Comparison
AI or human front desk: which to choose
· 7 min read
You don't have to choose “either/or”: a division of labour works best. The bot gets the standard questions and the routine (is there room, what time, where's the menu, how do I move my booking), and a person keeps everything that needs a decision, tact or an exception to the rules. The key thing in such a setup isn't the bot itself but the moment of handover: the guest shouldn't have to repeat everything, and the member of staff should see what has already been discussed.
Below we look at how to draw the line, how to arrange the handover and how to keep an eye on quality.
Standard questions: what can be given to the bot
Take a week of conversations and write down the guests' questions. Most likely, most of them repeat:
- “Do you have room on Friday?”
- “What time do you open?”
- “Can I see the menu?”
- “I need to move my booking an hour later”
- “What's your dress code?”
- “How do I cancel?”
These questions have a clear-cut answer, and answering them doesn't take mood or experience. They suit a bot well. In the evening, when the house is full, the manager spends time on them that could go to the guests standing at the door.
A good sign: the answer can be written in advance or comes from the data (free places, the guest's existing booking). A bad sign: the answer depends on who's asking and what mood they're in.
Unusual cases: what to leave to a person
Some situations are best handled by a member of staff from the start:
- A complaint or a conflict. The guest was unhappy with a previous visit. They need an apology and a solution, not polite wording.
- Exceptions to the rules. “There'll be two more of us than you allow in the room”, “We'll be an hour late, but please keep the table”.
- Large and special events. A corporate party, a birthday with decorations, hiring the whole room.
- Money and disputes. A refund, a confusing bill, questions about a deposit.
- Anything the bot failed to understand twice in a row.
A simple rule for the owner: if you yourself would say “let me check with the manager”, it's a question for a person.
For example, a guest writes at 23:40: “We'd like a table for six tomorrow, and it's one of our birthdays, could you do something?” The bot can handle a booking for six. But “do something” (a cake, candles, a special presentation) depends on a decision by the venue, and it's better to pass it to a member of staff with a clear note.
The handover: how not to lose the guest at the joint
The handover breaks in three places: the bot doesn't realise it's time to call a person; the person can't see what was being discussed; the bot and the person answer at the same time. You need clear rules.
- The bot calls a person itself. When a question is beyond its limits or the guest asks for a manager, the staff get a notification.
- The person sees the whole conversation. The member of staff opens the conversation and reads it from the start, without questioning the guest.
- While a person is talking, the bot stays quiet. Otherwise the guest gets two different answers.
- The bot comes back by itself. If the member of staff doesn't answer for a long time, the conversation shouldn't hang forever.
- A member of staff can step in at any moment, even if the bot didn't call: it's enough to take over the chat.
This is exactly how it works in GuestKeepr. When a question is beyond its limits, the virtual front desk hands the conversation to a person, and the staff get a notification in Telegram. A member of staff can take over the chat and answer from the dashboard; meanwhile the bot stays quiet and comes back after 12 hours of the staff member's silence. That's an important detail: if the member of staff forgot to close the question, the bot won't leave the guest without an answer forever.
Quality control: how to check that the bot answers correctly
You can't launch a bot and forget about it. A simple habit is enough:
- Read a few conversations once a week. Pick the latest ones and those where a person was called. You'll see where the bot didn't cope.
- Add to the ready answers. If guests keep asking the same thing and the bot answers imprecisely, add the answer to the venue description.
- Check what is said about money and rules. The bot should state the deposit and cancellation terms only the way you wrote them.
- Keep an eye on the log. Who changed what in bookings can be seen in the activity log.
The GuestKeepr dashboard has the Conversations section for this: it holds all the conversations with the bot, with search and filters. The bot also always asks the guest for a name, a phone number, and the place and time, and doesn't confirm a booking it didn't create; and it works only with the guest's own bookings, with ownership checked by a confirmed number.
If you want the bot to answer more precisely, start with good information about the venue: we have an article on how to set up a booking bot.
Costs: what to look at without figures
We don't give figures, because pricing depends on the venue, and GuestKeepr's prices haven't been set yet. But costs can be compared qualitatively too. Think about four items:
| Item | Human manager | Bot |
|---|---|---|
| Time on standard questions | Takes attention away from the floor | Answers at once |
| Night and out-of-shift messages | Needs someone on duty, or requests are lost | Always answers |
| Unusual situations | Decides on the spot | Passes them to a person |
| Setup and control | Training and shifts | Preparing data and reviewing conversations |
We don't give dates or prices. There are also limits against spam and protection against extra AI spending, so that a random flood of messages doesn't turn into an unpleasant bill.
The honest assessment is this: a bot doesn't replace a person; it takes the routine. If the manager manages everything now, the gain will be small. If requests are lost in the evening and at night, the gain is noticeable at once.
How to divide the work: a short checklist
- Write down the last twenty questions from guests.
- Mark the ones that have a clear-cut answer, and give them to the bot.
- Write down the rest as “we call a person”.
- Agree who on the shift gets the notifications and answers.
- A week later, read the conversations and add to the answers.
What exactly the bot can do is described in detail in the article about a booking bot for a restaurant. And if you're choosing between a chat and a form on the website, take a look at the comparison of a bot and a widget.
What to tell the team
Bringing in a bot often runs into not technology but the staff's mood. The manager may decide they're being replaced and start sabotaging it. Say it straight: the bot takes the repetitive questions and the night messages, and people keep the guests, the exceptions and the decisions. Show on a real conversation how many repeats it contains.
It also helps to agree on a tone. If the bot writes politely and briefly, a member of staff picking up the chat shouldn't switch abruptly to another style. Put together two or three sample answers that the whole team uses, and the guest won't notice the switch.
Frequently asked questions
Will the guest notice they're talking to a bot?
Most likely yes, and there's no need to hide it. A fast and accurate answer matters more to the guest, and when needed they should get a living person.
What if nobody's there at night and the question is difficult?
The bot will answer what it can and pass on the rest: a notification goes to the staff, and in the morning a member of staff opens the conversation and sees all of it. The guest won't be left without a response.
Can the bot cancel someone else's booking?
No. The bot works only with the guest's own bookings, and ownership is checked by a confirmed phone number. In Telegram there's a “share phone number” button for this.