A post-purchase travel experience that turns booking data into tailored next steps for each traveller.
Same booking system. Different traveller needs. Different checkout output.
Every day there’s a new AI tool, and most of them want you to know they’re AI. Chatbots, magic generate buttons, prompt interfaces. This project goes the other direction.
What if AI could reshape the content of a webpage based on who’s looking at it, without the user ever typing, clicking, or noticing? No prompts. No chat windows. Just a page that quietly adapts its language, priorities, and recommendations based on what’s already known about you: your age, destination, travel purpose, group type, experience level.
This project applies that idea to one specific moment: the post-purchase checkout page on travel platforms. The moment where personalisation usually disappears, and where it matters most.
The user never types a query or interacts with an AI interface. The system works without any input beyond what’s already known.
There’s no chatbot, no assistant panel, no “Ask AI” button. The AI operates behind the page, shaping what appears.
The page reads your booking data and generates content specific to your trip, your background, and your needs. It simply knows you.
From a 7-week university sprint to an independently rebuilt proof of concept.
This started as a university course exploring AI in design. The brief was open-ended: find a real problem and investigate how AI could meaningfully address it. I was responsible for every stage, from research and concept framing to system logic and technical build.
The aim wasn’t to create a polished product. It was to push the boundaries of emerging technology and explore how AI could unlock new opportunities for context-aware, user-adaptive design. Seven weeks, start to finish.
10 post-purchase confirmation pages from major travel platforms. All generic. All identical regardless of who booked.
Same booking system. Same UI. Completely different content, adapted to Dewi’s family trip to Tokyo.
You’ve just booked a trip. You’re excited. Then you hit the confirmation page: “Your purchase is complete.” A booking reference, a generic email, maybe an upsell for travel insurance. All that excitement, flattened into a transaction receipt.
Travel platforms know an enormous amount about you at this point. Your destination, dates, group type, where you’re flying from, how experienced a traveller you are. But none of that data shapes what happens after you click “book.” Personalisation powers the search and pricing engines, then vanishes the moment the transaction is complete.
The post-purchase moment is where traveller anxiety is highest (what do I need to prepare?), where context is richest (the platform knows exactly what trip you’ve booked), and where current platforms deliver the least value. Confirmation screens are designed for verification, not for the traveller. That gap is where this project sits.
With a 7-week timeline, there wasn’t space for extensive primary research. Instead, I mapped the data travel platforms already hold and analysed how they use it, or don’t.
The finding was clear: the issue isn’t data scarcity, it’s dormant data. Travel companies already have rich contextual information about every booking, but almost none of it gets activated after purchase. It’s used heavily for search, sort, and pricing. Then it stops.
Deep search personalisation, but purely transactional after booking.
Strengths
Gaps
Human touch through hosts, but no embedded AI layer.
Strengths
Gaps
Strong ecosystem data, but instrumental tone and no tailored guidance.
Strengths
Gaps
Cross-brand data sharing, but generic confirmation and deal-first language.
Strengths
Gaps
AI shouldn't always be a chatbot or a button. It can quietly shape what the page says and shows, based on who's looking at it.
The confirmation and pre-trip moments are emotionally loaded and under-designed. They're the best place to reinvest personalisation.
The opportunity isn't new data pipelines. It's using what platforms already know (age, trip purpose, destination, timing, group type) to generate context-aware content at scale.
This gap is exactly where this project sits: using AI to personalise the moments after purchase. The confirmation, the lead-up, and the support no platform currently designs for.
Use AI as an invisible layer that reads existing booking data and generates a personalised post-checkout page. One that tells each traveller what they specifically need to know, based on who they are and where they’re going.
A first-time family from Jakarta heading to Tokyo in winter needs visa guidance, cold weather prep, and kid-friendly logistics. A solo photographer revisiting Kyoto in spring needs insider cultural spots and efficient transit tips. Same system, same UI, completely different output. That kind of contextual adaptation at scale is what AI makes possible.
The system starts with what the travel platform already knows about you. Your destination, dates, where you’re flying from, group type, experience level, interests, and budget. In a real platform, this data comes from the booking flow and customer profile automatically. In the demo on this page, the form makes that data visible so you can test how different inputs change the output.
The front end collects traveller context through a structured form (11 fields: name, destination, origin, dates, length of stay, age range, purpose, experience level, budget, group type, interests). On submit, the form data is validated client-side (destination is required, all others optional) and sent as a POST request to a Next.js API route at /api/generate. An AbortController handles race conditions if the user resubmits before a previous request completes.
