Uber has a problem with its story. The company has spent the past three years explaining rising passenger fares and falling driver pay through a single lens: commercial insurance costs, it says, have surged. In California, Uber tells regulators that insurance now accounts for 31% of an average rider fare. Globally, it says, once insurance is stripped out, its true take is around 21% - unchanged for years.
Two research teams, working independently with very different data, have found that story difficult to square with the numbers.
The first is a peer-reviewed Oxford University study published at the ACM Conference on Fairness, Accountability and Transparency in June 2025. Led by Associate Professor Reuben Binns and DPhil student Jake Stein, it analysed 1.5 million trips from 258 UK Uber drivers between 2016 and 2024, obtained through data subject access requests under GDPR. Since Uber introduced dynamic pricing in London in early 2023, the study found, Uber's median take rate climbed from 25% to 29% - with some trips yielding Uber a cut above 50%. Inflation-adjusted driver pay fell from over £22 per hour to just over £19 in the year following the change. Uber's surplus per driver working hour rose 38%.
The second is a report by Len Sherman, adjunct professor at Columbia Business School, published in June 2026. Drawing on decade-long trip histories from three veteran US drivers who collectively completed 50,000 rides, Sherman found that Uber's average take rate in many US cities has passed 50%. One driver tracked since the mid-2010s saw his take rate rise from 15-20% to above 50% over a period in which Uber's free cash flow grew by nearly $10 billion and its stock price increased as much as fivefold.
The insurance argument is not without foundation. Rideshare commercial coverage is genuinely expensive: Uber carries up to $1 million in liability per trip, coverage that individual drivers could not replicate at comparable cost. US mobility insurance costs, Uber says, rose more than 50% per trip over recent years. Some of that is real.
What is harder to explain is what Sherman found when he drilled into 100 near-identical trips by a single driver - same vehicle (a Tesla Model Y), same route (Ithaca to Syracuse Airport, roughly 60 miles), same driver safety record, same service type. The reported "estimated commercial auto insurance and operational expenses" on those trips varied from $13.75 to $50.00. A factor of more than 3.6 on trips where the risk profile was as close to identical as real-world data allows.
He then ran a regression. Day of week, time of day and service type were statistically insignificant. What did predict the variance: the rider price and the driver pay. When Uber charged the passenger more, reported insurance costs went up. When the driver accepted lower pay, reported insurance costs also went up. On trips where both applied together, the insurance line was higher still.
Sherman's regression implies that if Uber charges a rider $10 above average on a given trip, reported insurance and operating expenses tend to increase by $3.00 - without any change in the underlying risk. If it finds a driver willing to accept $10 below average, those fees increase by $4.30. If both apply simultaneously, they increase further. There is no actuarial logic under which insurance costs should behave this way.
The Oxford study found the same structural pattern in UK data. The higher the fare charged to the passenger, the higher Uber's take rate, and the less drivers earn per minute in absolute terms.
Consumer Watchdog, the US activist group, reported in May 2026 that Uber's self-insurance reserves grew from $6.7 billion in its 2023 annual report to $12.5 billion in its 2025 annual report - $5.8 billion in additional reserves over two years. Over the same period, Consumer Watchdog reported, Uber released more than $4.1 billion from those reserves as unrestricted cash.
Anyone who prices insurance risk will recognise the concern this creates. When a company both sets the per-trip fee it charges for insurance on individual transactions and decides how much of its revenue to designate as reserve, the two levers interact. Sherman's analysis suggests Uber's reported insurance fees are highest precisely on the trips where its conventionally defined take rate would otherwise be highest - which is exactly where you would expect the fees to land if their primary function were to reduce reported take rates rather than price risk.
Uber denies this interpretation. It says personal injury lawyers have inflated liability exposure in certain states, that mandated coverage requirements are genuinely expensive, and that its take rate after pass-throughs is stable globally.
What the data does not support is the claim that the take rate "remained stable for many years." That is what an Uber spokesman said when the Oxford study was published. The Oxford paper found median take rates rose from 25% to 29% after dynamic pricing, with 50%-plus cuts appearing on some trips. The gap between Uber's public position and what researchers found in Uber's own backend data - accessed through legally mandated data requests - is hard to reconcile.
Under Uber's original model, fares and driver pay were both functions of time and distance, with Uber taking a fixed percentage. The alignment was structural: if Uber charged a passenger more, the driver benefited proportionally. As Uber's CEO put it in a call to investors, describing the shift: "you've gone from just flat time and distance to now kind of point estimates for every single trip based on the driver... targeting of different trips to different drivers based on their preferences or based on behavioral patterns that they're showing us."
That decoupling matters for underwriters. Rideshare insurance has been building as a product category for a decade, with pricing models built on a reasonably transparent relationship between fares, driver hours on road, and actual exposure. When fares and driver pay vary independently - and when the fee labelled "insurance" on individual trip receipts is correlated with fare levels rather than risk levels - those models lose an important anchor.
The Oxford team found that 93 of 114 drivers tracked across the year before and after dynamic pricing were earning less per hour. Standby time - logged on and waiting, without pay - has risen significantly. In most months since 2023, Oxford found, drivers spend more time waiting for the next trip than they do on journeys with passengers.
Uber's public response is consistent. Its January 2026 blog post put the "true economic take - after netting out pass-throughs - at approximately 21% globally, and under 20% in the US once insurance is excluded." Its response to Consumer Reports in June 2026 cited California insurance at 31% of the average fare and described its US mobility insurance costs as having risen more than 50% per trip in recent years.
At a San Francisco tech conference in 2023, Uber's CEO told journalist Kara Swisher that, excluding insurance costs, the US take rate was "about 15%." A driver in the audience called this out from the floor. The CEO moved on.
What Uber's position does not address is the variance. Insurance costs rising nationally is one thing. Insurance fees that track rider willingness to pay and driver willingness to accept - trip by trip, on identical routes - is another. Sherman's conclusion is measured: "these regression results don't necessarily prove causality," he writes. But he adds that there is no other explanation for why Uber should charge such widely varying insurance fees on trips with an identical risk profile and negligible variable operating costs.
Australia's rideshare insurance framework took shape from 2016, when NSW introduced dynamic premium pricing for CTP, with other states following. The structure here differs from the US: drivers pay into the CTP scheme, Uber carries additional commercial coverage, and insurers have progressively expanded personal policies to include rideshare activity. Uber has not, to date, faced the same minimum pay legislation pressure in Australia that has modified its behaviour in Massachusetts and other US states.
The Oxford and Columbia research does not claim to document the same dynamics in the Australian market. What it does is establish, with trip-level data from Uber's own backend systems, a mechanism - insurance fees correlated with margin, not risk - that a platform with pricing power on both sides of its marketplace can use. Whether that mechanism is operating in Australia, and whether the CTP structure provides sufficient insulation from it, are questions the research does not settle. The financial incentives that produced it are present here too.
