Solving train overcrowding through information and incentives
In an effort to reduce train car overcrowding in the Transbay Tube and throughout the system at large, BART deployed the first test phase of its ambitious FTA-funded Perks program in September 2016. Perks offered incentives to riders willing to shift their travel times outside of peak congestion periods toward less-crowded times, and while the seven-month pilot succeeded in shifting participant departure times, it was clear that the program’s operational costs could not sustain a full-scale launch.
To further reduce those costs and improve program performance, BART commissioned Metropia to employ its technology platform in support of the program’s second phase, Perks 2.
The Approach
Whereas phase one of Perks employed a ‘one size fits all’ approach, Metropia’s Artificial Intelligence (AI) backed platforms made it possible to target the optimal audiences with personalized incentive offers, a two-pronged approach which actually increased program participation while reducing operational costs for Perks 2.
The first step of the overall approach was to predict the daily time-dependent crowding pattern for each station-pair segment using Metropia’s machine-learning based crowding predictive algorithms based on available data sources such as historical Clipper Card gate-in/gate-out data, weather data and even special event data.
Next, Metropia’s behavior platform, INDUCE, used that crowding prediction data to determine new departure times for each individual rider within a 40-minute window of their routine departure time. After weighing the time shift’s imposition to the rider against its incremental benefit to the overall system, INDUCE then calculated the appropriate amount of reward points offered to each individual rider. Within two hours of the actual trip, INDUCE was able to read the Clipper Card data feed to determine if the user actually followed the recommended departure time and, if so, awarded points to the user’s account and the award records shown in the user’s trip log section. Additional opportunities to earn reward points were available for using BART for weekend activities and airport-bound trips as well as answering short in-app survey questions covering user demographics and travel behaviors. Riders were able to redeem their reward points for a variety of gift cards available in the BART app. The user experience, from travel time offers to reward point redemptions, can be seen in the Perks 2 section of BART’s official app, as shown in the images below.
Each month, train crowding information was updated and re-predicted along with the offers to participants. Several different adjustments to the incentive scheme were made during the program period to better understand behavioral responses.
Due to the experimental nature of this program, only 1900 users were recruited to participate in Perks 2, which ran from December 2018 through June 2019.
Project Accomplishments
Upon completion of the project, Perks 2 was deemed far more successful than the initial phase. The key highlights are summarized as below:
Later departure offers are more likely to be accepted by riders than earlier departure offers in both the AM and PM peaks
With limited PR and marketing effort and mostly relying on the user to discover the Perks program offer, nearly 20% of the users accepted and followed the offer monthly
Among eleven weekend offers evaluated, five resulted in statistically significant and positive increases ranging from 13% to more than 100% in the incentivized type of trip
More than 60% of the users reported that when they followed the suggested new departure time, they experienced better or equal crowding experience than before
The incentive cost per shifted trip varied over the course of the program but was approximately $1 overall, a significant improvement over the incentive cost of $10 per shifted trip in Perks Phase 1
Were the program scaled to include all BART riders, achieving a 5% reduction in peak-congestion through Metropia's behavior modification platform would have annual costs of $1.9M, or less than 1/3 of the $6M required to achieve the same results through a more traditional capital project.