The most complex single product I worked on at Lyft during my final year there. I owned this from ideation through ship in May 2023.
The idea for this product was to boost the overall efficiency of the airport sub-marketplace while improving the experience for both riders and drivers.
There are a few major problems with airport pickups for customers and for Lyft as a company.
- Each airport's operations and physical environment are unique. Signage, level, and even whether they require some tram or shuttle can all be challenges. Riders can't rely on their intuition to successfully meet their driver. So we have the first problem to solve: get riders to the right spot for pickup.
- Airports are strict about traffic flow and won't let a rideshare driver sit on the curb for an extended to wait for their riders. So when riders are late or the driver is early, the ride often ends up in a cancelation.
- Even when a pickup is successful, if there's any extended wait time for the driver waiting OR for the rider on the curb, that's inefficiency in the dispatch system with real business costs for Lyft in the form of wasted driver hours and rider dissatisfaction.
The answer to this problem is to totally reset the behavior and expectations of riders when booking their ride and shift towards pre-booking their pickup. By ordering as soon as they hit the runway, Lyft's marketplace systems have extra time to find the perfect driver for the pickup and riders can get better guidance for finding their pickup spot in the app.
*Mockup, for illustration. Real feature much more polished and rich 🙂
My work involved with bringing this product to users touched a TON of cross functional teams:
Rider app - obviously the bulk of the work. It was a big challenge to overcome users learned behaviors and communicate all the new wayfinding instructions and explain the new matching system through design and content.
Mapping - to do a better job with matching, Lyft's systems need to know where riders are in the airport and how long it will take for them to get to the pickup area. I had to work with the mapping organization to both create walking routes in the basemap and to then train ML models to predict the walk times for riders.
Dispatch and matching - these systems needed to take into account the new information about riders walking travel timing to more efficiently choose a driver. This required significant re-writes of the logic and scoring systems.