Uber knows exactly how little it can pay its drivers. That is the point

Uber knows exactly how little it can pay its drivers. That is the point

Two research teams - one from Oxford, one from Columbia Business School - have independently documented the same finding: Uber's algorithm is not pricing trips. It is pricing workers

When Uber launched, the deal was simple. The company took a fixed cut - first 20%, then 25% - and drivers kept the rest. If Uber charged the passenger more, the driver earned more. Transparent, predictable, structurally aligned.

That model no longer exists. In the United States from 2022 and in the United Kingdom from early 2023, Uber replaced it with dynamic pricing. Fares and driver pay are now calculated separately, by algorithms that incorporate location, time of day, probability of cancellation and other factors the company does not disclose. The driver sees the offer. They do not see what the passenger paid. They cannot calculate what Uber kept.

Sarah O'Connor, writing in the Financial Times in August 2025, put the question plainly: what if you took dynamic pricing logic and applied it to wages? What if your pay varied job by job, based on factors an algorithm understood and you did not?

It is not a hypothetical. It is what has happened. There is now data.

What Oxford found

Oxford University researchers Reuben Binns and Jake Stein spent two years working with the Worker Info Exchange, a non-profit that helps gig workers access data held about them under data protection law. Through legally mandated data access requests, 258 UK Uber drivers provided their complete trip records - 1.5 million trips between 2016 and 2024. The resulting peer-reviewed paper, published at the ACM Conference on Fairness, Accountability and Transparency in June 2025, is the most detailed independent audit of Uber's pay system ever conducted.

In the year after dynamic pricing was introduced in London, average driver pay fell from £22.20 to £19.06 per hour in inflation-adjusted terms. Uber's median take rate rose from 25% to 29%. On some trips, Uber took more than half. The surplus Uber extracted from each hour a driver spent on trip - the revenue generated for the platform minus the driver's labour cost - rose 38%, from £8.47 to £11.70 per hour. Passengers paid more. Drivers got less.

There is a more structural finding underneath those numbers. The higher the passenger's fare, the less the driver earned per minute in absolute terms. Uber's take is not evenly distributed - it is concentrated on the most valuable trips, precisely the journeys where drivers might reasonably have expected to earn most.

Reuben Binns told the Financial Times: "To have that uncertainty hanging over you all the time... it turns work into a gamble, basically."

What a Columbia Business School professor found

Independently, Len Sherman - adjunct professor at Columbia Business School - published an analysis in June 2026 based on decade-long trip histories from three veteran US drivers who had collectively completed 50,000 rides in Florida and Texas.

Uber's average take rate in many US cities has now passed 50%. One driver, who started in the mid-2010s when Uber was advertising "keep 80% of every fare," watched his share erode through the 2019 pay cuts made ahead of Uber's IPO, through the introduction of algorithmic upfront pricing from 2022, to the point where passengers are now handing over more than twice what he receives. Over the same period, Uber's free cash flow grew by nearly $10 billion. Its stock price increased as much as fivefold.

Sherman's most specific finding concerns 100 near-identical trips by a single driver on the same route - a Tesla Model Y, Ithaca to Syracuse Airport, roughly 60 miles on Interstate 81, same driver, same service type. Uber's reported "estimated commercial auto insurance and operational expenses" on those trips varied from $13.75 to $50.00. A regression found that day of week, time of day and service type were statistically insignificant. What predicted the variance: the rider price and the driver's pay. Higher fares, higher reported costs. Lower driver pay, higher costs still.

There is no actuarial logic under which insurance costs should behave this way on trips with an identical risk profile. Sherman's own framing is careful: there is no other reason to expect such widely varying insurance fees on trips with an identical risk profile and insignificant operating costs.

How the system actually works

Uber's chief executive described the mechanism to investors in a 2023 earnings call. The company had gone, he said, 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 behavioural patterns that they're showing us."

The algorithm accumulates data on how each individual driver responds to offers. It learns which drivers accept low pay and which reject it. It learns how long a driver will wait before their threshold drops. It uses that data to calibrate future offers individually - probing the minimum any given driver will accept, trip by trip, without the driver knowing the question is being asked.

The driver sees only the number. They do not see the passenger fare. They cannot know whether the offer reflects market conditions or reflects what the system has learnt about their particular tolerance for a low-paying job.

When Uber's CEO described this to investors, he called it offering "the right trip at the right price to the right driver." The right price for Uber is the lowest price that particular driver will take. Veena Dubal, law professor at the University of California Irvine, has described this as the "gamblification" of work - algorithmic systems that transform pay from something predictable into a series of bets in which, as she has written, the house always wins.

For any employer now building AI into how work is allocated or paid, the architecture Uber has developed is not a curiosity. It is a precedent. The same capability - accumulating behavioural data on individual workers' acceptance thresholds and using it to calibrate future offers - is available in workforce management software well beyond the ride-hailing sector.

What Uber says

Uber disputes the research. When the Oxford paper was published, a spokesperson said the company "did not recognise the figures in this report" and pointed to weekly earnings summaries as evidence of transparency. The company argues that its true take, once insurance costs and pass-throughs are deducted, is approximately 21% globally.

Sherman's analysis addresses this directly. Uber's reported insurance costs on individual trip receipts vary by a factor of more than three on trips with identical risk profiles - and that variance is correlated with fare levels and driver pay rather than with actual risk. Consumer Watchdog reported in May 2026 that Uber's self-insurance reserves grew from $6.7 billion in 2023 to $12.5 billion in 2025, while approximately $4.1 billion was simultaneously released to unrestricted cash. Higher passenger prices, lower driver pay, growing reserves and growing profits: the pattern is consistent across the Oxford UK data, the Sherman US data and the Consumer Watchdog financial analysis. The three efforts reach a different conclusion from Uber's.

The drivers already knew

The Oxford team did not arrive at their findings alone. They spent two years working alongside drivers and trade union organisers, conducting workshops and collecting testimony. The drivers already suspected what the data confirmed. In sessions with the research team, one driver described asking a passenger what they had been charged: "The lady paid £64 but I was getting exactly 50% which was £32." Another: "They want to hide how much commission they are taking from you." A third: "That's when you discover they are robbing us and the customer."

The research gave those suspicions a number. The number is 38% - the increase in surplus Uber extracts from each driver working hour since dynamic pricing began.