Korva is a scroll of the internet’s best ideas.
One voice note books the whole trip
Booking the trip
It starts with where the user is. If location is shared, the assistant sees that they're in San Francisco, then finds the flight options and applies the details a traveler normally enters by hand: preferred airline, seat type, class, meal, and which credit card to charge. Then it books.
Hotel and calendar
The hotel comes next, and it may be one the user has already stayed in the last seven times. Everything lands on the calendar, including the flight, the Uber to the airport, the Uber from the airport to the hotel, and the events the user might want to attend.
Preferences carry forward
Preferences carry forward. Anything the user has said about style or where they like to stay is stored, and the assistant extends that general taste to other bookings, so each future trip needs fewer decisions.
Why is travel so hard for software to book?
Airline inventory has long flowed through global distribution systems, which began as airline reservation networks. The Sabre system started in the late 1950s as a partnership between American Airlines and IBM. Fares, seat maps, and ancillary options come in different formats from different carriers. That patchwork is why booking travel by software has needed so much custom integration.
How does an assistant learn travel preferences?
Systems that generalize from past choices usually pull out attributes, such as neighborhood, price range, or brand, and score new options against them. This approach is called content-based filtering. It differs from collaborative filtering, which recommends items that similar users chose. Combining both is a standard design in recommendation systems.
More from Invest Like the Best with Patrick O'Shaughnessy
