How to choose AI receptionist software for plumbers

Put one question to any AI receptionist you are considering: what happens in the middle of the night when a caller says water is coming through the ceiling? The receptionist worth paying for books that job into your real calendar, at an address the map can find, after telling the caller it is an AI. The checklist below is the long form of that question, written for a business in Europe, where the AI Act's disclosure duty has applied since 2 August 2026 and the caller's data has rules of its own. SwiftGuard's own answers are given under each check, so you can hold it to the same standard.
What should it do with an emergency call?
Two answers, because two different calls get called an emergency. A gas smell or a carbon monoxide alarm is a safety event, and the right response is to stop booking. SwiftGuard's agent says it cannot handle an emergency, tells the caller to leave the building and then ring their national gas emergency line or the emergency services, and, if someone is unconscious or in danger or there is a fire, to hang up and call 112. It marks the call urgent and does not read out step-by-step instructions.
A leak, no heating or no hot water is the other kind: an urgent job. SwiftGuard's agent notes the urgency in the booking for the team and never promises how soon someone will come. That promise is yours to make when you see the booking.
Whether the urgent job becomes a booking tonight depends on a setting that is easy to skip in a demo: the earliest time the receptionist may offer. A new SwiftGuard account starts at two hours, which suits a business that plans a round, and its own calendar guide calls that the wrong answer for emergency work: a burst pipe at 9am is offered half past eleven and rings somebody else. If you take call-outs, turn it down to fifteen minutes, or to As soon as I'm free, which offers the next open half-hour slot. Ask any vendor to show you where that setting lives before you sign.
Does it book into your real calendar, or just take a message?
A real booking means the receptionist reads your calendar during the call, offers only times you are free, and writes the appointment before the caller hangs up. A message means you still do the booking, later. Both have a place; only one gets a job into your diary while you are under a sink. For the wider comparison of a human answering service against an AI one, read answering service vs AI receptionist.
SwiftGuard connects to Google Calendar, and today only Google Calendar. Once linked, the agent sees when you are already busy and offers a caller only the times you are genuinely free. New accounts start with business hours of Monday to Friday, 08:00 to 18:00, closed at the weekend, and you can change that, working weekends included. If no times fit, or the calendar cannot be reached, it takes a message. How an AI receptionist works, step by step walks through the whole call.
Ask the vendor which calendars it writes to, and whether it reads your free and busy time live or works from a schedule you typed in once.
How does it capture the address?
Postcode and house number. Those are the two things a phone line carries cleanly, so SwiftGuard's agent asks for them rather than the whole address; its address lookup returns the street and the town, the agent reads those back, and it waits for a clear yes before it books anything. The caller says the numbers, the agent says the street. A street name spelled down a bad line is the easiest thing in the call to get wrong, and getting it wrong sends the van to a real house that is not theirs.
A wrong address is still a real address, so nothing downstream will ever catch it: the plumber drives to a stranger's door, loses the slot, and the customer is left waiting.
That line is from the agent's own instructions. It is why the read-back is compulsory. In a demo, give a postcode and a house number and listen for the street coming back to you; a receptionist that writes down whatever it heard has no way of knowing it heard wrong.
Where does your callers' data live, and who else touches it?
Under the GDPR you are the controller of your callers' data, because you decide what it is collected for, and the receptionist vendor processes it on your behalf as your processor. That split carries obligations you can check in an afternoon. Article 28(3) requires a written contract binding the processor, and lists what it must set out, the purpose of the processing and your own obligations and rights among them: ask for the data processing agreement and read it before you sign. Article 28(2) says the vendor may not hand your callers' data to another processor without your written authorisation, and under a general authorisation must tell you before adding or replacing one so you can object: ask for the list of sub-processors, with where each one processes. And Chapter V says a transfer outside the EU may only happen if its conditions are met; absent an adequacy decision, the Commission's standard contractual clauses are one of the safeguards Article 46(2)(c) allows. So for each sub-processor outside the EU, ask what covers the transfer.