Using the combined input, the AI builds a richer picture of who you are as a traveller. A family from Jakarta visiting Tokyo in December isn’t just “family, Tokyo.” The model infers that they’re likely first-time overseas travellers from a tropical climate heading into winter, probably needing visa guidance, cold-weather clothing advice, kid-friendly transport options, and halal food recommendations. It identifies the gaps between your background and your destination.
The API route calls Claude Haiku (claude-haiku-4-5-20251001) through the Anthropic SDK with a structured prompt. The prompt instructs the model to perform three reasoning steps in sequence: first, expand the raw traveller data into a rich persona (inferring cultural context, climate familiarity, experience gaps, and likely concerns). Second, identify what this specific traveller probably doesn’t know about their destination. Third, generate structured content addressing those gaps. The model returns JSON matching a strict schema: a title, subtitle, trip info summary, and exactly three expandable cards, each with a heading, summary, tag, and detailed sections containing bullet points and tips.
The system generates a customised confirmation page. Instead of a generic “booking confirmed” screen, the traveller sees a personalised header addressing them by name and trip context, a trip summary card, and three expandable recommendation cards covering the specific areas most relevant to their situation. Different travellers seeing the same UI get completely different content, tone, and priorities.
The API route validates and normalises the Claude response server-side before returning it as { ok: true, checkout: CheckoutData }. The front end receives the structured JSON and renders it through a shared CheckoutOutput component: a browser-chrome-framed card layout with a confirmation header, trip info grid, and three expandable cards. Each card uses flex-grow animation for smooth expansion. The same rendering component is used by both the preloaded hero demo (with mock data) and the live generator (with API data), ensuring visual consistency. Icon names in the response are validated against a known set and fall back to a default compass icon.
The biggest question for this project: is the system generating genuinely personalised content, or just generic travel copy with names swapped in?
To test this, I created persona checklists. For each test traveller, I outlined what I expected to see and what should be absent, based on their specific background, experience level, and trip context. Then I compared the AI output against those criteria. This turned subjective impressions into something observable and repeatable.
| Persona | Profile | Key test | Verdict |
|---|---|---|---|
| Dewi | Family, first overseas trip, Jakarta → Tokyo, December | Does it provide family logistics, visa guidance, winter prep, and cultural food comfort? | ✅ Strong pass. Visa info, kid-specific transport, halal dining, and winter clothing all addressed. |
| Jordan | Solo photographer, frequent traveller, Sydney → Kyoto, April | Does it skip beginner guidance and provide insider cultural picks and efficient logistics? | ✴️ Partial pass. Strong on cultural curation, but includes some basic info a frequent traveller wouldn’t need. |
| Harold & Margaret | Retired couple, comfort-first, Melbourne → Paris, September | Does it prioritise accessibility, comfort, guided experiences, and avoid tech complexity? | ✅ Strong pass. Comfort transport, timed entries, calm tone. Some accessibility detail could go deeper. |
Across all three personas, the system demonstrates credible personalisation range. Dewi gets guided reassurance and family logistics. Jordan gets insider cultural picks with minimal hand-holding. Harold and Margaret get comfort-first planning with accessibility awareness. The system clearly adapts tone, content priority, and level of detail based on who’s travelling.
Where it falls short: some outputs include information the persona likely already knows (giving a frequent traveller basic airport transfer advice), and deeper accessibility cues are implied rather than made explicit. These are refinement issues, not conceptual failures. The core personalisation works. Future iterations should sharpen the model’s ability to omit what experienced travellers don’t need and deepen contextual sensitivity for accessibility requirements.
Enter a traveller profile and generate a personalised post-checkout travel summary using the live AI system.
Fill in your trip details and hit Generate to see your personalised summary
The concept is validated. Here’s what comes next.
The persona checklists demonstrated that the system genuinely adapts its content, tone, and priorities based on traveller context. Dewi gets family reassurance. Jordan gets insider cultural picks. Harold and Margaret get comfort-first planning. The core concept works. The question is no longer whether AI can personalise a checkout page, but how to refine and scale it.
This project turned a 7-week university sprint into something I genuinely believe in. It’s not a chatbot and it’s not a recommendation engine. It’s a quiet layer that makes the web feel more considered. That’s the kind of design problem I want to keep solving.
