Travel Inspiration Lives on TikTok. The Booking Still Lives in 47 Browser Tabs.

From the Travel Tech Podcast newsletter for episode 23, Travel Inspiration Lives on TikTok. The Booking Still Lives in 47 Browser Tabs..

A friend who has actually been somewhere will always beat an algorithm that has indexed everywhere. That is not a sentimental observation. It is a product architecture decision, and it turns out it has significant commercial implications for how $355 billion in annual travel commissions gets distributed. Many of you across aviation, travel tech, and AI care about exactly that kind of structural shift, and I wanted that perspective on the show.

My guest this week built the AI platform behind the Meta Ray-Ban glasses, ran enterprise AI partnerships at Microsoft, and served as CPO at a company deploying AI into physical world operations before deciding that the most interesting real world AI problem was travel. She started Boop after a frustrating honeymoon in Venice with 1,600 photos and no efficient way to share a single one of them as a bookable, actionable experience. She leads a team of 10 in Seattle and launched public beta 90 days ago. Listen on Apple Podcasts, also streaming on Spotify and YouTube.

The central tension running through the conversation is this: travel discovery already moved to social, but the booking infrastructure never followed. The gap between where intent is created and where the transaction happens is costing both travelers and smaller hospitality providers in ways that are still largely unaddressed.

The billion pieces of travel content that all end in a dead end

That tension has a structural cause. There are roughly a billion pieces of travel content on TikTok and Instagram. Every single one ends the same way: a screenshot, a Google Maps search, a separate hotel lookup, a price comparison, a WhatsApp thread, maybe an Excel sheet, and eventually a booking made on a platform that had nothing to do with where the original inspiration came from. The hotel that inspired the post often loses on its own brand keywords to Expedia or Booking.com, paying 15 to 25% back to the same platforms that outbid it. The person who inspired the trip earns nothing. The experience provider that made the trip memorable may not even be discoverable online, since roughly 80% of global experience providers have no online booking presence at all.

AI has not fixed this. It has mostly reproduced it at scale. The early version of any travel AI recommendation, built on general internet data without a verified source layer, forces the user to do the same trust evaluation they would do on any anonymous review: why did it recommend this, is this place actually open, is this the best option for my group? What does cut through is a real friend's verified camera roll, anchored in actual GPS metadata, with photos and booking links attached. That is the architecture Boop is building: an on-device AI that processes photo metadata locally, extracts trip structure without uploading anything to external servers, and connects the resulting itinerary to live pricing and availability across hotel, activity, and experience providers. The data from 90 days of public beta is notable. The most copied trips are not from accounts with the largest followings. A Chicago flight attendant's eight day Japan itinerary and an Amsterdam local's guide built around a Beyonce concert tour stop both significantly outperformed content from influencers with millions of followers.

This week I spoke with Nancy Li Smith, founder and CEO of Boop, about on device AI and privacy architecture, the 50/50 affiliate commission split that pays itinerary creators, the agentic universal cart the team is building to consolidate hotel, experience, and niche local provider bookings in a single transaction, why 50% of trips booked on the platform are four days or fewer, and how sports clubs, event brands, and small hospitality operators are starting to use fan-generated itinerary libraries as a distribution channel.

Key Takeaways

• Human curation is now the moat, not the feature: As general purpose AI travel recommendations become commoditized, an itinerary grounded in a verified real world experience from a trusted source consistently outperforms one generated from anonymous internet data, because it eliminates the user's need to evaluate the source.
• The commission stack can be redistributed without disintermediating providers: Boop's model generates affiliate commissions across hotel, experience, and activity categories and splits them 50/50 with the original itinerary creator, while keeping the underlying hospitality provider as merchant of record.
• Intent capture is an infrastructure problem, not a content problem: 250 million pieces of existing travel content already use language like "steal my itinerary" or "copy my trip." The missing layer is a transaction mechanism attached to that content at the point where intent is created.
• The experience economy gap is larger than the accommodation gap: Activity and experience providers capture only around 20% of actual experience bookings globally, leaving the majority of small operators outside any discoverable booking surface entirely.
• Frequency is the underrated metric in next-gen travel: Platform data shows 50% of trips are four days or fewer, suggesting the target audience is planning significantly more frequent short trips than the industry's annual-vacation framing implies, with 80% of users who copy one trip going on to plan a second.

🔧 Three to Build

Nancy Li Smith's argument is simple: inspiration is social, trust is personal, and booking tools are still 47 tabs behind both.

Build an on device personalization layer that never sends raw user data to your servers.

Federated learning lets you train recommendation models across millions of devices while the sensitive inputs, camera roll metadata, location history, browsing patterns, stay local. Each device contributes gradient updates, not personal files. For an airline or OTA trying to improve inspiration to booking conversion, this means you can personalize at scale without accumulating the liability that comes with a centralized data lake.

Catch: your model quality depends on update frequency, and users who rarely open the app contribute almost nothing, which quietly biases the model toward your most engaged segment.

→ Privacy preserving federated learning for passenger data

Build a social trust graph that ranks creators by downstream booking influence, not follower count.

A Chicago based flight attendant with 4,000 followers whose itineraries get copied 800 times is more valuable to your distribution strategy than a travel influencer with 400,000 followers and zero conversion. Network analysis on copy, share, and save behavior surfaces those nodes. You can then build targeted outreach or affiliate tooling aimed at the people who actually move bookings, before a competitor does.

Catch: the highest influence nodes are often everyday creators with no media kit and no interest in a formal partnership, so your outreach motion has to be lightweight or they ignore you.

→ Social network analysis to identify brand advocates

Build a conversational layer that lets users remix a copied itinerary into a confirmed booking without leaving the chat.

Most inspiration products stop at the point of desire. The user has a saved itinerary but faces a context switch into search, comparison, and tab hell. A conversational AI agent can hold the itinerary as structured context and let the user say things like "make it four days, skip the museum, add a dinner option under 60 dollars" before handing off to a booking flow with those parameters already set. The agent is not replacing search. It is collapsing the gap between a copied plan and a committed purchase.

Catch: the hardest part is not the conversation, it is keeping the itinerary data model rich enough that the agent can reason over it rather than just echoing text back.

→ Conversational AI for personalized booking assistance

Alex Brooker, Host of the Travel Tech Podcast, Founder of Airside Labs