Why its hard to disrupt the OTAs $BKNG $EXPE $TRVG
retained text
*There's a 56% chance users will leave money on the table by delegating vacation to AI* \- Autonomous AI agents excel at purchasing standardised commodities where preferences are relatively fixed upfront. They **struggle with high-stakes experience goods**, like family vacation hotels, for three reasons: \- Preferences are constructed, not predefined. Travellers discover what they want by browsing and comparing actual options. Expedia found that travelers **browse 141 pages and spend 303 minutes** researching in the 45 days before booking. Search is part of preference formation, not just execution. \- High-dimensional preferences compound prediction errors. Hotel choices involve dozens of subjective attributes: location, views, room layout, ambience, amenities, etc. Even assuming an illustrative 85% match on each attribute, **the probability of matching all five is just 44.4%**; across eight dimensions, it falls to 27.2%. Ellis & Huang (2026) also found AI predictions using actual past choices were more accurate (74%) than those using written prompts (71%). \- The cost of getting it wrong rises with the stakes. Worldpay's survey of 8,000 consumers found that 23–31% would trust an agent with travel purchases of £101–£500, falling to 19–24% up to £1,000 and **just 2–3% above £5,000**. Meanwhile, research across 200 London hotels found persistent price differences between booking platforms. Delegating a high-value booking without thorough cross-platform discovery risks both a poor experience and leaving money on the table. Disc: I am long $TRVG
document fetched 2026-11-04 00:44 · canonical https://reddit.com/r/ValueInvesting/comments/1xmq0ed/why_its_hard_to_disrupt_the_otas_bkng_expe_trvg