AI & Travel – Current Challenges & Opportunities

Everyone in the industry seems stuck at the same bottleneck: itineraries are easy to generate, but making them directly bookable remains unreachable. Even high-profile AI travel startups like Mindtrip, which have raised significant funding, still rely on affiliates or similar partnerships to complete transactions. The dream of a seamless, AI-driven "plan-and-book" experience hasn't materialised because the final booking step remains stubbornly manual or dependent on legacy integrations.

Currently, most major travel companies' AI efforts are concentrated on personalisation and back-end optimisation. A brief look at hiring trends at players like Booking.com shows a strong focus on improving relevance and efficiency, streamlining content decision-making, refining search results, and enhancing the underlying infrastructure that powers recommendations. These initiatives undoubtedly create operational benefits, but they are incremental improvements rather than transformative changes in how travel is discovered and booked.

The true advantage of AI in travel will be felt in the space between gross revenue and EBITDA. AI can help companies drive better margins, not necessarily by generating more bookings outright, but by reducing operational expenses and creating more integrated product experiences. Imagine tighter connections between flights, hotels, activities, and local experiences, all surfaced in ways that make cross-selling seamless. This is where AI's optimisation capabilities could generate a substantial business impact, without even having to solve the "make itineraries bookable" problem immediately.

However, there's a primary context concern. While AI models can make surprisingly accurate predictions about what a traveller might want, OTAs and traditional players only have access to booking data. They can see what was booked, but rarely why it was booked. This lack of contextual understanding makes it difficult to create truly personalised recommendations. By contrast, AI platforms like Gemini, which can access a user's full digital footprint—emails, Google Drive documents, past trip plans—have a far richer base to work from. If you've ever exchanged travel ideas over email or drafted an itinerary in a Google Doc, Gemini can surface and leverage that data instantly. Over time, tools like ChatGPT, which I've personally used daily for over two years, build a detailed, persistent understanding of how I travel, my preferences, and even my business priorities.

This presents a significant competitive challenge: how can new entrants building AI travel products overcome such a data moat? Long-term user context is challenging to replicate. Travel companies are operationally strong when it comes to booking, but have historically neglected discovery, while AI is now blurring the line between discovery and booking. The problem is that neither side is executing particularly well yet. My tests with AI booking agents, including OpenAI's, show how far we still have to go—simple tasks like searching for a domestic flight from Mumbai to Bangalore can stall because the agent can't navigate a date picker on a website.

Even where AI can create itineraries, quality is still a serious issue. Benchmarks indicate that only around 10% of AI-generated itineraries are on par with those made by experienced human agents. This is partly due to models being trained on outdated datasets—lacking the most recent travel information—and partly because travel content changes quickly. A destination visited last year might offer an entirely different experience this year. Human agents excel at navigating these nuances; AI still struggles.

The limitations extend to task performance and cost efficiency. In one published benchmark of 50 travel-specific tasks, the best-performing AI model could only complete 54% successfully. Worse, the compute cost per booking task ranged from $100 to $300. Traditional travel agents effectively subsidise their work through the margins they make on bookings. An AI agent, however, has no built-in revenue capture—there's no way to offset that cost unless the booking itself generates a profit. Without a monetisation structure that mirrors the travel industry's existing margin dynamics, it's unrealistic to expect consumers to pay such a high premium just for the convenience of an AI booking. Until both the capability and cost barriers are addressed, AI in travel will remain far more potent as an optimisation engine than as a direct booking tool.

Innovative standards or frameworks, such as the MCP, can be further enhanced to facilitate smooth and efficient access for AI to the diverse inventory of the travel industry. Can a travel-only LLM be fine-tuned to present better and process intricate travel details, making them easier for users to understand and navigate? Additionally, which company will be the first to provide AI complete access to real-time pricing and availability, adapting swiftly to the dynamic landscape of travel or radically changing it?

These are the things I'm excited to explore!

To further explore this thesis, I spoke with Hari Ganapathy, CEO & co-founder of PickYourTrail. He mentioned that the company built an internal AI agent on GPT Playground to handle repetitive but critical post-booking, pre-departure traveller queries (e.g., visa documents, currency to carry, certificates). By combining this with their own contextual data, they deliver instant, accurate answers: improving customer support, reducing the common issue of unanswered calls, and boosting brand recall.

On the operations side, they use AI-driven lead scoring (considering factors like days to departure, budget, flight pricing) to prioritise higher-quality leads in their assisted sales model. Their CRM also tracks and continuously scores calls to infer customer intent, further enhancing advisor efficiency and conversion rates.

Together, these AI interventions have:

• Improved conversion metrics year-on-year
• Increased efficiency and customer satisfaction
• Strengthened profitability
• Automated almost 50% of post-booking, pre-departure support chats through the AI agent.

In short, AI has become central in both customer support (instant query resolution, 50% automation) and sales efficiency (lead scoring, intent analysis), driving healthier conversions and profitability.

We often hear that the current AI moment is similar to the start of the internet. However, upon closer examination, the internet enabled access to customers that was previously impossible, resulting in a significant top-line impact for most businesses.

But can you say the same today? Just by being on top of an LLM or enabling sales via LLM, are you getting access to customers in a way that is cheaper than what was before? The answer would usually be a No.

If the above holds, then the actual impact of AI for some time will be in gross margin optimisation rather than top of the funnel.