Parking Citation Response: Lyft Bikeshare

By far one of the most challenging and nuanced projects I undertook while working on bikeshare was solving the problem of improperly parked bikes and the associated fines. We took a number of steps to reduce these citations via partnerships with the city, operations, and policy adjustments.

Ultimately, my work saved $500k in fines, while leading to an improved operating environment that will save millions of dollars through the program’s lifetime.

Additionally, my work resulted in 1000+ bike racks being installed throughout San Francisco, and were located according to my recommendations.

This was a project that occurred over four distinct steps, while also working with the city to improve the program policy requirements.

1. Prior to launching ebikes, I created a simple suitability analysis, showing where bike parking was, where our service area would expand over time, and where we would need to drop bikes to maintain our serviceability requirements. We sent the intersection-level results to the city where they later installed bike racks.

2. After launching ebikes, I enhanced the analysis to look at which intersections our bikes were collecting based on ridership patterns and compared that to existing bike infrastructure (a few images from the slide deck describing this analysis are to the left). Again, we sent the intersection-level results to the city where they later installed bike racks. This was a parametric analysis, so we were able to send additional requests at a moment’s notice, using new and/or historical data. I later added intersections with high citations as priority locations as well.

3. After one month of operating, we began getting fines from the city. We had no intake system, so I managed the product for intake and response, which also allowed us to collect key data. A citation would be received via email, then parsed for data collection, then sent to a routing program to allocate jobs to the ground team. The software collected additional data on operation performance to keep the team accountable and provide a basis for contesting citations.

4. Once the data about where and when infractions were occurring was flowing, I used machine learning to dynamically create a list of bikes for our operations team to “sweep” prior to the city’s investigators getting to the bikes and citing them. Instead of sweep the thousands of bikes on the ground, they just needed to focus on the ones most likely to receive a fine.

Throughout those steps, we worked with the city to show the regressive nature of the citations. Ultimately, we succeeded in changing the policy so that some bikes received a fine, whereas the others received a warning. Using the two processes in place, our operations team was set up to properly respond; and our fines dropped by about 65%.