Field notes · 10 April 2026

The Search Box Was the Wrong Answer

Every experiences platform built a search box. Nobody asked whether it was the right product.

When Viator scaled, when GetYourGuide expanded, when every travel platform added an experiences tab, they all reached for the same interface. A search box. A filter panel. A ranked list of results sorted by review score and popularity. The product looked like what they already knew how to build, and it looked like something consumers already knew how to use. That familiarity was mistaken for validation.

The search box works in other parts of travel because the user arrives with a query. When you open a flight booking platform, you know the origin, the destination, and the dates. The product you want is defined before you start searching. When you open a hotel booking platform, you have a city and a check-in window. You can filter by location and price because you know roughly what you are looking for. The search interface is a reasonable match for the decision you are actually making.

Experiences broke that assumption, and nobody stopped to notice.

A traveller in Rome on a Thursday afternoon with four hours free does not have a query. They have context. They are with their partner, they have already done the Vatican and the Colosseum, yesterday was long and they want something quieter, they are interested in food, they do not want to be in a group of forty people. None of that maps to a search box. The search box returns results for "Rome food experience" and presents 340 options with an average rating of 4.7 stars. The traveller scrolls for a while, feels overwhelmed, and asks the person at the hotel front desk.

That person at the front desk is worth understanding.

What the Concierge Actually Does

The hotel concierge at a good property is the highest-converting experiences distribution channel that has ever existed. They have no app. They have no inventory feed. They have no algorithm. And they consistently get people to part with money for experiences they had no intention of booking an hour earlier.

The mechanism is worth studying in detail, because it is the opposite of what every platform built.

The concierge does not start with options. They start with questions. Who are you here with? How long are you staying? Have you been to Rome before? What did you do yesterday? Are you looking for something active or something slower? Do you have a car? What time do you need to be back? The questions are not a form. They are a conversation, and the concierge is building a picture that has nothing to do with category filters or price brackets.

They are collecting context. And context is the thing that makes a recommendation useful rather than generic.

Once they have enough of it, they do not open a database. They recall from a curated mental list that probably contains twenty or thirty things they know well and trust personally. They have sent guests there before. They have heard back. They know which ones are reliable, which ones suit couples versus families, which ones require energy and which ones are easy. Their constraint is not access to supply. It is the quality of their own knowledge. A concierge who has not personally verified something does not recommend it.

From that list, they give you three things. Four at most. They do not give you a ranked list of thirty and leave you to work through it. They make a judgment about what is right given what they now know about you, and they present a small set with specific reasoning attached to each. "Given that you want something quieter, this one is better than the other. Maria only takes six people, it is in her apartment in Trastevere, very relaxed. I sent a couple last week who said it was the best thing they did in Rome."

That sentence does more conversion work than any star rating.

Then they close it. They call ahead if they know the operator. They tell you exactly where to go and what to expect. They remove the operational uncertainty that makes people hesitate. And their recommendation carries weight because their reputation is attached to it. A concierge at a good hotel does not send you somewhere they are not sure about, because they will see you at breakfast the next morning.

The entire model runs on context, curation, confidence, and skin in the game. The search box offers none of these things.

Why the Industry Chose the Search Box Anyway

The choice was not irrational given what was available at the time. In 2010 and 2015, building a personalised recommendation system at scale was genuinely hard. The AI infrastructure that would allow a platform to collect traveller context conversationally and reason about operator fit in real time did not exist in a usable form. What existed were search interfaces, review aggregation, and inventory management systems that had already been built for hotels and flights. The path of least resistance was to adapt those tools for a new category.

The deeper problem was a failure of category thinking. The platforms looked at the experiences market and saw a supply-and-demand matching problem, which is what their existing tools were built to solve. Feed enough supply into the platform, make it searchable, let review scores surface quality, and the market will clear. That model works when the product is standardised and the buyer arrives with a defined preference. It does not work when the buyer does not know what is possible until someone who knows the category tells them.

A cooking class in Rome is not a unit of inventory in the way a hotel room is. A hotel room in a given neighbourhood at a given price point has attributes that can be compared meaningfully across options. A cooking class has a host, a format, a group size, a setting, an energy, a type of food, and a teaching style, none of which are captured by a category tag and a star rating. Two classes with identical ratings can produce completely different experiences for the same person depending on who they are and what they want from the afternoon.

