Skip to main content
Make Data Work
Oct 15–17, 2014 • New York, NY

Generating Possible A/B Tests for Uber Via a City Simulation Framework

Bradley Voytek (UC San Diego and Uber, Inc.)
11:00am–11:40am Thursday, 10/16/2014
Connected World
Location: 1 E12/1 E13
Tags: geo_local
Average rating: ****.
(4.12, 8 ratings)

Uber has two main goals: 1) Get you a ride when you need it, and; 2) Make sure our driver partners are maximizing their earnings. Optimizing these two parameters requires modeling a number of complex, non-linear, interacting systems. Rather than actually confronting this difficult problem directly, Bradley made use of agent-based simulations of driver and passenger behaviors to see what combinations of parameters were best.

Here he introduced Uber’s city simulation framework and explain how and why we simulate Uber passenger/driver interactions. He will also discuss how this is used for “semi-automated science” to generate plausible A/B test options for Uber to explore.

Bradley’s simulations recommend optimal dispatch distances for pairing a driver with a passenger, a value that varies over time and differs across cities. Furthermore, the simulations suggest optimal behaviors for drivers to take between trips such that, when dispatch distances are very short drivers should navigate back toward demand density, however when dispatch distances are relatively longer drivers can maximize their earnings by using less gas by remaining stationary between trips.

Such plausible scenarios—which emerge purely from the simulations—provide Uber with a suite of testable A/B hypotheses. In other words, the city simulation framework generates possible A/B tests to optimize the Uber client experience and minimize gas usage to maximize driver partner earnings.

Photo of Bradley Voytek

Bradley Voytek

UC San Diego and Uber, Inc.

Brad is an professor of computational cognitive science and neuroscience at UC San Diego, and the Data Evangelist for Uber. He makes use of big data, mapping, and simulations to figure out cognition.

He’s created several research tools, most notably the neuroscience literature meta-analytic resource brainSCANr.com with his wife, Jessica Bolger Voytek.

He’s an avid science teacher and outreach advocate and he’s spoken at events ranging from elementary schools to venues such as Ignite, TEDxBerkeley, @GoogleTalks, and SciFoo. He runs the blog Oscillatory Thoughts (http://blog.ketyov.com) and his tongue-in-cheek book about the zombie brain, Do Zombies Dream of Undead Sheep? (Princeton University Press), comes out this fall.