From research to launch: an AI app for planning group trips

Planning a trip, especially with a group, is chaos spread across five apps that don't talk to each other: WhatsApp for the conversation, a spreadsheet for costs, Google Maps for places, Notion for the plan, and ChatGPT for ideas. And traveling alone has its own weight: not knowing who to connect with.

The market had a gap. AI planners like Wanderlog, Mindtrip, and Layla help you organize but don't connect you with anyone; social travel apps like Couchsurfing, JoinMyTrip, and GAFFL connect people but don't help them plan. Nobody combined both, and nobody did it natively in Spanish for Latin America and Spain.

Before designing, we ran a survey with more than 240 responses, 10 to 15 in-depth interviews with frequent travelers, a benchmark of more than 12 direct and indirect competitors, and desk research on solo travel, remote work, and AI.

Trust and verification also weighed heavily. Research surfaced distinct profiles, and they became our ideal customer profiles: the organizer planning for a group and the solo traveler, the planner and the improviser, the utilitarian and the social.

• A small team and a tight budget: six people, three of them developers.
• A seasonal deadline: launch had to land before the July travel season.
• Two markets, one language: neutral Spanish that works in Latin America and Spain.
• Trust without friction: identity verification with an ID and a selfie, handled with care for privacy.
• Store and legal requirements for a social app in several countries.

We could enter the market leading with community (matching travelers) or with utility (AI and organization). We chose utility because of the cold start: a social product has no value until enough people are on it, while a good itinerary is useful from the first session. The benchmark backed it, and so did the risk of promising something the product couldn't deliver yet.

The cost: a narrower message at launch, with our most distinctive feature kept in the background.

We didn't want to be "an AI app". The AI is the hook that gets someone to try Canter; the product is everything that happens after. That shaped monetization too. Instead of charging from day one or limiting the number of generations, we gated by trip length: itineraries are free up to seven days per trip, and Canter Premium (one-, three-, or six-month plans, plus add-ons for bigger groups and more active trips) makes them unlimited.

The cost: short trips stay free, so the AI earns less directly.

Image coming soon: Canter Premium paywall

The social side is the most attractive part of Canter, and the riskiest to lead with: without enough travelers it disappoints, and pitched loudly it reads like a dating app. So we planted it in the product, with joinable trips and verified profiles, communicated organization first, and were explicit that Canter is not a dating app.

The cost: the differentiator that draws the most attention doesn't do the work of attracting people at launch.

Image coming soon: Group chat and trip management

We went from discovery to defining the AI's role and the core flows, then wireframes, a prototype, and iteration. The branding and design system started with the other product designer, and we built on them together.

The AI turns a request into a day-by-day itinerary made of place cards, which you can see as a map or as a list. A trip already knows its destinations, dates, and group from the moment it's created, so asking the AI only takes what nobody else can fill in: what you want, in your own words. The app is in Spanish, so the quotes below are translated.

• For the whole trip: a sheet titled "Create a draft itinerary for your whole trip" lists the trip's destinations, all selected by default, and a free-text field that starts with "I want…".
• For a single day: each day opens a free-text field, "Write what you want to create". The app adds that day's city on its own; if the day has no city yet, it asks for one first.

There are no chips or presets: the only structured choice is which destinations to include. The group's wishes go into that same text, written by whoever generates the plan. It keeps the request fast; the trade-off is that one person speaks for the group.

Image coming soon: Asking the AI: the whole-trip sheet and the single-day field

Planning a whole trip can take up to a couple of minutes, so the wait had to feel like progress, not like a frozen screen. A full-screen state shows an animated map under "We're building your trip", with lines that rotate every few seconds ("Tracing the best route", "Researching attractions", "Fitting in distances"), a clock with the time elapsed, and a card with the trip's flags, name, and dates.

Most importantly, nobody has to stay and watch it. You can go back home and keep using the app, and Canter lets you know when the itinerary is ready, or cancel. Planning a single day is faster, so it waits in place, inside the itinerary, through three steps: "Generating itinerary", "Choosing attractions", and "Completing each day".

Image coming soon: Loading state while the whole trip is planned

Nothing the AI generates lands directly in the trip. It first appears as a preview on top of the itinerary, where you can confirm it, discard it, or ask again with different words. The confirmation bar says it plainly: "It will be added without deleting or overwriting what you already have on each day." Regenerating a day follows the same rule: it adds, it never replaces.

Once applied, an AI suggestion is a block like any other. You can change its title, description, notes, address, or icon, drag it to reorder the day, delete it, or add new places from search. Permissions apply to the AI too: the trip's admin and anyone who can edit can use it, while people with read-only access see the itinerary locked.

Image coming soon: Preview of the AI suggestions with Confirm and Discard

Every suggestion is linked to a real place from Google Places, through Canter's server, with its name, address, coordinates, and a link to Google Maps. Opening a place shows its opening hours for each day, rating, number of reviews, and photos, which gives people what they need to spot a place that's closed and remove or swap it by hand. There's no automatic check for that.

Failure has its own states. If the AI comes back empty, the app says "No suggestions" and invites you to try different words. If planning the whole trip fails, it says "Our trip got cut short" and offers to go back home or contact support.

Image coming soon: Place sheet with opening hours, rating, and photos

We soft-launched on iOS and Android around July 1, 2026, to a waitlist of 300 to 350 people, iterated on what we saw, and then opened to the public in Latin America and Spain.

Two and a half weeks after opening to the public, Canter had 300 users and 178 trips. Of those trips, 80 were public, an early sign that the social layer works even though we didn't lead with it, and 6% of profiles had already verified their identity with an ID and a selfie.

• AI lowers the barrier to entry, but it doesn't retain on its own. What keeps people is the product around it.
• Designing with AI starts with deciding its role: what it proposes, what people decide, and where it stays out on purpose.
• What people see while it generates, how they edit the result, and how errors show up are design decisions, not only engineering ones.

• Agree on ownership and roles from day one. In a small founding team, clear decision rights save a lot of time once the pace picks up.
• Don't underestimate distribution. In consumer products, getting in front of people is as much work as building the product.
• Treat AI as the hook from the start, instead of arriving there along the way.