Take-Home Design Task: AI-native Airbnb
Problems
Planning a trip is manual work. People research the area, the stays and the experiences themselves, then check by hand that availability, distance from the stay, and the order of activities line up. With a separate checkout for each item, the total cost is hard to keep track of. This creates major friction on decision making.
Users might still think of Airbnb mainly for stays, not experiences or services, so they often default to looking for stays on Airbnb but planning and booking experiences somewhere else.
People who plan with an AI tool get ideas in one place, but often read reviews and book somewhere else. Each handoff loses context, and it’s part of why 56% try AI to plan but only 8% find its answers enough.
What success looks like
A stay plus at least one experience or service, booked together before arrival. It only rises when the trip is planned on Airbnb, and add-ons offered after check-in don’t count, so it’s hard to inflate.
A clear signal for the one-stop shop Airbnb is building towards.
More of each trip’s spend is booked on Airbnb, without needing more travelers. I’d use it to support the north star not to replace it, since this one can be influenced by other things.
Solution
A trip planner inside today’s Airbnb search. People describe their trip in plain language, and AI guides them through a few quick choices: the area, then a stay, then experiences and services that fit around it. It does the matching of distance, timing and availability, and ends with the whole trip on one map with one total, ready to book together.
All
Homes
Experiences
Services
Guest favorite
Guest favoriteEntry point is the same place as today's search in home screen.
All
Homes
Experiences
Services
Where to?
User can describe what they're looking for in plain language, typed or spoken. A few words are enough to start.
Lisbon in May is a lovely time to go, with warm days and long evenings. What do you two enjoy? I’ll plan the weekend around it.
To work around the blank page effect, AI guides the user through a couple of questions in order to help them plan their trip end-to-end. The flow goes: interests -> recommend an area -> pick stay -> upsell activities -> reserve -> book.
Besides text, I'd give AI widget-like tools to call to ease readability and decision making for users.
Food and walking, you’ll love Lisbon for that. These three areas suit you best, and each has its own feel. Where would you like to stay?
Starting with neighbourhood recommendation to narrow down search. With minimal helper text to ease decision making.
Great choice. Bica has tascas on every corner and views over the river. These homes are free May 14–17, all a short walk from the food:
Recommended
Stay recommendations at a glance based on the answers so far
Great choice. Bica has tascas on every corner and views over the river. These homes are free May 14–17, all a short walk from the food:
Recommended
1 / 14
favorite
Any item in the widgets can be expanded to see more details about it before making a decision. Decision can either be via a UI (eg. Choose CTA), or conversationally.
Lovely pick, that terrace looks right over the river. Here’s what’s near the flat and matches what you two enjoy. I’ve added three to start:
Recommending relevant experiences and services that tie nicely to the stay, which takes the overhead of having to manually matching them by distance, times, availability, etc...
Here’s your weekend in Bica. Everything is a short walk from the terrace flat, and I kept Sunday loose so you can wander at your own pace.
A quick overview of the itinerary
Lisbon
The map shows every stop at a glance and in proximity.
Itinerary shown in the form of a timeline, something Airbnb already shows but after you've booked each item alone by itself. It works well with my approach but now you escape the tedious manual work of searching and matching.
One total for everything.
Thoughts and considerations
Blank page effect: In the vast majority of cases a simple text area for a prompt won’t be effective to capture user’s intent properly and really get answers in order to give them back value. Design here extends to the system prompt as it becomes an integral part of the experience, making AI proactive in asking for things like interests, suggesting activities, considering weather, and so on.
Idle aversion: In conversational interfaces, often AI takes time to reply and the user is left idle, we already build around that in Konvo, whether showing the agent’s to-do’s, loading states, or delightful visuals, there would need to be something to mitigate the idleness.
Motion: There are multiple key moments where delight could fit very nicely in the flow like while AI is first finding matches for the trip for example (since it might take a few seconds). I decided to leave motion out for this presentation due to time constraint, I’d rather use motion delightfully and with a bit of thought instead of just adding AI slop for the sake of adding it.
Voice chat: Would be curious to validate the voice chat route where I could have a casual voice conversation with Airbnb agent while it guides me through and help me decide. Since a lot of the experience requires visuals (check pictures of areas, stays, etc...) I’d mainly explore a route where it combines voice and visual interfaces (like the latest Chatgpt voice chat for example).
*Rough sketch*User can speak and express more naturally with voiceVoice controlsWhy more than the 3 screens you asked for? Once I got Claude to hit the Airbnb system properly, it was easy to prototype in high fidelity already, hence I showed more screens, yet still there are a lot of things I left out.