AI agents and the future of online travel agencies

Most of the largest companies in the world are platforms. Most of the growth has been in the last 15 years, driven by the internet, which accelerates their fundamental advantages over the pipeline model.

Online travel agencies are the platform play of the travel industry. The principles are the same as any other sector. So what makes the platform model so much better than what we had before?

• Economies of scale
• Network effects
• Low marginal costs - no inventory to keep
• Innovation by suppliers on the platform. Supplier product or feature innovations become fresh ideas.
• Diversified revenue streams. Additional verticals within travel, payments, media sales etc.
• Optimized data feedback loops

The first two are the key. Economies of scale are crucial — where each individual travel site has to maintain its own website, marketing, sales process, customer service department, tech stack etc, the OTA can do this once across millions of products. It's not a bit cheaper to operate, it is thousands of times more efficient. 

The network effect is what turbo-charges platforms. The more products they have, the more customers they appeal to. The more customers they have, the more products they can onboard and at optimal commercial value. This then feeds back into economies of scale. Non-platform businesses can’t easily compete with this loop.  

These drivers allow scale which leads to a very attractive offering to customers. It brings broad product offerings, great user experience, marketing budgets, impactful branding strategies, efficient payments and expansive distribution. OTAs gain a massive advantage in the core areas of distribution and customer trust.

Agents enabled

What’s an AI agent? Let’s be clear, there is no agreed definition. In the very broadest definition, you can argue that ChatGPT browsing for results is a type of agent. At the other end of the scale, and where it gets interesting, is when agents can complete complex and multi-step tasks, often with some reasoning required in decision making as part of that process. The agent will represent an individual or an organization, and will complete increasingly complex tasks on command.

In many ways an AI travel agent doesn’t need to be overly complex. It’s 100% feasible with today’s technology. So as ever improving LLMs and agents come to market, how will they affect the complex travel transaction landscape that is today largely owned by OTAs.

First of all, some things won’t change. A lot of travel discovery is enjoyable for humans. Here in the dreaming phase, things will adapt more slowly. Some humans will choose not to use agents at all and stick to their old habits.

Also, the booking process isn’t particularly broken today. If pushed, I bet I could make a decent choice for a hotel in any location in the world on a good OTA, and check out within a minute. For a flight I might need an extra 30 seconds (and that’s just for baggage regulations).

But that being said, agents are coming. Large language models (LLMs) will keep advancing. Companies will spend billions of dollars on agents of all types to see what sticks. 

The agents, with a quick text or voice prompt, can conduct a search for any travel product from front to back, requiring input only when the human chooses to. They can browse the internet, access APIs, talk to other agents, and access third party applications (if given credentials by the human). So any product that can be booked online today can be booked by an agent, without the supplier having to do anything at all. That’s all possible today — although it’s imperfect right now.

Initially, agents will be given access to APIs, and will browse the internet to fill gaps. The agent will also look for the new middleware agents, which will pop up to aggregate direct supply links. So instead of a human looking at 38 websites and maybe comparing prices on three to four, my agent will check every product bookable on the internet, serving me a personalized selection of all possible options, from all sources.

The agent ends up with a list of choices, which might look a little bit like a metasearch result. It compares all parameters and prices. All things being equal, the lowest price gets the booking. But my AI agent also filters out products or sites that might seem risky. It looks at all upgrade options, analyzes all loyalty programs to break them down to an exact cash value, and it looks at specials and offers around all of the booking paths. It completely ignores the brand. It makes a choice, assisted by me, until I work out that it’s better at making decisions than I am.

So what happens to all of the OTAs' natural advantages?

Economies of scale. Gone. The AI agent has all of the scale from day one. It can book any travel product available online. Each single OTA is just a subset of that. The agent doesn’t need product, contracts, payments, customer service, or any of that messy stuff. Monetization will be easy — there’s plenty of commission for the agent to negotiate during the 1.5 seconds it takes to make the booking. It only has to cover the cost of a few tokens used, either directly, or by the other AI agents which assisted. 

Network effects. Also gone. The AI agent doesn’t need this loop to grow.

Low marginal costs, innovation by suppliers on the platform, diversified revenue streams, optimized data feedback loops, and all of the other principles — almost all vanish overnight. 

It’s a pretty dire situation for an OTA, or any other intermediary.

In the short term, the OTAs might be those early trusted sources. They have the granular data that the direct websites won’t have for a while. They have the payment methods and the lower rates. But very quickly, there will be agents with all of the direct sources of those products and without the human having to go through any pain to find them. Suppliers will work this out and give them what they need.

The fundamental core value prop of trust is pretty much zero as soon as humans aren’t in the loop. The AI doesn’t see marketing messages and persuasion taglines. It doesn’t care about the nice rounded blue button, with “buy now” in a special font that has been split tested over 20 years. It doesn’t even notice the slick new $2 million logo.

Agent immunity

Of course, the answer is that OTAs will pivot. None of this is going to happen overnight. The OTAs will have many months to revolutionize their entire business model and find a new one immune from the AI agents. But what is immune from AI agents? The connected trip would have been a good one. Too much talking, not enough doing on that one. The problem is you could think of 100 potential pivots or moves that they could make, but finding just one that makes some money and can’t be steamrolled? Not so easy.

Maybe a part three is required for Google. In the same way that the OTAs are transaction platforms, Google is in exactly the same position on the discovery phase of travel. All intermediaries are in trouble.

Personally, I think most of the OTAs will find a way. But in five years, will they still be recognizable as the OTAs of 2024? Probably not.