Goldilocks Recommendations for Hotels: Finding Guests the Option That’s “Just Right”

In 2025 new AI capabilities are making all of us rethink user experience and the underlying technologies behind “search.” At Mobi.AI, we are packing many of these advancements into one search box. If you want to search for a hotel, you can describe any aspect of your trip that matters in natural language directly into this text box to start the search. Prompt it with “hotels in tropical destinations, hiking distance from the beach, with organic food options nearby and a max 8 hour flying time from Miami” and we can help you discover cities with hotels that match these criteria. Or maybe you’ve already planned every detail of your trip and are looking for a hotel that meets all your travel constraints. If you ask, for example, “best place with my dog in downtown Manhattan that won’t charge me hidden fees,” our search system can make an excellent localized recommendation.

We refer to this minimalist starting point and all of its breadth of functionality as Intent Driven Search (IDS) for travel. This technology works by first taking the search prompt and breaking the text down into its “intents.” On a user interaction level, intents are each idea within a travel query that expresses some kind of desire or constraint (for example, "beach" is a specific desire and "within 8 hours flight of Miami" is a constraint). On a technical level, potential intents are detected using Natural Language Processing technology. Each intent is then mapped to concepts that are more easily processed by computers like search filters, optimization problems, or planning constraints. With these intents in your trip (tropical destination, hiking + distance from beach, organic food nearby, flights from Miami), we are able to make a candidate pool of hotels. From this candidate pool, we then want to make what we call “Goldilocks Recommendations.” Goldilocks Recommendations are our way of making the IDS output as intelligible as the input. We are used to seeing search results displayed as a seemingly infinite number of pages full of candidates, ordered by some high-level idea like “lowest cost” or “most popular.” Instead of presenting an infinite list, we present a handful of options that represent the diversity of the candidate pool. So, instead of returning thousands of tropical, beachside hotels that match our “vacation discovery” prompt from earlier, we would recommend only three and explain how each one represents a different kind of trip while still matching the initial request.

How is this done? To start, Intent Driven Search provides us with a targeted and manageably sized candidate pool. We then take this candidate domain and analyze the dimensions of similarity and difference. For example, if I am looking for a restaurant in a small countryside town in Italy, I might see that on the dimension “type of food” everything looks very similar (it is hard to find anything that is not Italian). However, if we look at the dimension of “pricing,” we might see that there is quite a lot of variation within the candidate pool. How do we choose the dimensions to look at when we have a candidate pool? Mobi starts by deriving dimensions from the original prompt; for example, because “hotels in tropical destinations, hiking distance from the beach, with organic food options nearby and a max 8hr flying time from Miami” mentions flights, we could analyze the candidate pool based on different flight characteristics. This allows us to show you the best candidate destinations with direct flights, the cheapest flight deal, and the best flights for a weekend getaway. For the prompt “best place with my dog in downtown Manhattan that won’t charge me hidden fees” we can analyze the candidate pool based on dog compatibility. Maybe you are shown a hotel with designated rooms for your pet, a hotel noted in travel guides as pet-oriented, and a hotel right next to a dog park. If the candidate pool looks fairly similar along dimensions related to the prompt, we can also compare across common requests such as cost, proximity to the city, or energy level.

But why only output three options? Presenting Goldilocks Recommendations for hotels doesn’t just simplify and optimize the output response, it is a decision backed by travel and design experts at Mobi. We’re specifically pulling from research on the psychology of decision-making to come up with the number three.

Giving one recommendation, even if it is very good, does not give users a sense of agency or exploration. Two choices on the other hand makes us over-fixate on the polarity of the options. When we give three options, we either tend to feel like we’re choosing from a spectrum of choices (with a high, middle, and low option) and facilitate customers forming ideas about the “design space” they’re choosing from. For example, if Mobi provides one option that “anchors high” the user subconsciously adjusts their understanding of what trips of this type could cost, allowing us to increase our basket size and conversion rate. Why not present more? In psychologist Barry Schwartz’s book “The Paradox of Choice—Why More is Less,” he describes a concept that many Americans are very familiar with: too much choice can be exhausting and anxiety inducing. We prefer to have fewer choices—especially if the choices being presented fulfill our needs or align well with our personalities.

The underlying design and technology behind Goldilocks Recommendations is highly configurable to specific business use cases. This is because Mobi’s AI systems combine multiple AI technologies, including explainable AI algorithms. Mobi’s search system is intentionally not a machine learning black box from input-to-output, allowing us to add functionality at critical intermediate points in the recommendation generation. For example, if you are a hotel company and using Mobi’s IDS platform to let your customers directly explore options, we can seamlessly integrate and prioritize the “dimensions” you choose to show your customers into the AI system. Maybe for our beach vacation prompt, we analyze results along a “loyalty” dimension. From the pool of hotel options that match our request, we present users with the “best overall deal,” “most extras for your rewards-membership level,” and “best usage of points.”

While our previous examples have a customer directly interacting with our application for simplicity, we can also think of examples where this tool is really useful to a business or agent making a trip for a client. Let’s imagine that, instead of using a travel portal, a traveler calls a travel agent to ask for “hotels in tropical destinations, hiking distance from the beach, with organic food options nearby and a max 8 hour flying time from Miami.” A typical hotel search system does not have many of these modifiers in their advanced search filters—and even when they do, the client will be left on the line while the agent configures the search. Then, when the search has been done, the agent is probably looking at a list of hundreds or thousands of options with very little additional context. With our system, a client’s request can be directly input into the search bar without modification and the agent will get three options reflecting different aspects of a trip dimension. This allows agents to respond quickly and with some context or rationale behind the choice related to or beyond the initial request. Since these dimensions are configurable we can also incorporate dimensions that help the travel company increase profits or service quality. For example, Mobi sometimes works with Travel Management Companies (TMCs), specialized travel agencies that focus on arranging and managing corporate travel for businesses. For these customers, we can set the Goldilocks Recommendations to show their agents one recommendation that makes the TMC the most money, one recommendation that is best for the traveler in terms of convenience and duty of care, and one recommendation that is best for the client company who employs the traveler. Alternatively, when a travel agent is using the tool they can see the best three options that make them the most commission. The service is fully configurable and defined with and for our partners to show the results that they need and want to show.

Mobi’s Intent Driven Search is possible because of recent advances in AI technologies and intentional design decisions made by our team of travel experts. It is even more powerful when paired with Mobi’s exceptional and extensive datasets and planning/routing engines. Whether your guest is trying to find the perfect local hotel or looking to discover new places to stay—Mobi’s IDS tool elevates your search process.