University of Oxford Department of Computer Science Oxford Martin School Worker Info Exchange

“Not Even Nice Work If You Can Get It”

A Longitudinal Study of Uber's Algorithmic Pay and Pricing

Reuben Binns*, Jake Stein*, Siddhartha Datta, Max Van Kleek & Nigel Shadbolt* Shared first authors · University of Oxford · FAccT '25, Athens
258 drivers1.5M+ trips2016–2024

Your data, on its own, says little

A driver can ask Uber for their own data: trips, fares, hours online. But one person's records can't answer the questions that matter. Am I paid fairly? Is the algorithm treating me differently from everyone else?

The platform sees everyone

Personal data only becomes meaningful in context. Uber holds every driver's data at once, so it can place each person among their peers, spot patterns, and set pay accordingly. Drivers only ever see themselves.

This gap is an information asymmetry, and it is a core way platforms manage workers.

Rosenblat, A. & Stark, L. (2016). Algorithmic Labor and Information Asymmetries: A Case Study of Uber's Drivers. International Journal of Communication, 10.

Our goal: give drivers the controller's view

When drivers pool their data, each person can finally see where they stand among their peers: the same vantage point the data controller has always had.

We call this mutual legibility: seeing your own data in the same context the data controller sees it.

Calacci, D. & Stein, J. (2023). From Access to Understanding: Collective Data Governance for Workers. European Labour Law Journal. doi:10.1177/20319525231167981

A driver's day has three parts

When a driver logs in to the app, their time splits three ways:

On a trip with a passenger, En route to pick someone up, and Waiting for the app to offer a job.

§1 Introduction

Uber only counts some of it

Uber pays for time on trips and driving to pickups. The waiting time is unpaid, and invisible in Uber's version of "hourly pay".

§4.1 Pay

The courts count all of it

In 2021 the UK Supreme Court ruled that Uber drivers are workers, not independent contractors. The Employment Tribunal said working time includes every minute a driver is logged in and ready to work.

§4.1 Pay

Same work. Two very different wages.

Averaged over 2020 to 2024, drivers earned £29.46 an hour by Uber's definition.

Count waiting time, as the Tribunal does, and it falls to £15.98, before costs like insurance, repairs and fuel.

§4.1 Pay

How we know: drivers took back their data

Under GDPR, anyone can ask a company for a copy of the data it holds about them. Working with the non-profit Worker Info Exchange, drivers did exactly that and pooled it for this audit.

Everything was analysed on a secure laptop at WIE's office so no driver could be identified.

§3.2 Data Source

A data request returns a mess, not a dataset

Each driver's response arrived as dozens of loosely documented files: device logs, ratings, saved locations, messages. Only a handful say anything about pay.

§3.2 Data Source

So we rebuilt the picture ourselves

Four tables had to be cleaned, matched and stitched together before a single pay figure could be calculated. Problems surfaced at every level:

  • SCHEMA No unique trip IDs → time-based fuzzy matching
  • SCHEMA Tables renamed or missing between responses
  • TABLE Column names don't match → regex normalisation
  • TABLE Different aggregation (dispatches per hour vs. per record)
  • CELL Mixed formats in one column (Unix vs. timestamps) → regex normalisation
  • CELL Metrics don't line up with their definitions
  • CELL Personal details removed, encrypted ID added

§3.2 Data Source

Then the algorithm changed

Before 2023, the split was simple: of every fare, drivers got 75% and Uber kept 25%.

§4.3 Commission / Take Rates / Surplus

"Dynamic pricing" broke the fixed split

From early 2023, an algorithm set the passenger's price and the driver's pay separately, trip by trip. Uber's cut now varies every time.

The typical (median) cut rose to 29%. Only 46% of drivers still got 75% or more on average.

§4.3 Commission / Take Rates / Surplus

The bigger the fare, the bigger Uber's slice

Uber's cut was largest on the most valuable trips. The jobs drivers most hope for are exactly where Uber takes the most.

Bars are illustrative, showing the pattern found in the paper.

§4.3 Commission / Take Rates / Surplus

Uber's margin grew by 38%

For every hour a driver spent on a trip, what Uber kept after paying the driver rose from £8.47 to £11.70.

§4.3 Commission / Take Rates / Surplus

Meanwhile, more waiting

Time on trips stayed flat at around two hours a day. Unpaid waiting kept climbing, and since 2023 drivers usually spend more time waiting than driving passengers.

Approximate daily averages redrawn from Figure 2.

§4.2 Utilisation Rates

Real pay went down

Adjusted for inflation, comparing the year before dynamic pricing with the year after, Tribunal-definition pay fell from £22.20 to £19.06 an hour.

§4.1 Pay

Not everyone lost equally

We followed 114 drivers who worked a full year either side of the change. Each square is one of them.

§4.4 Inequality Between Drivers

Most were worse off

93 drivers earned less per hour. 21 earned more. The gap between drivers grew, and hours or experience didn't explain who won or lost.

§4.4 Inequality Between Drivers

Pay used to be predictable

Before 2023, a simple model could predict a trip's pay quite well from distance, duration and time of day. Drivers could plan.

Each dot: predicted vs actual pay (illustrative).

§4.5 Predictability of Pay

Then the old rules stopped working

Trained on 2022 trips, the model failed completely on 2023: it did far worse than just guessing the average. Whatever drivers had learned about their pay no longer held.

§4.5 Predictability of Pay

In short

After Uber introduced algorithmic "dynamic pricing" in 2023, the drivers in our study saw lower real pay, a bigger cut for Uber, more unpaid waiting, less predictable work, and more inequality between drivers. We can't prove the algorithm caused every change, since demand, traffic and the driver pool changed too, but the pattern is consistent.

It's also proof that workers can audit the algorithms that manage them, using data protection rights they already have.

What would help

Trip-level transparencyShow drivers the fare and Uber's cut for every trip, not just weekly averages.
Know before you acceptShare fare, distance and destination up front, as Colorado now requires.
Open trip dataPublish anonymised trip records, as New York and Chicago do.
Pay for waitingRecognise logged-in time as working time, and reduce idle hours.
Read the paper arXiv Worker Info Exchange