How one attraction cut 70% of its phone inquiries

The reservations team at a thermal spa and water park had a problem every reservations manager will recognise. Long phone queues, guests waiting minutes to be answered, and the same questions arriving over and over: is tomorrow night available, what time is the animal feeding, can we bring children.

In peak summer and over holidays the front desk was splitting its attention between the phone and the people standing in front of it. And when that happens, bookings fall through for no better reason than nobody picked up in time.

This is what changed over four weeks, including the parts that took longer than expected.

What the queue was actually made of

Before changing anything, the team counted what the calls were about. That step matters more than the technology choice, and auditing the questions your team answers forty times a day sets out how to run it.

The answer was unglamorous: a short, repetitive list. Opening hours. Availability for a specific night. Whether children were allowed. Feeding times. Prices for the spa against the hotel.

None of those needs a voice, and none of them needs judgement. That is the case for automating them, and it is also the map for doing it.

What was put in place

Three things, in the order they mattered.

Answers in natural language, not a menu. A guest does not phrase a question the way a form expects. The assistant had to understand "is there anything free this weekend for two adults and a kid" as the same question as "availability Saturday", and answer both. The difference between that and a scripted bot is covered in how to tell what you are buying.

Any language, including voice notes. A meaningful share of guests wrote in languages nobody on the night shift spoke, and a lot of them sent voice messages rather than typing. Both had to work, in the property's own tone. Voice notes and languages you do not speak covers why that combination breaks most setups.

A live link to availability. An answer about a free room is only worth as much as the system behind it. Reading the real booking system, rather than a list typed out once and left to drift, is what let a conversation finish as a booking instead of a handover.

What changed, and over what period

The property's own before-and-after, three months in:

Around 70% of inquiries ended without a person typing. The remaining 30% is the interesting part: anything the assistant could not answer, and anything where being wrong would be expensive, went to an agent with the conversation attached, and the guest got a call back. Which requests should always take that route is worth deciding before launch, not after: the requests you must not automate.

The four weeks, honestly

Week 1, information and connections. Collecting what the team knew, setting the tone so replies sounded like the property rather than a vendor, connecting the booking and property systems, and two hours of training for staff.

Weeks 2 and 3, the real work. Watching actual conversations, finding the questions the assistant answered badly, and fixing the underlying facts rather than bolting on a rule each time. This is the stage people underestimate, and skipping it is the most common reason a deployment disappoints. Changing what your AI knows covers how to do it without breaking answers that already work.

Week 4, rollout. By the end of the first month most inquiries were handled without a person, and the load on the desk had dropped noticeably.

What it did to the team, which was the point

The team did not shrink. What changed was where its attention went.

Fewer interruptions from repeat questions. More time for the guest at the desk, who notices whether they have your attention. Consistent answers, so a new starter gives the same parking directions as the duty manager. And a quieter phone, which matters more for retention than it sounds: a front desk where five lines ring at once is a job people leave.

The reservations manager's summary of it was that the work became calmer. That is a smaller claim than the numbers above and probably the more useful one.

Where this works, and where it does not

It suits properties with high volumes of the same few questions: resorts and spas with several services, attractions with seasonal peaks, and anywhere a large share of guests write in languages the team does not speak.

It is poor value below roughly ten inquiries a day. At that volume a person can simply answer them, and the effort of writing everything down is not repaid.

If you run a water park or thermal bath rather than a hotel, why the water park phone never stops sets out the same problem from the attraction's side, where one question can arrive three hundred times on a Saturday.

What to do first

Spend a week counting what your phone is actually asked, split by question. Whatever you decide afterwards, that list is the most useful document you will produce, and it is the best brief any vendor will ever get from you.