Lyft challenged CR’s findings, citing an “observer effect,” meaning that by having dozens of people checking prices for the same route at the same time, CR may have artificially inflated demand for that ride and influenced the final prices our volunteers saw. Uber said that because its ride prices change “nearly every second,” it was “impossible” for us to ensure that trip requests happened at exactly the same time.
In short, Uber and Lyft argue that no two trips on their platforms—no matter how seemingly close in time and location they are to each other—can ever truly be the same.
“In an open, dynamic marketplace like ours, with nearly 1.7 million mobility and delivery trips per hour, a trip is defined just as much by when it is requested and what’s happening nearby as where it is going,” Uber said in a statement to CR.
But several experts we shared our findings with dispute that argument. On nearly every route we tested, they noted, we found that at least some of our volunteers converged on the same price for the same ride at almost the same time. And it would be difficult for CR’s tests alone to create artificial spikes in demand, given the relatively small number of volunteers we used and the mostly large and densely populated places we chose for our test rides, experts said.
“You’re saying that a few dozen people caused such a dramatic effect? Maybe if it was the heat of rush hour, from the airport to downtown, a truly hot surge area, but that doesn’t apply here,” says Christo Wilson, a computer science professor and associate dean at Northeastern University in Boston who previously audited Uber and Lyft’s pricing models for the city of San Francisco.
So what explains the different prices our volunteers saw, according to the companies? Uber and Lyft said that a wide variety of factors—rider demand, the supply of available drivers, location, time, estimated trip time and distance, weather, promotional offers, and traffic patterns, among them—all play a part in both original and final prices.
“Price differences reflect real marketplace dynamics,” Lyft’s Sid Patil, executive vice president of the company’s marketplace division, said in a statement. “At any given moment, more drivers may be available in a specific area, different demand levels, or different promotional activity. All in all, our marketplace ebbs and flows, depending on locations, times, events, weather, and other factors.”
Uber and Lyft said the only truly personalized pricing on their platforms is through their promotional offers, such as new-rider discounts and “re-engagement offers,” which they use to entice back customers who haven’t used the app in a while. Neither company provided a complete list of all the factors they use to personalize promotional offerings.
But elsewhere, both Uber and Lyft have detailed the types of data they collect and how it could be used, in their U.S. patent filings and company privacy policies.
Lyft said in a statement that it doesn’t group, or “segment,” its customers or use behavioral data to set base prices. But the company acknowledged that it uses a “broad set of signals” for its promotions and discounts. Lyft’s privacy policy goes into detail about some of the customer data it collects: how you interact with the Lyft app; your address book and calendar, if you consent; and the creation of inferences about who you are. Lyft provides two examples in its privacy policy: If you frequently ride to and from airports, you may be identified as a frequent traveler. Lyft says it may also infer your gender based on your first name.
Lyft’s patents go much further, outlining “sensitivity” scores and models, which can be used to predict the “importance” or “priority” of a given trip, arrival, or drop-off location; an “intent” model, which is capable of using your demographic info to predict a ride before it is even requested; and a willingness-to-pay score, defined as the “willingness by the mobile requestor device to pay a higher transportation service amount.”
Uber said in a statement that it, too, doesn’t use “protected characteristics,” such as race, gender, ethnicity, and disability status, for base prices or promotions; nor does it use “rider-specific behavioral characteristics.” But that did not address the use of behavioral data of larger customer groups. Uber did acknowledge that it uses personal data for promotions and discounts. Uber’s patents show it can use a phone’s sensor data and your past behavior for its models. That data can include how quickly and accurately you type an address; your gait and walking speed, which can be used to infer your height, weight, and body type; and even the precise angle at which you hold your phone, to spot any deviation from the norm. Your ride history is also a powerful predictor of both who you are and where you’re likely to go. Uber outlines one such example in one of its advertising patents: If someone routinely requests an Uber to a day care center or school before going to a workplace or university, they could be identified as a single working parent. From there, the age, gender, and sex of the rider, and the rough ages of the rider’s children, can be determined through the ride history alone, Uber says.
