How to choose an AI receptionist for a small business: a buyer's checklist

There is no single best AI receptionist, and any article claiming otherwise is ranking vendors it has not run for six months. What does exist is a short list of properties that reliably predict whether a given product will suit a small business, and a test you can run yourself in an afternoon. We build one of these products, so treat this as a checklist to hold us to as much as anyone else.

Start from your own call log, not a feature list

Before comparing anything, pull one full month from your phone provider. You need three numbers: total inbound calls, how many went unanswered, and when they clustered. A business missing eight calls a month has a completely different problem from one missing two hundred, and the right answer differs accordingly.

The nine questions

1. 1Can you keep your existing number? Conditional forwarding should be enough. If you are told to hand out a new number, the switching cost is on you and it is larger than it sounds.
2. 2How is it priced? A fixed monthly plan with included minutes makes cost per call fall as you grow. Per-call pricing rises exactly in the months a campaign works.
3. 3What happens when the agent does not know something? The honest answer is a warm handover or a captured callback. A product that improvises answers about your prices will eventually cost you a customer.
4. 4Which languages, on the same line? For Dutch, Polish or mixed-language markets this is often the deciding factor, and it is rarely on the pricing page.
5. 5What do you receive after a call? A transcript and a structured summary with your own fields beats a recording you have to listen to. You will not listen to recordings by week three.
6. 6How are transfers and emergencies configured? You should be able to define rules yourself: a keyword, a matter type, an explicit request for a human.
7. 7Where is data stored, and will they sign a processing agreement? For EU businesses this is not optional, and vagueness here is a real signal.
8. 8Can you test it before paying, with your own script? If you cannot dial the agent yourself before entering card details, you are buying a demo.
9. 9What is the exit? How quickly can you point the number back and export your call history.

The pricing traps

• Per-minute billing with a low headline price and no included bundle — cheap until the line is used.
• Setup fees framed as onboarding, on a product that is meant to configure in minutes.
• Charges per configured agent rather than per organisation, which punishes having a separate out-of-hours script.
• Annual commitments before you have handled a single real call.
• Overage that is invisible until the invoice. Ask where you see remaining minutes during the month.

The afternoon test

1. 1Configure the agent with your real opening hours, prices and service area — not the sample script.
2. 2Call it yourself and ask the three questions your customers actually ask most.
3. 3Call again and behave awkwardly: interrupt, change your mind, mumble a postcode, ask for a human.
4. 4Read what arrives afterwards. Is it a note your team could act on without listening to anything?
5. 5Check timing. Replies over roughly two seconds make a call feel laboured.
6. 6Only then compare prices. Most products fail step three, which makes step six irrelevant.

The product that wins is not the one with the best demo. It is the one whose transcripts you still trust after fifty real calls.

When a human service is the better buy

If most of your calls are emotionally sensitive, highly variable, or need someone to exercise judgement on your behalf mid-conversation, a human answering service is the honest recommendation — or a split where AI handles overflow and closed hours while people take the sensitive queue. Choosing on that basis costs less than discovering it after migrating your number.