Back to Blog

Book it or hand it over? Every hotel draws the line in a different place

Every hotel that puts AI on its chat has to answer one question before anything else: when should the assistant stop selling and call a person?

MN

Mushon Nachmani

vGuest

4 min read

Every hotel that puts AI on its chat has to answer one question before anything else: when should the assistant stop selling and call a person?

It sounds like a setting. It is actually your business model.

Draw the line well and your team spends its day on the enquiries that genuinely need judgement. Draw it badly and you have either rebuilt your old switchboard, or handed a machine the authority to price the business you make your margin on.

The useful thing is that there is no industry answer. There is only your answer. Three properties we have worked with drew three completely different lines, and all three were right.

Three properties, three lines

A 4-hotel boutique chain lets the AI complete bookings up to three rooms. Four rooms or more goes to a person. So do corporate clients, returning guests on negotiated rates, and waitlist requests. Every hand-off is tagged by hotel and by topic, so it lands with the desk that owns it rather than in a general queue.

The reasoning is straightforward. Up to three rooms, the rate on the booking engine is the right rate and there is nothing to negotiate. Above that, someone is going to ask about a meeting room, a group breakfast sitting, or a late release date. That conversation is worth a human.

A country hotel shows at most two specific room options, each with a direct booking link. Three rooms or more is a group, and a group goes to the reservations team.

Two options, not eight. The hotel found that a long list of room types in a chat window produces questions, not bookings. Narrowing to two and linking each one straight through to payment is a deliberate sales decision, not a limitation.

A visitor centre defines a group as 20 or more people of the same kind: one company, one pensioners' club, one school. Only groups get a tailored quote and access to exclusive activities. Twenty-two strangers who happen to book the same time slot are not a group, and must not be treated as one.

That distinction matters commercially. The group price buys a dedicated guide and a private slot. Give it away to twenty-two unrelated walk-ins and you have discounted your busiest hour for nothing.

None of these rules is wrong. Each reflects how that property makes money and where its staff genuinely add value.

What goes wrong with a generic bot

It hands over too early. The bot is cautious, so anything with a number in it goes to a person. Within a fortnight your team is answering "what time is breakfast?" again, and the project has bought you nothing. This is the most common failure and the hardest to spot, because nobody complains. The work just quietly comes back.

It hands over too late. The bot keeps selling past the point where it has authority, and promises something only a manager can approve: a discount, a late checkout on a sold-out Saturday, a price for forty people. Now you either honour a commitment you did not make, or start the guest relationship with a retraction.

It uses the wrong definition. It sends a family of twelve to the group desk, who quote a group rate for what is really three connecting rooms. Or it books a genuine forty-person coach party at retail, through the booking engine, on the one weekend you had held for a wedding.

All three failures have the same root cause: the rule existed in someone's head and was never written down.

What to write down before you launch

1. The largest booking the AI may complete on its own. A number, not a principle. "Up to three rooms." "Up to six people on one ticket."

2. Who counts as a group, in your words. If your definition has a qualifier (same organisation, same arrival, same rate), write it down. This is where generic rules fail most often.

3. The questions that always go to a person. Cancellations, invoices and VAT queries, negotiated or agency rates, complaints, anything involving money already taken. These are not hard questions; they are questions where being wrong is expensive.

4. What counts as urgent. This is the one properties forget. One attraction escalates "we're running late today" by phone rather than by message, because a late arrival can mean a missed slot and a cancelled visit with no refund. A message sitting unread for twenty minutes is a lost booking and an angry guest. Decide which sentences deserve a ringing phone.

5. Where each hand-off lands. Reservations, reception, or a named person. Decide what happens to it outside their hours too. A rule that only works Monday to Friday, nine to five, is half a rule.

Two things worth checking after you write it

Test the edges, not the middle. Anyone can check that a two-night double booking works. Ask the assistant for four rooms, for a quote for a coach party, for a discount, for an invoice to a company. The edges are where your line either holds or does not exist.

Re-read it in three months. Hand-off rules go stale. You add a meetings space, you change who owns corporate, a seasonal desk closes. The line that was right in March is wrong by October, and nobody notices until a group has been booked at the wrong price.

Takeaway

The hand-off line is not a technical detail and it is not something a supplier can choose for you. Write it down the way you would brief a new reservations agent on their first morning: plainly, with numbers, with the exceptions spelled out. Then make sure your AI follows that brief exactly.

That brief needs two more layers: your tone and house rules, and the way your hotel actually runs.

Common questions

When should a hotel AI assistant hand a guest over to a person?

There is no industry answer, only yours. Write down the largest booking the assistant may complete on its own as a number rather than a principle, who counts as a group in your words, the questions that always go to a person, what counts as urgent, and where each hand-off lands, including outside working hours.

What happens if the hand-off line is drawn too early?

The assistant becomes cautious, so anything with a number in it goes to a person. Within a fortnight your team is answering "what time is breakfast?" again and the project has bought you nothing. It is the most common failure and the hardest to spot, because nobody complains. The work just quietly comes back.

What happens if the assistant keeps selling past its authority?

It promises something only a manager can approve: a discount, a late checkout on a sold-out Saturday, a price for forty people. You then either honour a commitment you did not make, or start the guest relationship with a retraction.

How should a hotel define a group for its AI assistant?

In your own words, with the qualifier included. One visitor centre defines a group as 20 or more people of the same kind, one company, one club or one school, because the group price buys a dedicated guide and a private slot. Twenty-two strangers who happen to book the same time slot are not a group and must not be treated as one.

How do you test whether the hand-off rules actually work?

Test the edges, not the middle. Anyone can check that a two-night double booking works. Ask the assistant for four rooms, for a quote for a coach party, for a discount, for an invoice to a company. Then re-read the rules in three months, because hand-off rules go stale when you add a meetings space or change who owns corporate.

All articles

190+ properties · 6 countries

See it answering your guests

Bring your own property. We will show you what the assistant does with your real questions, in your languages, on WhatsApp.