In February 2026, we set out to try something at LIU Resorts: route every point of contact between the guest and the hotel through a single AI layer.

Seven months later, we have 282 hours and 23 minutes of direct guest conversation. That is roughly thirty-five eight-hour shifts.

But that number isn't the story. The story is what became visible over those seven months, after years of being invisible.

The Shift That Never Gets Logged

Every hotel runs a shift that appears in no report.

A guest asks a staff member in the elevator what time breakfast ends tomorrow. A towel is requested at the pool. Over dinner, someone mentions the air conditioning in their room is bothering them, and the waiter forgets to pass it on to the front desk. A Russian-speaking guest gives up on a question because nobody at the desk speaks Russian.

None of this comes from poor management. The team works hard and most requests do get handled. But none of it gets written down.

What isn't recorded can't be measured. What can't be measured can't be improved.

When we started at LIU Resorts, three issues were on the table, and all three led back to the same place.

Requests couldn't be tracked. They arrived through different channels and moved forward verbally or as scribbled notes. Where a request stood, when it closed, which department it went to, none of that lived in one place.

Peak hours stretched capacity. During check-in rushes, evening hours, and high-occupancy periods, incoming questions exceeded what the front desk could absorb. Guests waiting, or requests carrying over to the next shift, became unavoidable.

Feedback only surfaced after the guest had left. Guest opinions were tracked through online platforms. Which means after checkout. Anything said during the stay went unrecorded, so there was no way to act on it while it was still fixable.

What We Built

We configured RecepAI for LIU Resorts to cover every point where a guest touches the hotel.

Four channels, one assistant. Web chat, web voice, WhatsApp, and Google Business Profile all run through the same assistant. A request sent over WhatsApp during a stay and a review left on Google afterward are handled with the same underlying knowledge.

The hotel's information, not general information. The system runs on the hotel's own live knowledge base through a RAG architecture. Room types, facilities, restaurant and activity hours, policies. Answers don't begin with "hotels usually." They say what is true at LIU Resorts.

A request doesn't end in the chat. The assistant doesn't just answer. It records the request, routes it to the right department, tracks its status, and makes it reportable.

Feedback is collected during the stay. Guest opinions are gathered in the natural flow of conversation and reach the hotel directly. While the guest is still on the property.

Google reviews are answered automatically. Reviews posted through Google Business Profile are handled by the system. The number of unanswered reviews drops to zero.

No language limit. Whatever language a guest writes in, that's the language they're met in. Regardless of who happens to be at the desk.

Not one model, the right model. RecepAI runs on two of Anthropic's Claude models working together: Claude Fable and Claude Haiku. Not every guest question carries the same weight. Some answers call for speed, others for reasoning. Splitting the work across two models lets us hold on to response quality and speed at the same time.

Seven Months In

282 hours, 23 minutes. Total direct conversation time with guests. Continuous, including after-hours and the busiest moments at the front desk.

233 requests, all of them without human intervention. Over the last 30 days, 233 guest requests were recorded and routed to the relevant departments. Every one of them was taken directly from the guest, without the front desk having to step in.

56 pieces of feedback. Collected over the last 30 days with no additional human involvement and delivered to the hotel. Some of it was actionable while the guest was still on the property.

330 conversations in 8 days. Guest conversations during the first eight days of September. An average of forty-one guest interactions a day.

The Numbers Aren't the Point

These requests existed seven months ago too.

Guests were asking. The team was answering. Things were requested and, most of the time, delivered. What changed isn't that the requests exist. It's that they're visible.

Now it's possible to see which request came in when, how long it took to handle, and which department it landed on. Management discusses guest experience with data instead of intuition.

For a hotel, the practical meaning is this: the problem shows up while it can still be fixed, not in a Booking review.

Not Instead of the Team, Alongside It

The concern we hear most often in projects like this is always the same. Will AI replace the receptionist?

At LIU Resorts, the opposite happened.

The front desk team was freed from repetitive questions and the burden of record-keeping. In exchange, they spend more time on the work that actually requires a person. The system took over part of the load the team was carrying and made the rest visible.

The biggest contribution an AI receptionist makes is returning the human receptionist to human work.

What Comes Next

The LIU Resorts deployment is live and still developing. Response times shorten each month, the knowledge base deepens, and the range of requests the system can handle keeps widening.

But the point of this piece isn't specific to LIU Resorts.

Every hotel runs a shift nobody sees. Requests that never get logged, feedback that never gets heard, peak hours nobody can quite keep up with. The gap forms between the hotels that start measuring it and the ones that don't.

And that gap is the most expensive one to close once it shows up in a guest review.

RecepAI is the AI receptionist platform Hotel Linkage builds for hotels. Get in touch to talk about how it would work at your property.