Why AI travel scenarios miss the $3T opportunity

Mario Gavira's recent piece on four scenarios for agentic artificial intelligence (AI) in travel offers a compelling glimpse into how the industry envisions its future. But it also reflects a familiar blind spot: thinking about AI disruption from inside the industry bubble rather than understanding where travelers actually begin their journey.

The scenarios range from online travel agents (OTAs) adding chatbots to autonomous AI agents negotiating directly with suppliers. What they all miss is the fundamental question: Where do travelers discover and plan trips today?

Spoiler: It's not on Expedia.

The OTA fortress isn't going anywhere

Let's get one thing straight—OTAs aren't getting disintermediated anytime soon. Gavira's scenarios three and four, where suppliers bypass OTAs entirely, ignore the brutal economics of travel distribution.

Here's what OTAs actually excel at: single-item booking—one hotel stay, one flight, one experience. The classic "search, detail, price, compare, book" workflow that dominates travel today.

Booking.com commands 69.3% of the European hotel OTA market, while Expedia holds 11.5%. In the United States, these two giants account for roughly 93% of the core OTA leisure market. Their scale isn't the result of breakthrough tech—it's operational mastery at brutal scale. They solved the hardest problem in travel: aggregating fragmented inventory while handling payments, cancellations and customer service across thousands of suppliers.

Eighty percent of travelers visit OTAs to research and compare prices, and OTAs spent $17.8 billion on marketing in 2024 to maintain that traffic dominance.

But try to plan a complex, week-long itinerary using an OTA? The systems completely break down.

Even current large language models (LLMs) break down under this weight. OpenAI's most advanced model achieves only a 10% success rate on complex travel planning benchmarks, while earlier LLMs manage less than 1% accuracy compared to humans at 100%.

Travel planning happens in communities, not comparison sites

Here's what these scenarios fundamentally misunderstand: Travelers don't start their journey on OTAs. In fact, 89% of travelers turn to social media for travel inspiration, with 75% relying on social platforms to discover destinations.

They start in Facebook groups asking, "Where should I go for my anniversary?" They browse Instagram stories from friends' trips. They ask for recommendations in Slack channels and Discord servers. Importantly, 55% get inspiration from family and friends' accounts—far more than the 18% who follow celebrity influencers.

During my time at TripAdvisor, this pattern was crystal clear: Most of our traffic was using our reviews to validate travel choices people had already made elsewhere. We weren't the starting point—we were the confirmation step. OTAs follow the same pattern. People aren't browsing Booking.com to discover their next destination. They're going there to book a hotel they've already settled on as part of a trip they've already planned elsewhere.

What's happening in those "elsewhere" conversations is "anti-planning." Instead of search-driven queries, people are dreaming out loud. They're sharing half-formed ideas, describing ideal experiences without concrete constraints.

"I want to go somewhere with great food and history, maybe Europe, in the fall, with my partner who hates crowds..."

That's not a search query. That's a conversation containing far more valuable intent data than any keyword-based search. Of travelers who use social media for trip planning, 62% make specific decisions after viewing content.

The gap between this kind of dreaming and actually having a bookable itinerary is massive.

The technical reality check

What strikes me most about these scenarios is the assumption of travel tech infrastructure that simply doesn't exist.

Gavira envisions AI agents from airlines and hotels conducting "complex, instantaneous algorithmic negotiations." Have you tried to change a United flight recently? Some airline legacy components date back to the 1970s. Airlines globally spend $37 billion annually on IT—much of it just maintaining infrastructure held together with "rubber bands and crossed fingers."

These systems were designed around batch processing and physical paper documents. Customer records aren't even in one place. Most hotel application programming interfaces can't handle room preference requests consistently.

Yet, we're supposed to believe these same companies will suddenly deploy sophisticated AI agents that can negotiate dynamic pricing and custom packages? The technical debt alone would take years to unwind.

The real disruption play

Rather than chasing OTA disruption, the smart play is building AI that works where travelers actually are—in communities and social platforms—and can handle the complexity that breaks current systems.

The technical challenge is significant. You need AI that can:

• Monitor and capture the nuanced conversations where travel dreams are shared
• Break down complex, multi-week itineraries into manageable components
• Manage the combinatorial weight of coordinating flights, accommodations, timing, experiences and logistics
• Reassemble everything into executable plans that actually work

This isn't just a user experience problem—it's a computational architecture problem. You're building a travel planning compiler that can decompose complex intent into bookable components, then optimize the entire system.

Current OTAs can't do this because their systems are built for single-item transactions. Standard LLMs can't do this because they lack persistent state management for complex travel logistics.

But here's the key insight: Once you solve this technical problem, the data becomes incredibly valuable. A detailed, coordinated week-long itinerary isn't just a travel plan—it's a high-value product that multiple parties will pay for.

A two-sided marketplace opportunity

Build a platform that uses AI to create complex travel plans from community conversations, then operate as a two-sided marketplace for that high-value data.

On one side, you're capturing travel intent from communities and social platforms, using AI to transform vague dreams into detailed, coordinated itineraries that individuals couldn't create themselves.

On the other side, you're distributing these plans to multiple buyers. OTAs want qualified leads with higher gross booking values. But more importantly, travel agencies and specialists want access to the detailed planning work—they know how to take a sophisticated itinerary and execute it profitably.

A detailed week-long Japan itinerary with optimized train schedules, restaurant reservations and experience coordination is worth far more to a boutique travel agency than a simple hotel booking is to an OTA.

You're not trying to replace the booking infrastructure—you're creating more valuable inventory to flow through multiple channels while aligning incentives instead of fighting them.

Why this actually makes sense

The travel industry has always been about intermediation, not disintermediation. The most successful companies succeed by creating new forms of valuable intermediation.

There's a genuine technical moat available to whoever builds specialized AI that can capture conversational travel intent, decompose it into coordinated components and manage the complexity of reassembly. Current systems simply can't handle this computational problem.

I remember trying to plan a complex trip across multiple Japanese cities with my family. We spent hours coordinating train times, hotel locations near stations, restaurant reservations that fit our schedule and activities that worked for both adults and kids. Each decision cascaded into a dozen others. No single platform could handle it all.

That's exactly the kind of complexity that AI could solve—but only if it's built specifically for this problem, not bolted onto existing booking systems.

The real disruption

While everyone's debating whether AI will replace OTAs, the actual disruption will come from whoever figures out how to use AI to create more valuable travel products in the first place.

That company won't look like Gavira's scenarios. It won't be an autonomous booking bot or an incremental improvement to existing OTA search.

It'll be a platform that meets travelers where they actually are, uses AI to help them envision better trips than they knew were possible, then efficiently distributes that value through the existing travel ecosystem.

Full disclosure: I'm an engineering advisor to Globe Thrivers, which is building exactly this model under CEO Shir Ibgui. They're creating a community-driven platform that captures conversational travel intent and uses AI to transform it into detailed, bookable itineraries. Their B2B2C approach already has interest from enterprise partnerships—proving that this two-sided marketplace model resonates with both travelers and the industry.

The winners won't replace the OTA fortress. They'll render it obsolete by turning inspiration into itineraries before the first search bar is touched.

This story original appeared on Matsuoka's website, Hyperdev.