AI Receptionist and Voice Agent
An AI receptionist answers inbound calls, captures the caller and the reason, books or transfers, and logs the result to your CRM. OperStack sets the transfer threshold deliberately, because a bot that keeps a confused caller talking costs more than one that hands over early. Most engagements take 3 to 5 weeks including evaluation on real recordings.
| Entry point | Setup · Custom quote Custom quote. Most teams start with out-of-hours only, then extend. |
|---|---|
| Typical timeline | 3 to 5 weeks including evaluation on real recordings Depends on how much of your data is already clean. |
| Updated | Reviewed by Maksim Shchegolev |
The metric vendors quote is the wrong one
Voice vendors sell containment: the share of calls handled without a human. Containment is easy to raise and easy to raise badly, because a caller who gives up is contained.
The number that matters is resolved contacts, plus the transfer rate at the moment the caller first signalled confusion. A high containment rate with a rising abandon rate is a worse outcome than a bot that transfers a third of calls cleanly.
Out-of-hours is where the honest value is
During business hours you are comparing the agent against a person who answers. That is a hard comparison to win and the failure is visible to a customer.
Outside hours you are comparing it against voicemail, or against nothing. Capturing the caller, the reason, and a callback slot at 22:40 is a genuine improvement over a message nobody returns until Thursday. Start there.
Where the response clock starts and pauses, which decides whether that callback counts as fast, is in the speed-to-lead guide.
Failure has to sound like something
Text agents fail quietly. Voice agents fail out loud, to a person, in real time. That changes the design: the fallback is not a log entry, it is a sentence the caller hears and a transfer that actually connects.
We script the failure explicitly, test it, and make sure the transfer target is staffed. An agent that says “let me put you through” to an unstaffed queue is worse than no agent.
Where a voicemail still wins
Low call volume, highly technical callers, or a business where the first question is always about price. In those cases the agent adds a layer between the caller and the answer, and a short voicemail with a fast callback beats it.
We will tell you that on the audit call. It is a small piece of revenue for us and a large saving for you.
What the engagement covers
- Call flow design: what the agent answers, what it captures, when it transfers
- Transfer thresholds, including an immediate human path the caller can always reach
- Calendar integration for booking, with rules for the slots it may offer
- CRM writeback: caller, reason, outcome, and the transcript attached to the record
- Out-of-hours behaviour, which is where most of the value actually sits
- Evaluation on recordings of your own calls before it answers a real one
What you keep
- The call flow in writing, reviewable by someone who has never seen the tool
- Measured transfer and containment rates from your own evaluation set
- A kill switch that routes everything to a human, owned by a named person
What this does not cover
- Outbound calling campaigns. Inbound answering only
- Telephony migration. We integrate with the phone system you already run
- Any commitment to a caller. Quotes, availability promises, and pricing stay human
If one of these is the actual problem, say so on the audit call and we will point you somewhere better rather than sell you the wrong module.
Start with the written assessment
Free and in writing: send the site and how inbound works today, get the gap we see, the module that fixes it and a fixed price within two working days. No call at this stage.
Frequently asked questions
- What can an AI receptionist reliably handle?
- Identifying the caller, capturing why they called, answering questions your documentation already answers, booking against calendar rules, and transferring. Those have testable outcomes. Negotiation and exceptions do not.
- When should it transfer to a human?
- Earlier than most vendors configure. Two failed intent matches, any pricing question, any complaint, and any caller who asks for a person. A bot that keeps a frustrated caller in a loop costs more than the call it saved.
- Is starting with out-of-hours a good idea?
- Usually the best starting point. Those calls currently reach voicemail or nothing, so the comparison is against zero rather than against a good human answer. The risk of a mediocre first version is much lower.
- How do we test it before going live?
- On recordings of your own calls, scored for whether the agent would have captured the right reason and transferred at the right moment. Vendor demos use clean audio and cooperative callers, which is not your Tuesday afternoon.
- Will callers know it is not a person?
- They should, and we configure it to say so. Callers who discover it later report the experience worse than callers told upfront, and in several jurisdictions disclosure is a legal question you should check locally.