Search cannot process this. The concierge can, because the concierge asks before they answer.

The result of a decade of search-based experiences platforms is an inventory problem that presents itself as a discovery problem. There is no shortage of supply on these platforms. The traveller who opens Viator in Rome is not failing to find experiences. They are finding too many, with too little signal to distinguish them, and insufficient confidence to commit to any of them. The conversion rates in the category reflect this. Experiences platforms convert at a fraction of what hotel platforms do, and the gap is not explained by supply quality. It is explained by the interface design.

What Digitising the Concierge Actually Means

The concierge model is not replicable at scale through human labour. A platform cannot hire a person who knows every city personally and talks to every traveller. But the model is replicable through a different kind of product, and the components that make it work are now buildable in ways they were not ten years ago.

The first component is context collection. The concierge asks questions before making a recommendation. A platform can do the same thing through a conversation rather than a search box. Who are you travelling with, what have you already done, what kind of afternoon do you want. This is not a novel idea. What is novel is that the AI infrastructure to have that conversation, reason about the answers, and match them against a curated supply set is now good enough to be useful.

The second component is curation. The concierge's power comes from the selectivity of their knowledge. They know twenty things well rather than eight hundred things superficially. A platform that curates its supply, vets operators personally, and lists fifty well-understood experiences in a city rather than five hundred aggregated ones is replicating the constraint that makes the concierge's recommendation credible. The platform that has fewer listings but knows them better is building the same kind of trust.

The third component is reasoning at the point of delivery. The concierge does not hand over a shortlist. They explain why a particular option is right for this specific person given what they just said. A platform that generates a recommendation with specific reasoning attached, "given that you mentioned you want something slower after a busy two days, this one suits you better than the other," is doing something categorically different from a ranked list of options. The reasoning is the product.

The fourth component is operational follow-through. The concierge removes uncertainty by closing the practical details. A platform that sends pre-trip briefings, handles logistics, confirms the meeting point, and manages disruption when something changes is doing the same work. The booking is not the end of the transaction. It is the beginning of an operational responsibility.

None of these four things require inventing new behaviour. They require building the digital equivalent of something that already works, rather than borrowing a template from a category where the underlying assumptions do not hold.

The Operators Who Were Never Going to List Anyway

There is a secondary consequence of the search box model that rarely gets examined. The operators who are best suited to the concierge model are the ones least likely to appear on a search-based platform.

An operator who fills through hotel concierge relationships and word of mouth from previous guests has no real need for a platform commission that runs at 20% to 30%. They are already full, or close enough to it. The operators who list on major platforms tend to be those with excess capacity, and excess capacity in a quality-variable category like experiences correlates with product issues more often than it should. The platforms end up with a skewed version of the available supply, and the traveller who searches gets a skewed version of what actually exists in the city.

The concierge does not have this problem. Their recommendations come from direct knowledge and a closed feedback loop. When a guest returns and says the experience was not what they expected, the concierge updates their mental model. The feedback is specific, immediate, and personal in a way that aggregate review scores are not.

A platform that curates rather than aggregates, that builds direct relationships with operators rather than scraping inventory from booking systems, and that maintains its own quality standard rather than outsourcing quality assessment to traveller reviews is replicating this closed loop. The operators on that platform are there because the platform selected them, and they stay because the channel works for them economically. That is a fundamentally different supply relationship than the one that search-based aggregation produces.

The Version That Should Have Been Built

The experiences market did not need a flight search engine with a different inventory. It needed a product that behaves like the person at the front desk of a good hotel.

A conversation that collects context before making a recommendation. A curated supply of things the platform knows well rather than everything that can be aggregated. A small number of specific suggestions with reasoning attached, not a ranked list of hundreds. Operational support through the logistics that make people hesitate. A quality standard maintained by the platform rather than delegated to star ratings.

The enabling conditions for building this did not fully exist in 2010. They exist now. The AI infrastructure for conversational context collection and personalised recommendation is usable. The operator relationships that underpin genuine curation are buildable. The gap between what experiences platforms have built and what the concierge model offers is as wide as it has ever been, and the tools to close it are available for the first time.

The search box was not a bad decision given what was possible when it was made. It is a bad decision to keep making now.