What you should never leave to an AI at the front desk

You should never leave cancellations, refunds, payments, complaints or anything about safety to an AI. These messages need a person who can make a decision, take responsibility and judge the situation. What an AI can do well is recognise them, send the guest an honest holding reply and pass a clear summary to your team, so the person who picks it up can act at once.

That line is not a weakness of automation. It is what makes automation safe to use at a small property, where one wrong promise can cost more than a whole week of fast replies saves.

Why some messages need a person

Most guest questions have one correct answer that does not change: check-in time, parking, breakfast, Wi-Fi. A system can answer those reliably from the information you give it. The messages below are different. They depend on your policies, on the specific booking, on money that has already moved, or on how a guest is feeling. A plausible answer is not good enough.

The five things to keep in human hands

• “I need to cancel my booking.” Cancellation terms depend on the rate, the channel and sometimes your goodwill. A person should check the booking and decide.
• “When will I get my money back?” Refunds involve your payment provider, sometimes a booking platform, and real timelines you must be able to stand behind.
• “My card was charged twice.” Payment questions need someone who can see the transactions. Never ask for card details in a chat message.
• “The room was not clean and nobody helped us.” A complaint needs listening, an apology in your own words and often a gesture only you can offer.
• “There is a smell of gas in the kitchen.” Anything about safety, health, injury or a lost child is urgent and must reach a person immediately, by phone if needed, with emergency numbers given straight away.

Date changes and special requests that cost money, such as an extra night or an airport transfer you arrange by hand, often belong on this list too. Decide in advance, write it down, and let the rule be clear.

What a good hand-off looks like

A hand-off has two halves: what the guest sees, and what your team sees.

For the guest, a holding reply that is honest and specific. It confirms that the message has been read, says who will deal with it and roughly when, and promises nothing about the outcome:

“Thank you, Mr Byrne. I have passed your request to cancel to our manager, who will check your booking and reply to you here this morning. You do not need to do anything else for now.”

For an urgent safety message, the holding reply changes completely: it gives the emergency number and the property's direct line first, and alerts the person on call at once.

A good holding reply promises a person, not a result.

For your team, a summary that saves them reading the whole conversation:

• Who: guest name, booking reference, room, dates and channel.
• What: the request in one sentence, in your team's language.
• Context: anything already said, any earlier messages, the guest's tone.
• Urgency: whether it can wait for the morning or needs someone now.

Mistakes to avoid

• Pretending to decide. “Your refund has been approved” sent automatically, when nobody has approved anything.
• Silence. Holding back a reply until a person is free leaves the guest wondering whether anyone has seen the message at all.
• Endless loops. A guest who repeats “I want to speak to someone” and receives the same answer three times will remember it in their review.
• Lost hand-offs. A summary sent to an email address nobody checks at weekends is not a hand-off. It should land where the team already works. Our piece on a day with one inbox shows what that looks like.

Start with draft mode

If you are introducing an AI receptionist, the safest beginning is to read every reply before it is sent. After a couple of weeks you will see which answers are always right and can go out on their own, and which categories should stay with your team for good. The list above should stay with a person whatever you decide about the rest.

Where Concierva fits

Concierva is built around this rule. Cancellations, refunds, payments and complaints always go to a person, with a summary of the booking and the conversation; the guest receives a short holding reply in their own language. The routine questions are answered from your property's information. You can read more about how it works on the product page.

Everything arrives in one inbox, with handed-off conversations clearly marked, and you can start in draft mode, approving each reply until you are comfortable letting more go out automatically.