Luismi Design
--:-- · MAD
← Work / 02 / 06 Case study / LATAM Airlines · Concierge

Continuous improvement of access to Concierge.

LATAM's in-app AI travel assistant lived behind one card on Home — and almost nobody tapped it. This is how a content-and-accessibility pass, shipped without a single new component, multiplied access and first messages.

Role
Senior Product Designer, AI
Team
Product, Eng, Data, Content
Year
2025 — shipped
Methods
Content design · A/B · Accessibility
Outcome
Up to +102% access · +66% first messages
Concierge — LATAM Airlines' in-app AI travel assistant, shown on the Home tab of the LATAM app
01 — The problem

One card. Almost nobody tapped it.

Problem2025 · Q1

Concierge — LATAM Airlines' in-app AI travel assistant — lived behind a single card on the Home tab. One button on that card was the whole way in.

Baseline access rate
0.5%
Logged-in users · Home card
Worst-case peer
10%
Reference banner on Home
02 — Why it mattered

The ceiling wasn't the assistant. It was the invitation.

StakesBusiness + user

The worst-performing reference banner sitting beside the Concierge card was still clearing 10% — ten times the card's rate, for a far less useful feature. That gap said the problem wasn't Concierge's quality; it was that almost nobody knew to start.

For the business: Concierge was a real product bet with engineering behind it, returning nothing while the invitation kept failing. For the user: travellers who'd have gotten real value from the assistant — inspiration, flights, a place to plan from — never saw a reason to open the conversation.

03 — What I owned

Content, contrast and the suggestion strategy.

ScopeSix weeks, no new components

Content design across every Concierge entry point on Home and Shop > Flights — the title, description and CTA on each card. The accessibility fix on the access button. And the quick-reply content strategy: mining chat intents in BigQuery and writing the suggestion sets for each user state.

The constraint was real — six weeks, no new components, minimal engineering. Whatever shipped had to be words and pixels on a card that already existed.

04 — The key decision

Replace the description with quick replies.

DecisionIntervention №2

Inside the chat, suggested prompts were already driving 40–50% of all interactions. So we asked a sharper question: what if the card showed those quick replies instead of a paragraph of description? Suggestions don't just save typing — they teach people what Concierge can actually do.

Smaller surface. Less risk. More leverage.

05 — The solution

Rewrite the message, raise the contrast.

SolutionScreens · before & after

Three moves, each picked because it needed no new component. A content pass on every card's title, description and CTA — drafted with an internal content assistant trained on our voice & tone, a golden set of vetted responses, and outside content-design references. A copy change on the Shop > Flights access button. And an accessibility pass on the button itself, whose tri-colour gradient was failing AA contrast for the label sitting on top of it.

Original Concierge card on Home — old title and generic call to action
Before — Home, defaultQuestion headline, generic CTA, paragraph of description
Redesigned Concierge card on Home — new title, sharper description and conversational CTA
After — Home, defaultConfident headline, tighter description, conversational CTA
Title — Home, default

"¿Buscas inspiración para viajar?"

"Descubre tu próxima aventura"

Title — Home, next trip

"¿Buscas inspiración para viajar?"

"Eleva tu viaje con IA ✈"

Description — Home, default

"Descubre ideas de destinos, actividades imperdibles y reseñas de lugares…"

"Inspírate con destinos, crea tu itinerario, encuentra ofertas de vuelos y alojamientos."

CTA — Home / Shop > Flights

"Inspírate con LATAM AI" · "Descubre con IA"

"Conversa con LATAM AI" · "Buscar vuelos con LATAM AI"

Accessibility · contrast pass

The button's label sat on a three-colour gradient that ran from magenta through a light mid-blue. Over those light tones, white text measured 2.8 : 1 — under the WCAG AA minimum. Re-anchoring it to a two-stop gradient — Purple 700 into Teal 500 — kept the brand energy and pushed contrast to 4.9 : 1.

Original Concierge button — tri-colour magenta-to-blue gradient with white label
Before — tri-colour gradientWhite label over the light mid-tones · 2.8 : 1, fails AA
Redesigned Concierge button — two-stop deep purple to teal gradient with white label
After — two-stop gradient · Purple 700 → Teal 500Same label · 4.9 : 1, passes AA

Shop > Flights · the biggest single lift

On the flight-search surface, the access button said only "Discover with AI." We made it say exactly what it does for someone already booking a flight — "Buscar vuelos con LATAM AI." Same button, same place, one honest line of copy.

Shop > Flights search screen — original access button reading Discover with AI
Before — "Discover with AI"Vague label on the flight-search header
Shop > Flights search screen — rewritten access button reading Buscar vuelos con LATAM AI
After — "Buscar vuelos con LATAM AI"Says exactly what it does, where it's needed

Suggestions · by user state

We branched the quick-reply content by state: no upcoming trip, an upcoming trip, and the 48-hour pre-trip window.

Concierge card with quick replies for a user with no upcoming trip — flights and destination inspiration
No upcoming tripQuick replies skew to inspiration & flights
Concierge card with quick replies for a user with an upcoming trip — organise the trip and things to do
Upcoming tripQuick replies skew to organising & activities
Concierge card with quick replies inside the 48-hour pre-trip window
48-hour pre-trip windowSame set, tested closer to departure
06 — Validation

Grounded in data, tested per state.

ValidationA/B · per user state

Every move ran as a controlled experiment, measured on access rate and first-message rate, split by user state. The suggestion sets themselves were grounded in data before a single one was written — seeded from the top user intents in chat, mined in BigQuery.

Top intent — Flights
43%
Avg. default + next-trip
Destinations & activities
29%
Inspiration intents
Hotels
15%
Stay-related intents

The 48-hour pre-trip window was the one inconclusive cell — it's being re-tested with a suggestions-only variant.

07 — Outcome

What shipped.

OutcomePost-experiment
Content — Home, default
+26%
Clicks on access button
Content — Home, next trip
+75%
Clicks on access button
Content — Shop > Flights
+102%
Clicks on access button
Accessibility — all surfaces
+11%
Home & Shop > Flights
Suggestions — no upcoming trip
+66%
First message sent
Suggestions — upcoming trip
+12%
First message sent
Shipped Concierge card on Home for a user with a next trip — rewritten copy and AA-safe button
Shipped — Home, next tripRewritten card and AA-safe button, promoted to default
08 — What I learned

The lift was in the invitation, not the UI.

Reflection

The biggest lift didn't come from new UI — it came from saying, in the right place, exactly what a button already did. Suggestions went further: they didn't just save a tap, they taught people what Concierge could do.

Content & accessibility, promoted to default. Suggestions set the foundation for the next big bet.

© 2026 — Luis Miguel Bello García Senior Product Designer, AI contact@luismi.design
Next — 03 / 06 Travel Planner for retention