SwiftGuard's answers, from its privacy policy: the core database is in Frankfurt, inside the European Union. Listening to the caller and transcribing the call happen inside the EU, and so does the summary written afterwards. The step that works out what the caller means is the part that leaves: the voice platform passes the words, as text, to a language model in the United States (Google's, with OpenAI's as the fallback). Twilio carries the call and holds the recording, in the United States. The transcript, the summary and the recording are deleted 90 days after the call; messages and booked appointments stay until you delete them. The data processing agreement lists the sub-processors in Annex III, gives at least 30 days' notice before one is added or replaced so you can object, and says plainly that SwiftGuard does not yet hold a signed data-protection contract with every one of them. Judge a vendor by whether it will tell you that kind of thing unprompted.
Does it tell callers they are talking to AI?
In the EU the law requires it. Article 50(1) of the AI Act puts the duty on the provider of an AI system that interacts directly with people: design it so that those people are informed they are dealing with an AI, unless that is obvious from the circumstances. The Act applies from 2 August 2026. The European Commission's own FAQ confirms Article 50 applies from that date, with a grace period only for the separate marking duty in Article 50(2), and only for systems already on the market before then. The duty to build the disclosure in sits with the provider, and the Act also applies to deployers of AI systems established in the Union; SwiftGuard's own disclosure page names the business whose calls it answers as the deployer. So the check is whether the disclosure is built in, and whether anyone can turn it off.
SwiftGuard's agent opens with a fixed line saying it is the AI assistant for your business and that the call is recorded and written down so the job gets booked correctly. Its AI transparency disclosure says that line cannot be switched off - by SwiftGuard or by the business using it - and names what can get in its way: a caller who talks over it, a noisy line, or a fault that drops the connection at the start of the call, may mean they hear part of it or none of it, and they can ask at any point whether they are talking to a person. That page also sets out the roles: SwiftGuard is the provider, and the business whose calls it answers is the deployer, with nothing to configure for its callers to be told.
Which phone number, which language, which currency?
Three small questions decide whether it works on your line at all. The number: whether the receptionist takes over your existing number or hands you a new one to print on the van. With SwiftGuard you keep your number and your phone. Calls reach the agent through forwarding: on a mobile, one short code dialled on your own handset; a landline, VoIP line or hosted phone system generally does not accept dial codes, and forwarding is a setting in that provider's control panel instead. A line in the Netherlands is the country it supports forwarding for today, and the setup screen says so if you are somewhere else.
The language: which language it opens in, and what it can switch to. SwiftGuard's agent opens in the language you choose for your business (English or Dutch today), and it can speak six languages, switching to German, French, Italian or Spanish when a caller answers in one of them.
The currency: a price in dollars is a sign the vendor has not priced for you yet, and a price with no tax basis is one you cannot compare. SwiftGuard's plans are in euros, ex VAT, on a flat monthly fee with a bundle of minutes that is a hard cap with no overage: the Starter plan is €69 a month ex VAT for 150 minutes. The full plans and prices are on one page.
The checklist to take into a demo
If the guide you found prices in dollars and mentions neither a data processing agreement nor the AI Act, it was written for a different market. These are the checks that matter here:
- Gas or carbon monoxide: does it stop booking and send the caller out of the building?
- Urgent jobs: does it flag the urgency, and can you set the earliest slot it may offer to minutes rather than hours?
- Booking: does it read your live calendar and write the appointment during the call, and which calendars does it support?
- Address: does it ask for postcode and house number, and read the street back for a yes?
- Data: is there a data processing agreement, a sub-processor list with locations, and a named safeguard for each transfer outside the EU?
- Disclosure: is the AI announcement built in, and can anyone switch it off?
- Line, language, price: does it work on your existing number, open in your language, and quote in euros with the tax basis stated?
Run SwiftGuard through it. Its answers are in this post and on the linked pages, and if one of them falls short for your business, that is worth knowing before the first call comes in.
