How Uber steers its drivers toward better performance

New research shows that the app’s ratings and incentive system has made drivers in Chicago as safe and reliable as taxi drivers. The findings suggest regulators may want to consider new quality-control measures.

Under normal circumstances, most of us wouldn’t hop into a stranger’s car, sleep in their guest room, or invite them into our home. But thanks to platforms like Uber, Airbnb, and TaskRabbit, these once-unusual acts are now routine.

Unlike traditional service industries, companies that hire gig workers often bypass extensive vetting or formal licensing requirements to speed up onboarding. So how can they ensure quality when almost anyone can sign up to work?

Uber, for one, tries to keep drivers accountable by sending notifications when riders give them low ratings, threatening to remove them from the platform if they don’t improve, and providing reports that compare their performance with their peers.

A recent study of Uber drivers in Chicago finds that these quality control mechanisms are working—motivating poor drivers to get better, weeding out those who don’t, and leading to a level of service that’s comparable to local taxis. The research is available as a working paper in the SSRN Electronic Journal.

“Services marketplaces are an important part of the economy, but they won’t fulfill their potential unless they can ensure quality while allowing the service workers to enter and exit flexibly,” says Susan Athey, Ph.D., a professor of economics at Stanford Graduate School of Business. “Workers are very responsive to information and feedback about their performance.”

When hiring for UberX, its main ride-hailing service, the company does a basic criminal history and driving record check and ensures drivers have a license, registration, and insurance. Unlike Chicago’s taxi drivers, Uber drivers don’t need to take a two-week course or pass a licensing exam.

Once new hires are on the road, Uber customers can rate each trip on a scale of one to five stars. Drivers with low ratings are notified that they need to improve and receive links to resources that can help. If they fail to raise their scores sufficiently, they risk being deactivated.

Athey and her colleagues Juan Camilo Castillo, Ph.D., and Bharat Chandar, Ph.D., examined around 6.9 million UberX rides near downtown Chicago over several months in early 2017.

Controlling for factors like destination and time of the week, they found that riders tended to give higher ratings to safer trips—those in which the driver maintained a steady, moderate speed, didn’t often brake or accelerate suddenly, and didn’t handle a cell phone. Customers also preferred shorter trips and being picked up and dropped off close to their desired locations.

When drivers got notifications about low ratings, they improved substantially on these measures of quality, Athey and her team found. By analyzing telemetry data on drivers’ mobile phones, they discovered that those who received warnings used their phones less, maintained a steadier speed, sped up less, took more efficient routes, and picked up and dropped off riders closer to their requested locations.

Drivers improved after just one notification, the researchers found, and they continued to be better drivers even after Uber alerted them that they were no longer at risk of being booted off the platform.

“Workers may not know or be paying attention to these dimensions of quality. A driver may not realize that they have more hard brakes than others,” Athey says. “Learning that they are below average has a powerful effect.”

How am I driving?

Another sign that the system of ratings and notifications was working: Telemetry data revealed that the drivers Uber removed due to consistently low ratings were significantly worse than average, based on the measures of quality the researchers observed.

Athey and her team also identified another way to get drivers to improve: providing detailed, objective information about their behavior on the road. Uber was already sending drivers weekly reports summarizing their performance based on telemetry data and comparing them with other drivers.

The researchers conducted a randomized experiment in which they provided some drivers with a more detailed dashboard tracking their behavior on individual trips. Drivers with access to this granular data improved more than a control group, and those who were in the bottom 10th percentile based on objective measures of their driving quality improved the most.

“A reminder of your past behavior seems to be a good thing in terms of future behavior,” says Castillo, an assistant professor of economics at the University of Pennsylvania.

Yet can feedback alone ensure quality when drivers don’t undergo extensive screening? To find out, Athey and her colleagues compared the performance of UberX drivers in Chicago to that of traditional cab drivers hailed through the Uber Taxi app. Taxi drivers go through a rigorous licensing process, and their customer ratings don’t lead to notifications about performance or the risk of deactivation.

Based on telemetry data, the researchers found that UberX drivers did better with speeding, sudden braking and acceleration, and convenient pickups and drop-offs.

Taxi drivers, on the other hand, used their phones less and took quicker routes. Overall, the quality of driving was roughly the same based on the measures riders prioritize, the researchers found.

“The conventional wisdom was that there might be a tradeoff between quality, which might come from experience and training in the case of taxis, and cost, which is generally lower with Uber,” Athey says. “We don’t find evidence that taxi quality is higher.”

In this case, other factors—such as demographics and financial incentives—could account for some of the variation in drivers’ behavior. Still, Athey says the findings suggest that regulators could consider replacing burdensome screening and licensing requirements with after-the-fact quality controls in some industries, especially when artificial intelligence tools can track compliance.

“We might be moving to a world where occupational licensing and other barriers, which may prevent workers from entering professions and keep prices artificially high, can be removed while maintaining service quality,” she says. “This may have fewer safety downsides than ever before.”