Case study · Conversation design
A new destination.
Every single day.
How a daily travel skill became a content system that could handle whatever the world threw at it.
150k
Interactions
Across the US, UK, Australia and Hong Kong
100+
Destinations
Cities, islands, countries, regions
$1M
Marketing budgets
Holiday Inn, Shangri-La, KLM, Visa, Choice Hotels and more
What it refused to be
A menu sorted by type of traveller
Somewhere new to explore every single day
Sorting by who you travelled with told users nothing about a place. The draw was simpler: a new destination to wake up to every morning.
The skill
Daily travel inspiration, delivered by voice.
Tripadvisor has one of the world's largest travel content libraries. Destination of the Day brought that content to Alexa. A new place every morning, a fun fact, region-specific music, three ways to explore further, and a Tripadvisor link sent to the user's phone if they wanted one. It ran daily in the US and UK. Other markets switched on for sponsored takeovers.
I led content and conversation design for the full run of the skill, across three years and four markets. I wrote over a hundred destinations, built the content system, designed the conversation architecture, and managed the sponsor integrations.
A bit of show, not tell
The honest version
Me vs the platform, the team, and myself.
Me vs the platform
Anangu and Jawoyn. Two indigenous place names the client needed right, and Alexa couldn't say either. Not a thing I could write my way around. The engineer saw a text-to-speech problem with no clean fix. Three rounds of SSML, YouTube references and audio demos later, we got it on the third pass.
Me vs the team's default
The team's answer was always Alexa, for speed and familiarity. For daily destinations I agreed. For sponsored takeovers I didn't. The points of interest were fixed, so a mangled name couldn't be swapped out, and a paying partner deserved better than a robotic read. I argued sponsors who gave us lead time should get a real voice. I won that.
Me vs my own call
Shangri-La Hong Kong was the test of it. I'd pushed hardest for a real voice here, the team and client backed me, the actor was booked. The clips came back with an accent so strong it risked reading as caricature from a brand. I couldn't promise a Hong Kong audience wouldn't take it that way, so days from launch I pulled my own choice. The fix was simple: Chinese instrumental, up and under throughout. It carried the takeover and gave it the very authenticity we’d been chasing in the voiceover artist, without the risk.
The format
The format took three tries.
The skill went through three iterations before the categories held. Each version shipped to real users, and researchers brought back the dropoff and return data we needed to redesign.
What counted as “held” got decided before V1 shipped. We measured dropoff after the reveal, return rate the next day, and link requests at the end. My intuition said V1 was sound. The data said users left after the reveal and didn't come back. The data won, twice.
V1 · Solo, Couple, Family
Didn't hold
Organised by traveller type. Looked logical on paper but didn't survive in the wild. The categories overlapped too much. A solo traveller eats, a couple eats, a family eats. Filing a place under who you were with told users nothing useful about what was actually there. Live data showed users dropping off after the reveal and few coming back the next day.
V2 · Hotels, Restaurants, Attractions
Closer, but too rigid
Organised by point of interest. Not every eatery is a restaurant. A waterfall is its own thing. The Eiffel Tower doesn't sit comfortably under attraction or anywhere else. Too much of any given destination ended up outside its own categories.
V3 · Hidden Gems, Must-Sees, Local Cuisine
This one held
Softer categories that bent to fit what destinations actually contained. A waterfall could be a hidden gem. A street food stall fit local cuisine even though it wasn't a restaurant. The Eiffel Tower was a must-see and didn't need further explanation.
Dropoff fell, link requests went up, and returners climbed. Within a year, Tripadvisor was negotiating marketing opportunities with Holiday Inn, Shangri-La, KLM, Visa, Choice Hotels and others.
The conversation architecture
Simple to use. A lot of thinking underneath.
One destination revealed, three ways into it, and a loop with no dead ends, you went as deep as you liked or left cleanly. Each category opened with its own soundscape, so you knew where you were before Alexa spoke.
The reveal
A new place, a fun fact, and destination-specific music. Amapiano for Cape Town, reggae for Jamaica. The place landed before a word was said. On sponsored days, the partner sat inside the reveal as part of the local picture.
Three ways in
Hidden Gems, Must-Sees, Local Cuisine. Each with its own soundscape. Once a category was heard it wasn't offered again, the loop only surfaced what was still unexplored.
Footsteps fading · bustling ambience · restaurant noise
The loop
Go deeper into another category, or exit. Every path led somewhere useful. On Echo Show, each category had matching visuals.
The handoff
The link to mobile. We first asked for the phone number right after the fun fact, and most users said no, fair, they hadn't chosen anything yet. Moved to the end, after all three categories, the yes carried real intent.
Where the writing meets the thinking
The conversation design in practice.
The scripts were where the conversation design lived. Every line was a decision. Four of them are below.
Hidden Gems as trust builder
“Let's uncover some hidden gems in Nashville. Start with Fort Nashborough, a historic replica of Nashville's original settlement. Then head to The John Seigenthaler Pedestrian Bridge for stunning views of the skyline and Cumberland River.”
Hidden Gems built the most trust. It signalled this wasn't a tourism board script. Real knowledge, the kind a local would share. That credibility carried into everything that followed, including the sponsor.
