Hospitality

What Sofiane Ghorbel's team does after a client lands in Rome

Luxury trip follow-up runs on phone calls to every supplier. Sofiane Ghorbel: it takes a lot of time, it creates loyalty, and no tool exists.

Luxury trip follow-up is still done by phone, one supplier at a time, and no product does it. Sofiane Ghorbel, founder of AL Hospitality Group and a hospitality executive with more than twenty-five years in luxury hotels, travel and destination management, says his team does not call the client at all. It calls every provider the client is using, and that is where his time goes.

Asked whether a tool exists for it: "I don't have this tool till now. I hope that we will find it, but it will be great to have it."

What does luxury trip follow-up actually involve?

Checking every supplier in the itinerary, in advance, without the client knowing it is happening. Ghorbel's description:

"We are his shadow. In the luxury we are always here."

"We don't call them at all, but we call all those providers that they are using that we booked ... we are already in contact with the concierge of the hotel in Rome ... Is everything alright? Do they have the welcome envelope letter? Do they have their bottle of champagne already there? Everything is checked."

How embedded that is in the client's own operation:

"It's like [I am] part of their team and I'm in their briefing meeting in the morning ... my seat is there. So this is how we behave."

And the cost of doing it that way:

"It takes us a lot of time. This is the problem."

Why can't the work simply be dropped?

Because it is the step that produces loyalty, which is the whole economics of the business. Ghorbel is explicit about it:

"This step of follow up is very important. It is where the client will be loyal."

"Don't send people in Italy and leave them there."

This is an unmet need rather than an AI failure. Nothing broke. There is simply no product between the booking system and the phone call, and the most expensive person in the company is the one making the calls.

Why don't general-purpose tools cover it?

They are not built around the workflow. Ghorbel's assessment is one sentence long:

"Most AI tools are generic currently. And hospitality needs workflow-specific solutions."

Fendikevich framed the downstream consequence. With only general-purpose models available, using them well requires prompting skill, iteration on output, and knowing what good looks like well enough to recognize when the output is wrong. That last requirement is exactly what junior staff do not yet have, which produces a talent gap where the people most fluent with the tools are the least equipped to catch their errors. Ghorbel's answer is that training is not optional and employers have to run it, but that is a mitigation, not a fix.

What would the product actually have to know?

Things that were never written down. This is the operator-side data problem, and it is separate from the client-privacy problem it usually gets bundled with:

"[That means that] hospitality remains highly fragmented and much of the industry knowledge still lives in relationships, private networks, human expertise and local market knowledge."

So a follow-up product would need to hold, per trip, at minimum:

Every supplier in the itinerary and the named human contact at each one, not the switchboard.
What was promised at each step — the welcome envelope, the champagne, the room category — in checkable form.
When each check has to happen relative to arrival, which is judgment currently held by the advisor.
What was confirmed, by whom, and what is still open, visible without a phone call.

None of that requires client preference data, which is the part of hospitality where consent makes the problem intractable. It requires the operator's own supplier network, which the operator already owns.

What to build

Build the checkable half. Confirmation status against a promised itinerary is verifiable, non-confidential and entirely inside the advisory firm's own records, which is why it is buildable now while guest personalization is not. The accuracy bar is unforgiving: at eighty percent, in a business where one incorrect recommendation can damage a relationship that took years to build, the twenty percent is disqualifying rather than annoying. Design for a system that flags what it could not confirm instead of asserting that everything is fine.

The commercial signal here is unusually clean. A named operator with a twenty-five-year network has described the workflow in detail, said plainly that it consumes his time, said it cannot be dropped because loyalty depends on it, and said no tool exists. That is a lead, not a think-piece.

So the question for anyone building in this category: are you trying to automate the recommendation, which this industry will not give you the data for, or the follow-up, which it will?