What an AI answering service replaces
Most businesses do not choose an answering service against another answering service. They choose it against what happens now — which is usually voicemail, a diverted mobile, or nothing.
That is the comparison worth making honestly. Voicemail does not lose you calls because it is old-fashioned; it loses you calls because the person ringing has a live problem and a list of other numbers. They do not leave a message and then wait. They ring the next name, and you never learn it happened, which is precisely what makes the loss so easy to ignore. It appears in no report.
An AI answering service changes that in the least glamorous way possible: the phone gets answered. Every time, including at 7:43pm on a Saturday, which is when the leak in the ceiling and the lockout and the broken tooth actually happen.
What happens on the call
The call is answered in your business name, with a statement that this is your AI assistant. The caller says what they want in their own words — no menus, no keypresses.
From there it does one of three things. It answers, from a knowledge base built by reading your own website and refreshed nightly. It takes a structured message — name, number, email, company, reason, urgency — and emails it to you with the transcript. Or, if the matter is urgent and your plan includes call transfer, it rings you, tells you who is calling and why, and lets you choose: press 1 to take the call, 2 to have a message taken, 3 to promise a call back in five minutes.
Two details do a disproportionate amount of work. The caller’s number is captured from the network rather than from what they say, so a mumbled mobile at a bad moment still reaches you as a real number. And anything outside the knowledge base produces “I don’t know, let me take a message” rather than an invented answer — the failure mode that does most damage to a business’s reputation is a confident wrong price, not an admitted gap.
Human bureau, AI, or both
Human answering services are good, and they are priced for what they are: people, billed by the call or by the minute, above a monthly minimum, usually with 24/7 as a paid extra. That model is worth it when your calls are long, sensitive and consultative. It is poor value when your calls are two minutes of “are you open Saturday” and “can someone come and look at this”.
The arrangement plenty of firms end up with is neither purist: AI on the first line, answering the routine majority instantly and around the clock, with transfer to a human — you, or your team — for the calls that need one. You get the availability without paying human rates for the enquiry that took ninety seconds.
What we would not claim is that AI is better at every call. It is not. It is better at being there.
The economics, stated plainly
Per-minute answering has a structural quirk that is rarely discussed: the service is paid more the longer your callers stay on the line. Nobody is acting in bad faith, but the incentive points the wrong way, and it is your phone line.
Odette is a flat licence — £19, £49 or £99 a month — so it earns exactly the same whether a call takes forty seconds or four minutes. Telephony and voice are yours, held in your own Twilio and ElevenLabs accounts and paid to them at their rates: about £2.80 a month for a number, roughly 1p a minute of call, around 8p a minute of voice. On a busy line that is near £30 a month, putting a typical all-in cost under £50. We take no margin on any of it.
Concurrency — how many calls can be handled at the same moment — is set by your own Twilio plan rather than by us, so it is worth checking that figure against your busiest hour if simultaneous calls are the specific problem you are solving.
Hear it before you buy it
Our own number is 0330 912 3400 and it is answered by this product, not by a salesperson. Ask what it costs. Ask something it will not know. Leave a test message and watch what lands. It takes ten seconds and it will tell you more than this page can.