it (model usage, hosting) get paid directly by the owner at market rates, not marked up and hidden in a bundle.

When this isn't the right call

Not every business is leaking enough after-hours volume to justify this. A shop that already answers its own phones fine, or one running under a call volume where a missed lead is a rare event rather than a weekly one, doesn't need an always-on system — it needs an answering habit fixed, which costs nothing. The honest version of this pitch includes saying so out loud, not just when asked.

For the businesses where it does apply, the test is simple: how many calls actually went unanswered last month, and what was each of those jobs worth. If that number bothers the owner, that's the conversation worth having next.

That conversation happens on the 20-minute call — no pitch deck, no discovery-call theater. By minute 20, the actual investment is on the table next to the owner's own missed-lead math, so the decision is arithmetic, not a guess. Questions before booking are answered at the FAQ.