Designing a Conversational Interface to Make Travel Planning Easy

Building a startup is not the usual UX/UI brief — and that is the point. Affordable travel is a wicked problem: product, tech, business model, and brand all pull on each other. I chose to hold that tension in the design process rather than pretend UX lived in a silo — otherwise we would have shipped a pretty search grid that ignored how the company had to survive.

We fell in love with the problem before the product. I mapped how people search for, choose, and order travel online — then dug into choosing, not searching or checkout. Searching was noisy and site-specific; ordering was hard to observe without a real purchase. Choosing was where preference showed up: time spent comparing, which factors won, what was merely nice to have.

Affordability beat convenience. A stable connection along the route mattered for people working from the road. And the planning styles split: many backpackers map a trip start→finish; digital nomads often plan fluidly mid-trip — long stays, then decide the next country by flight, train, ferry, or bus. That split mattered for product shape later: a tool that only helped "book the whole tour upfront" would miss the nomad.

I chose to fill the customer canvas before the value proposition. The offer side stayed empty on purpose — a design decision against jumping to features. Untested ideas (like whether layovers could feel like a gain) stayed marked as hypotheses, not requirements.

Dark mint with a "hack the system" vibe read clever to us and cold to travellers. I chose against that direction once research made the miss obvious: we were signalling fare-gaming, not going somewhere. Clever-for-founders is not the same as usable-for-travellers.

We rebuilt around adventure and affordability with quiet trust: warm orange and teal, paper texture, modern-analog. The bird with a leaf-wing was a deliberate tradeoff — flight and lightness without adrenaline cosplay, so safety and openness could sit in one mark. Then I pushed past mood boards into a real component system: states, interactions, consistency across flows. I implemented that system in React/Vite while my co-founders owned backend and data — so design intent and frontend behavior stayed one conversation, not a handoff document.

CheapVoyage 1.0 was a shipped product chapter, not a brand exercise. The limit of 1.0 was already visible: even with the right mood, we were still competing as a search surface.

By spring 2025 we were still on a fare-grid path. In summer 2025 Google Flights launched Flight Deals — essentially the surface we were building. Three founders cannot win an arms race against Google on that UI pattern.

The competitive shock forced a product decision, not a rebrand. We could have doubled down on "slightly different search" — better filters, nicer cards, more modes. We chose not to. Differentiation had to be the experience of planning itself.

The deeper reframe: the burden isn't finding one cheap ticket. It's months of lifestyle travel as a second job — especially for digital nomads deciding the next country mid-trip. CheapVoyage 2.0 had to lighten planning, not sharpen the grid.

I chose a conversational interface over another advanced form. Forms force travellers to know too much too early: exact cities, exact dates, exact modes. Conversation lets intent arrive incomplete — where you are, where you might go, for how long, flights or trains — and stay editable mid-thread.

Lightness was a hard constraint, not decoration. Every chrome addition competed with the traveller's mental load. Complexity had to stay in the system: routing, comparisons, data freshness — not in the UI's demand for perfect inputs. "Where to next?" is the product promise in one line: forward motion without a locked itinerary.

I scoped to Europe on purpose. Intercontinental ambition looked impressive and made the product worse: weaker rail, messier data, and a promise we couldn't keep as three founders. Europe is where our users already were and where rail is a real alternative, not a novelty.

Fly versus train had to live in one thread, including sustainability beside price — otherwise "cheapest flight" would always win by default. One conversation can hold a whole loop (Copenhagen → Paris → Amsterdam → back) and land as inspectable legs with plane and train options per segment.

Inspectability was the trust decision. Magical chat that hides the itinerary fails the first time a price or connection is wrong. I treated the model as a compositor, not a source of truth: every suggestion points at a concrete itinerary; no false precision; no locked destination until the traveller is ready. People forgive a slower answer if they can see why it was chosen.