As artificial intelligence (AI) is reshapes how travelers find answers to their air travel queries, the look-to-book (L2B) ratio is exploding.
Aviation analytics firm OAG shared a chart on LinkedIn showing travel's L2B ratio on a 10x growth trajectory, with growth projected to continue over the next several years.
“In the early days of online booking, airlines processed roughly 100-200 searches to generate one ticket sale,” OAG wrote on LinkedIn.
“Today, that number is closer to 10,000-20,000. And as agentic AI begins to reshape travel planning, we project the next step to be around 200,000 searches per booking (if not more).”
That data and projection raise the question of whether the travel sector’s infrastructure can keep up with traveler expectations and AI-fueled booking demand, according to OAG.
The change comes as AI models read content across the web without users clicking through to the information source, potentially resulting in a surge in the cost of serving travelers.
And the uptick in “machine traffic per human outcome” poses a threat to airline pricing infrastructure, Jeremy Burke, chief corporate development officer for OAG, wrote in a blog post. That’s because airfare is not static content. The data fluctuates.
“Look-to-book matters, because every time somebody looks, it costs cash,” Filip Filipov, CEO of OAG, told PhocusWire. “Once this gets out of hand, you don't want the search cost to serve this inquiry [be] $200 … [when] the actual ticket is $100.”
In order to fix the problem, the sector will need to think creatively, according to Filipov.
“Maybe the entire experience changes, maybe we always start with indicative pricing,” he said.
Caching could also help address the hurdle for repeat searches made close together, Filipov added.
An ongoing discussion
Industry stakeholders have been discussing the impact of AI on L2B and how to address it for more than a year.
Juan Manuel Agudo Carrizo, VP of product for Travelport, told PhocusWire at Phocuswright Europe last month that the company is seeing “skyrocketing” L2B ratios and hosting costs.
Like its peers, the global distribution system (GDS) is considering how to address the conundrum.
“We are taking that approach and we're saying, hey, ‘what's our moat?’” he said. “We're sitting [on an immense] amount of data. And we believe the way to approach this is to [add] an intelligence layer that sits between those LLMs.”
That layer could theoretically separate intent to travel from vague queries such as when it’s better to travel or the best destinations to visit.
“We're creating preemptive servers of data to be able to entertain and guide the end user into building their trip and only open it to those more deterministic queries when we know for sure the ... moment of truth when the end user is going to book,” Carrizo said.
Carrizo believes that GDSs could add a lot more value that way, rather than creating a model context protocol server.
Last September, Madhavan Kasthuri, head of global solution engineering at Sabre, told PhocusWire that AI could actually help resolve the problem it is causing.
“We are actively extending our cache technology, but we are also using AI itself to better manage calls to that cache,” Kasthuri said.
“Where things get more complex is when personalization comes into play. In those scenarios, the traditional cache model has limits.”
“The cache is nothing but a repository of offers,” Kasthuri said. “The cache should have the intelligence to constantly go through the millions of offers that are available in its store and figure out which have expired. This clean-up process has to be done in such a way that the airline is not hit with traffic at a single point in time. So the idea is to avoid peaks of traffic on the supplier, and here is where there is another layer of intelligence which figures that out.”
Sebastien Gibergues, an independent consultant of Amadeus, added in a March 2025 op-ed that the travel industry needs to move past application programming interface polling structures.
“It is the only way to absorb the inevitable exponential traffic growth coming up soon, and it will enable flight search technology to finally step into the age of generative AI,” Gibergues wrote.