--- title: Why I Chose AI Over a Traditional OTA description: We could have built another Expedia clone. We almost did. Here is the pivot moment that changed everything — and the risks we accepted. canonical: https://nowah.xyz/blog/chose-ai-over-traditional-ota lastModified: "2026-08-07T08:32:22.361Z" --- # Why I Chose AI Over a Traditional OTA We could have built another Expedia clone. We almost did. Here is the pivot moment that changed everything — and the risks we accepted. The original plan was a traditional OTA. Better design, cleaner interface, fewer [dark patterns](/blog/dark-patterns-flash-sales-never). We would compete on user experience while using the same fundamental model everyone else used: search forms, results pages, and a booking checkout flow. We had wireframes. We had a roadmap. We were three weeks into building it. Then we stopped. ## The dead end ![Illustration for this section](https://pics.nowah.xyz/website-media/founder-009-img-1.webp) The more we built the traditional model, the more we realized we were building a better version of something fundamentally broken. A cleaner search form is still a search form. A prettier results page still requires the user to compare fifty options and somehow make a confident decision. A smoother checkout still asks you to re-enter information you already provided. We were polishing the wrong thing. The problem was not that existing OTAs were poorly designed. Some of them are beautifully designed. The problem was that the entire interaction model, search, filter, compare, decide, was wrong for how people actually think about travel. OTA [conversion rates](/blog/low-conversion-rates-ai-fix) have been stuck at 2 to 3 percent for over a decade. The 800-billion-dollar online travel market is dominated by a UX paradigm from the 1990s. Most innovation in travel has been on the supply side, think alternative accommodation platforms, not on the booking experience itself. Building another OTA, even a really good one, would mean competing for market share within a broken paradigm. We wanted to change the paradigm. ## The threshold moment What made us bet on AI agents was the realization that the technology had crossed a usability threshold. AI models could now reliably [handle multi](/blog/how-we-handle-multi-currency-travel)-step tool chains: [search flights](/blog/launching-[tool-calling](/blog/tool-calling-at-scale-ai-travel-search)-layer-ai-agent-search-flights), compare results, check hotel availability, reason about trade-offs, and present a coherent recommendation. Earlier attempts at AI travel assistants had failed because the models could not actually do anything useful. They could talk about travel. They could not book travel. By the time we were prototyping, that had changed. The agent could take a vague natural language request, decompose it into structured queries, execute those queries against live inventory, and synthesize the results into a clear recommendation. Not perfectly every time, but well enough that the experience was already better than the 15-to-25-click traditional flow. The booking flow dropped from 15 to 25 clicks to 3 to 5 messages. That was the number that convinced us. ## The risks we accepted ![Supporting diagram](https://pics.nowah.xyz/website-media/founder-009-img-2.webp) This was not a risk-free decision. The technical risk was significant. AI outputs are non-deterministic. The same question can produce different answers. Testing is fundamentally harder. Latency is harder to control. The entire engineering discipline of building reliable software on top of probabilistic systems was, and still is, evolving rapidly. The business risk was equally real. Nobody had proven AI-first travel at scale. We were betting that users would trust an AI agent enough to let it book flights costing thousands of dollars. We were betting that the technology would keep improving fast enough that today's limitations would not be tomorrow's. We were betting on a paradigm shift that might be real or might be premature. The trust risk was the most sobering. Travel involves real money and real consequences. An AI that gives a bad movie recommendation loses a user's attention. An AI that books the wrong flight loses their money, their time, and potentially their trust in AI products altogether. We accepted all three risks because the alternative, building another OTA in a market full of OTAs with a structural 97 percent drop-off rate, felt like the bigger risk. ## Eighteen months of evidence Looking back, the bet was the right call. Not because everything went smoothly, but because the trajectory validated the thesis. Users engage with the conversational model more deeply than they would with a search form. They describe trips they would never have typed into a form. They explore options they would not have found through filters. And they complete bookings at a rate that makes the traditional OTA funnel look even more broken than the industry already knows it is. The technology has improved faster than we expected. Capabilities that were unreliable eighteen months ago are now robust. The agent handles edge cases that used to require human intervention. The accuracy of recommendations has improved measurably with every model update and every conversation in our training data. ## For founders weighing a similar bet If you are considering whether to build with AI as a core product versus a bolt-on feature, here is what I would evaluate. Has the technology crossed the reliability threshold for your domain? Not "can it generate plausible text" but "can it take reliable action?" If yes, building AI-first gives you a structural advantage over incumbents who are constrained by their existing model. Is the existing paradigm in your market demonstrably broken? If the incumbents have low conversion rates, high user frustration, or stagnant UX despite decades of optimization, there is an opening for a fundamentally different approach. Are you willing to accept non-deterministic behavior as a feature, not a bug? AI-[first products](/blog/accessibility-ai-first-products) are harder to build, harder to test, and harder to debug than traditional software. If you need deterministic outputs, AI-first might not be right yet. We chose AI over a traditional OTA because the technology was ready, the market was broken, and we were willing to accept the risks. Eighteen months in, the only regret is not making the call sooner. --- Nowah is an AI travel agent that searches and books real flights and hotels through conversation — no filters, no thirty open tabs. [Plan your next trip](https://app.nowah.xyz).