Must-Sees as headline writing
“Edinburgh's must-sees start with the Castle, perched atop Castle Rock with panoramic views over the city. From there, the Royal Mile connects the Castle to the Palace of Holyroodhouse, the official residence of the monarch in Scotland, lined with shops, cafes and historic landmarks. The Botanic Garden offers a slower stop, with plant collections from around the world and pathways through them.”
Must-Sees was the headline version of a place. The script covered what defined it. The crowd ambience told users they were where everyone goes.
Local Cuisine as sensory writing
“Cancun's local food scene is a culinary adventure that delights the senses with vibrant flavours and fresh ingredients. One of the standout dishes is ceviche, prepared with marinated raw fish or shrimp, lime juice, onions, and a medley of spices.”
Local Cuisine had to make people taste the destination. The restaurant soundscape set the scene. The copy did the rest. The goal was always the same. Make the user feel they were already there.
The link as a designed handoff
“If you'd like a Tripadvisor link sent to your phone so you can explore more of Nashville, please say your phone number.”
The link offer was where the experience moved from voice to mobile. Positioned after all categories had played out, when the user's interest was at its highest. Timing as conversation design.
The sponsorship integration
Partners that felt like local knowledge.
As the skill's audience grew, sponsors followed. KLM, Visa, Holiday Inn, Shangri-La, Choice Hotels, Tourism Northern Territory and others put budgets of up to $1M behind scripts they hadn't yet seen, on a daily skill they didn't directly control.
The brief was always the same. The partner had to feel like part of the experience. The hotel became the insider's recommended base. The restaurant became the local dining tip. The copy had to work hard enough that nobody could tell where the editorial ended and the partnership began.
The two interests pointed in opposite directions. The sponsor was paying to be seen. The user had opened a skill that worked because it never sounded like it was selling anything. The script only earned its place when it served both at once. Lean toward the brand and you lost the user. Lean toward the user and you shortchanged the people funding the skill.
The brands ranged widely. Shangri-La took us to high-end Hong Kong on one takeover. Choice Hotels took us to budget-friendly Dallas and Nashville on another. Different feel each time, same skill underneath.
Sponsored destination script, Los Angeles
“Did you know that LA is not only the entertainment capital of the world but also a city of diverse neighborhoods and rich culture? That's why the new Holiday Inn and Suites Monterey Park is the perfect home base, located just minutes from top spots like the LA Arboretum, Dodger Stadium, and Citadel Outlet Mall.”
The sponsor sat inside the destination story. The hotel was positioned as the logical base for the experience Alexa had just described.
In partnership with Holiday Inn
Production logic
Shipping daily set the rules.
Shipping every morning meant strict rules. Library music. Regional Alexa as the default voice. The trade was a synthetic voice and pre-cleared music in exchange for being able to edit and reload within minutes.
Most of what mattered lived in the detail. SSML corrections on sponsor names, place names and points of interest, usually caught before a client flagged them. The conflicts that detail threw up are above. The craft was in the fixing, and in one question under all of it. Could we back what we were about to ship?
Built to respond
The system could handle the real world.
We wrote and built a week ahead in batches. The real world rarely batches that neatly.
Same-day edition
Cornwall, overnight
The Northern Lights appeared overnight over Cornwall and made the morning news. We swapped that day's destination within hours. “Did you know that last night, for the first time in nearly three decades, the skies over Cornwall lit up with the Northern Lights? Reports are coming in that they may appear again tonight.” Live before midday.
Pulled
Spain
When devastating floods hit Spain, we pulled the Spanish destination before it went out.
Same-day edition
Glendale, Arizona
Rihanna headlined the Super Bowl in Glendale, Arizona. Glendale became the destination of the day. The fun fact wrote itself.
The team
How we worked.
I led conversation design across the run. The work shipped because of a team.
Product Manager
Kept product and business goals aligned with what users needed. Essential on sponsored takeovers where brand, product and content had to move together.
Conversation Engineers
Built every destination inside the Alexa console. The rhythm was simple. I'd write, they'd build, I'd listen. SSML corrections happened inside this loop, including the three rounds for Anangu and Jawoyn.
Sound Engineer
Chose every soundscape and music cue. The footsteps fading on Hidden Gems, the bustle on Must-Sees, the restaurant ambience on Local Cuisine. The Chinese instrumental that held up the Shangri-La Hong Kong takeover.
Researchers
Brought the live user data that drove the format from V1 through to V3. Their work also kept the skill consistent across very different sponsors, from Shangri-La in Hong Kong to Choice Hotels in Dallas.
Visual Designers
Designed the visual layer for Fire TV and Echo Show. On-screen elements matched the audio of each destination.
Tripadvisor Commercial Team
Managed the relationship between the skill's editorial voice and the needs of brand partners on sponsored takeovers.
Watch it in action
The skill, running.
What this shows
Three years of this taught me a few things.
Most of the design happened between people
Not in the script. Engineers pushed for what shipped fast, researchers brought data that contradicted my copy, sponsors paid for prominence on a skill that worked because it never felt paid for. None of them were wrong. The job was finding the line each could live with.
Cultural sensitivity isn't a final checkbox
Sometimes it was three rounds of SSML for two indigenous names. Sometimes it was pulling a recorded voice when the fit didn't hold. Same question under both: could we back what we were about to ship?
A daily product makes you design systems
After a few months I stopped writing destinations and started writing a way of writing them.
Scale and quality are not opposites
You just have to build the thing that holds both.