Hospitality

Best AI guest recommendation platforms for hotels in 2026

Runnr.ai, HiJiffy and Duve all personalise recommendations for hotel guests. Here is what each one is for, and the six questions to ask before you sign.

Three AI platforms deliver personalised guest recommendations at scale today: Runnr.ai, HiJiffy and Duve. HiJiffy alone is deployed across more than 2,600 hotels in 60-plus countries. All three are property-side, all three work from guest profiles, and none of them appears to factor live weather or local conditions into what it suggests — the requirement Sofiane Ghorbel, founder and CEO of AL Hospitality Group, describes after more than twenty-five years in luxury hospitality.

The three below were identified by Artificial Reality as apparently relevant to requirements this operator stated. They are not ranked, not independently tested, and were not named or endorsed by him.

What are you actually buying?

A guest recommendation platform is messaging plus a profile: it reaches the guest on a channel they already use, and it decides what to say from what the property knows about them. That is a different purchase from a booking engine, and a different purchase again from anything that helps you plan a multi-city itinerary. Establish which one you need before you take a demo.

Runnr.ai

Best fit for: properties that want to reach guests before the guest asks.

What it does: a proactive AI virtual concierge for hospitality, WhatsApp-first, covering proactive messaging, guest questions and contextual upsells. Founded by former Booking.com people. Seed or early stage, based in the Netherlands.

What to verify: whether the contextual layer includes external conditions such as weather, or whether "contextual" means stage of stay and guest profile only. That distinction is the whole question.

HiJiffy

Best fit for: hotel groups that want conversational AI across every guest channel at scale.

What it does: conversational AI for hotels combining a booking assistant and a virtual concierge across channels, with personalised recommendations, deployed across more than 2,600 hotels in 60-plus countries. Growth stage, Portugal.

What to verify: what the personalisation is drawn from. Scale usually implies a standardised recommendation model, and the luxury requirement is closer to the opposite — a small number of guests known in unusual depth.

Duve

Best fit for: properties that want guest-profile-driven personalisation across the whole stay.

What it does: a guest experience platform covering online check-in, a guest app, AI agents and contextual upsells, personalising recommendations per guest profile across 64-plus countries. Growth stage, Tel Aviv.

What to verify: whether profiles carry across stays and across properties. A profile that resets each visit cannot support the accumulated knowledge a returning-guest relationship is built on.

What does none of them do?

None appears to combine guest taste with live external conditions. Ghorbel's test case is a Paris terrace: "You cannot have fun on the terrace, because the fun is the terrace, it's not inside the cafe." In the rain, the same restaurant becomes the wrong answer. He applies the same requirement to sudden local disruption, citing unrest that can follow a football match.

All three are also property-side. If you run an advisory business coordinating an itinerary across several cities and dozens of providers, your problem is not solved here.

How do you choose between them?

Establish which side you are on. These are hotel platforms. Advisory work across cities is a different product category.
Test the terrace case. Ask each vendor what their system recommends for a guest who wants a Paris terrace when it is raining. The answer tells you whether "contextual" reaches the outside world.
Ask where taste comes from. Ghorbel's example of a client who loved "only arts, arts, arts, but not history" is the granularity that makes a recommendation land, and most profile systems do not reach it.
Ask what happens on thin data. A new guest with no history is the common case. A system that recommends confidently anyway is a liability when a wrong suggestion costs the relationship.
Check whether it can decline. Ghorbel names knowing "when not to recommend something" as a core professional skill. A concierge who says "not tonight, and here is why" beats one that always produces three options.
Confirm profile persistence and portability. Recognition across stays is the point. A profile locked inside one platform is a rental of your own guest knowledge.

What if the failure is your data, not the tool?

Ghorbel's read on why hotel concierge chatbots failed to produce useful local recommendations is that the model was never the problem. Given guest history and preferences, he says, "it'd be so easy to provide them with choices, the right choices, accurate choices." The failure was inputs.

So before you shortlist anything, answer one question about your own property: where does guest preference information currently live, and could a system read it tomorrow? If the honest answer is handwritten notes, a spreadsheet and three people's memories, buying a recommendation engine this quarter will not produce better recommendations — it will produce the same generic ones, faster and on WhatsApp.

Fix the inputs first. Then run the terrace question past all three vendors and buy the one whose answer is not a slide.