Luismi Design
--:-- · MAD
← Work / 03 / 06 Case study / LATAM Airlines · Concierge

Travel Planner — a reason to come back.

Concierge could hold a conversation, but not your trip. The Travel Planner gives it a memory: save any flight, stay or place — by tapping a card or simply asking — and a scattered trip assembles itself into one plan worth returning to.

Role
Senior Product Designer, AI
Team
Product, AI, Eng, Research
Year
2025–2026 · in production
Methods
Service design · Agentic UX · Research
Bet
+50% 7-day return rate to Concierge
Concierge chat proposing Caribbean destinations as saveable cards
In the chatConcierge recommends flights, stays & places as cards
A saved trip plan named Caribe 2026, with destinations and flights grouped together
In your planOne tap saves them — grouped, dated, yours
01 — The problem

A conversation with no memory.

ProblemConcierge · retention

Concierge answered questions beautifully — but the moment you closed the chat, everything it found was gone. There was no place to keep things, and so no reason to return.

7-day return rate
Belowtarget
No reason to come back
Differentiator
None
No unique value vs. chat alone
02 — Why it mattered

No memory, no reason to return.

StakesBusiness + user

For the business: the 7-day return rate sat below target, and the assistant had no unique value to pull anyone back — a real retention problem for a strategic AI product. A scan of indirect competitors like Kayak.ai exposed the opening: deliver Concierge's promise in one place.

For the user: planning a trip is complex, asynchronous and scattered — notes in one app, tickets in another, a hotel quote screenshotted in a third. People arrive at the gate already overwhelmed, and a chat that forgets everything the moment it closes doesn't help.

Give the conversation a memory, and you give the user a reason to return.

03 — What I owned

Concept, structure and validation.

ScopeService design + agentic UX

Senior Product Designer on Concierge's AI team, owning the Travel Planner concept end to end: the save-loop interaction model — card tap or natural-language prompt, both routed through Concierge's agentic layer — the plan object's structure and full state matrix, and the moderated research that validated it. Shipped to production in May 2026.

04 — The key decision

A playlist, for a trip.

DecisionUI + agentic AI

The mental model came from music: saving a recommendation to a trip should feel exactly like adding a song to a playlist. Every flight, hotel, destination and activity Concierge surfaces is a saveable card — bookmark it into a plan and move on.

There are two ways in, and they land in the same place. Tap the save control on a card, or just say it — "guarda ese hotel en mi plan." Concierge's agentic layer performs the save either way, so structured UI and natural language stop competing and start cooperating.

Save it like a song. Tap the card — or just ask.

05 — The solution

Inside a trip plan.

SolutionScreens & states

Every flight, hotel, destination and activity Concierge surfaces is a saveable card:

Concierge proposing round-trip flights to Punta Cana as a saveable card
FlightsRoutes, cabins & fares — in cash or miles
Concierge recommending a Caribbean activity, Isla Saona, as a saveable card
ActivitiesPlaces to go, reasoned from preferences
Concierge recommending hotels in Cancún, powered by Booking.com, as saveable cards
StaysHotels, scored — powered by Booking.com

Even a chat answer is worth keeping

Not everything useful is a card. A great reply — an itinerary, a comparison, a tip — can be pinned straight from the conversation into the plan's highlighted messages, so the reasoning behind a decision travels with it.

A Concierge chat reply with a Save to plan action alongside continue conversation
Guardar en PlanesPin any answer to the plan, right from the chat

The object — a lightweight container

A plan is a title, optional dates, and a stream of saved cards grouped by what they are. Whatever Concierge surfaces in chat can be promoted into it with one tap or one sentence, then read back as a single, ordered itinerary.

  • 01Title — editable, generated from the first prompt
  • 02Dates — optional; the plan reads fine with or without them
  • 03Grouped itemsdestinations · flights · activities · stays
  • 04Highlighted messages — chat answers worth keeping
  • 05One save path — card tap or prompt, both run the same agentic action
The full Caribe 2026 trip plan, grouping a destination, flight, activity and hotel with highlighted messages
One trip, fully groupedDestinations · flights · activities · stays · highlights
The plans list showing Caribe 2026, Japón soñado en primavera and Navidades en Brasil
All your tripsSwitch plans, or start a new one

Designed for the edges, not just the happy path

A planner is only trustworthy if it behaves when the data doesn't. I designed the full state matrix alongside the ideal flow — a warm first-run, an empty plan that still tells you what to do next, graceful load failures, and skeleton loaders so nothing ever flashes blank.

First-run empty state inviting the user to create their first plan
First run"Create your first plan" — one clear next step
An empty plan with per-section prompts guiding the user to add destinations and flights
Empty planEach section nudges the next save
A load-error state offering a retry
Load errorHonest message, one-tap retry
06 — Validation

Does it bring them back?

ValidationModerated · qualitative

Moderated usability testing pointed twice at the same need: users asked for a drag-and-drop calendar and a way to save and share quotes. Both were really one request — give me a container I can come back to. That request shaped the highlighted-messages pattern and the grouped plan.

After two iteration rounds — mostly sharper feedback states and clearer ownership cues — the major usability issues were resolved. Attitudes were positive, and participants named the planning hub itself as the reason they'd return.

"I want to come back and keep editing my plan." — Participant 03 · moderated test
07 — Outcome

Shipped — the bet is now tracking.

OutcomeMay 2026
Status
Live
Shipped to production · May ’26
Usability issues resolved
8 / 9
After two iterations
Retention target
+50%
7-day return rate · bet, now tracking live
08 — What I learned

A moment vs. a reason to return.

Reflection

A great answer is a moment; a place to keep it is a relationship. The save loop's real unlock wasn't storage — it was making Concierge's intelligence feel like something the user owns, not something that evaporates when the chat closes.

A smart answer is a moment. A plan is a reason to return.

© 2026 — Luis Miguel Bello García Senior Product Designer, AI contact@luismi.design
Next — 04 / 06 Tears of the Kingdom upgrade flow