“Earlier generations of these pricing systems really focused on time, supply and demand, and price elasticities and efficiencies. But now, many companies actively use behavioral and context data to inform their models. They don’t even necessarily need your personal data,” says M. Keith Chen, a behavioral economist and professor at the University of California, Los Angeles, who previously worked as Uber’s head of economic research and helped create its surge pricing algorithm.
Both Uber and Lyft also denied offering their customers fictitious discounts. Lyft attributed our findings on these discounts to the fact that “prices change constantly based on real-time marketplace conditions.” Uber called our testing “fundamentally flawed” because, in its view, you can’t establish a true baseline price on its platform.
“If one user’s undiscounted price matches another user’s discounted price, that’s simply because these prices were different to start with, due to changes in real-time marketplace conditions,” Uber said in its statement.
Experts also dispute those arguments, saying the fact that many volunteers saw exactly the same final price for many of the routes we chose suggested that there was, in at least some cases, a true algorithmically determined starting price.
“I don’t agree that there is no baseline price, even with ride-share,” says Chen at UCLA. “What you’re seeing with your data is indeed a baseline.”
Uber also took issue with our fake discount analysis. We counted fares as discounted when what appeared to be an original price had a strikethrough and a lower price was displayed. Uber said in a statement that when these prices are accompanied by labels such as “Fares lower than usual,” they are not meant to suggest a discount but instead a “historical” or “informational” comparison.
Lastly, Uber and Lyft said the percentage of each fare they take is much lower than what CR calculated. They put their U.S. “take rates” at “around 20%” and “significantly lower than 30%,” respectively.
The disagreement largely comes down to different accounting practices: Uber and Lyft say our calculations don’t acknowledge the large and growing amounts they spend on auto insurance to cover drivers when they’re en route to and during rides.
But experts we spoke to say the companies’ insurance expenses are simply a cost of doing business that, under standard accounting practices, shouldn’t be excluded. (Sherman at Columbia says excluding them is “very misleading.”) And they note that both companies have created their own in-house insurance subsidiaries, with billions in reserves available for claims.
Uber also argued that because the driver and riders were in the same location, the experiment minimized pickup distance and “created an artificial scenario that isn’t representative of reality.” While our tests were designed to have riders and volunteers near one another, our analysis of how much Uber and Lyft take from each fare is similar to other studies. Experts say Uber and Lyft do, in fact, take nearly half of customer fares. An analysis of rides provided by three Uber drivers with over 50,000 rides between them, by Sherman of Columbia, found that Uber’s share has now risen to more than 50 percent in many cities. In a separate analysis conducted for CR using ride-hail trip data from Oregon, Princeton’s Workers Algorithm Observatory calculated that, on average, Uber took 44 percent and Lyft took 52 percent of the amount a rider paid for a trip.
Technical accounting issues aside, Uber and Lyft drivers we spoke to say their take-home pay is an ever-shrinking portion of what their riders actually pay—and far less than what they’ve been led to believe they would make. In 2024, for example, Lyft announced it would guarantee drivers 70 percent or more of rider payments each week, “after external fees.” (Later that year, the company settled with the Federal Trade Commission and paid a $2.1 million fine for what the agency described as “deceptive earnings claims” about how much drivers could expect to make, and this year announced it would cap its fee at 30 percent.)
Before Uber changed how it pays drivers, its drivers expected to keep 80 percent of their fares, according to lawsuits filed against the company.
Mohamed Drissi, a 43-year-old driver in Portland, Ore., who is originally from Morocco, says his take-home pay has gradually decreased over the six years he’s driven for Uber. “There’s the insurance fee, the city fees, the Uber fee, whatever that is. And after all that, it’s not $70 [out of $100]. It’s a lot, lot less,” he says.
Indeed, on Mohamed’s six test trips with us, his passengers paid about $126 in fares, not counting tips. Of that amount, $66.73 went to Mohamed, $58.41 went to Uber, and $16.66 went to city and airport fees. Not including government fees, then, about 53 percent went to Mohamed and 46 percent went to Uber. (Again, Uber says a lot of that 46 percent goes to insurance.)
