The Problem

Booking travel is high-stakes and high-friction. Research showed clearly that asking customers to commit to an entire trip in one transaction drives bounce, not conversion. Planning is fragmented: flights, hotels, cars, activities, all in separate searches, leaving the customer to hold the whole trip together in their head. The AI tools entering the space were leaning into urgency and cross-sell, which the research suggested would amplify bounce.

The brief, as framed: how do you build an AI that accelerates planning without driving urgency, integrates with the booking surface without disrupting it, and earns its place next to a search bar that already converts?

We didn't ask "how do we add AI to the booking flow?" It was "how do we replace urgency with confidence at the planning stage?"

The Approach: three positions I guided the work toward

Confidence over urgency.

The AI builds the trip with the customer so they can visualise it before committing. Chronological itinerary view, customers see the trip the way they'll live it. Save now, book later removes the all-or-nothing pressure that research showed was costing conversions.

Helpful by default, not pushy.

Cross-sell exists, but it never leads. When the AI proposes a transfer from the airport over a taxi, it explains why: distance, late arrival, luggage. Context earns the suggestion. The customer can take it or skip it without friction.

Infer first, ask only what's needed.

The AI uses what it knows (trip details, traveller profile, prior conversation) before opening its mouth. Surface tiles in the UI accelerate chat rather than replace it. System prompts encode product strategy, rules like "over 5 nights, suggest multi-location" or "confirm visa requirements before any purchase link" are commercial and compliance decisions in prompt form.

Structure at transactional moments. Flexibility everywhere else.