CHI Special Investigation · Uber · Transportation
The Number in the Middle
How Uber Turned One Fare Into Two
A rider in Colorado paid $14.38 for an Uber ride. The driver got $6.06. That does not mean Uber pocketed the remaining $8.32. Insurance and other charges sat inside that difference, and Uber was very keen to explain that when it used this exact trip in federal court.1 Fair enough.
Shown before booking. The rider was not shown what the driver would receive.
Shown as an offer before accepting. The driver was not shown what the rider paid.
It is not Uber’s profit on this trip, and this investigation does not treat it that way. Insurance and other charges sat inside that difference, and Uber was very keen to explain that when it used this exact trip in federal court. Fair enough. The finding here is about who could see which number — not about how much of the gap Uber kept.
But here is the part Uber would rather you not stare at for too long: the rider knew $14.38. The driver knew $6.06. Uber knew both. And the two people actually involved in the ride did not know the other number when they agreed to the transaction.
That is not a minor detail. It is the business model.
Uber no longer simply takes a published percentage from one fare. It can effectively answer two separate questions: What will this rider pay? And what will a driver accept to do the trip? The space between those answers becomes the economic spread Uber manages.2
Uber itself says the rider’s payment is “distinct” from what the driver earns. Its service fee “varies from trip to trip.”3 Its 2024 Driver Fare Addendum is even cleaner: “The Rider Fare will not match the Driver Fare amount.”4
The architecture
Rider fare in. Driver fare out. Uber in the middle.
Uber sets the machinery on both sides of the same trip. The relationship between the two numbers is variable, and neither participant can evaluate it at the moment they decide.
There is a Rider Fare. There is a Driver Fare. They are different numbers, and Uber controls the machinery producing both.
One upfront price for the trip. The rider decides whether the ride is worth $37.80.
Cannot seeWhat the driver is being offered.
Uber prices the rider, prices the driver, and manages the space between the two. The size of that space changes from trip to trip.
Taxes, regulatory charges, booking fees, commercial insurance and other direct costs also sit inside it.
A separate screen offering $16.25. The driver decides whether the job is worth $16.25.
Cannot seeWhat the rider is paying.
That is a very different proposition from the one most customers probably have in their heads, which is something like: I pay Uber $30, Uber takes its commission, my driver gets the rest.
How it got built
It wasn’t always like this
Early Uber was comparatively easy to understand. There was a rate card: base fare, miles, minutes and, when applicable, a visible surge multiplier. Uber took a disclosed percentage. The company could change the rules, experimented with commissions and was hardly a model of perfect contractual purity, but the basic economics could still be reconstructed with a calculator. The passenger fare and driver pay lived inside the same recognizable equation.
Then Uber improved the product. In 2016, upfront pricing arrived.5 This was genuinely useful. No more getting into a car and discovering the final price later. Uber showed you what the ride would cost before you booked. But the arithmetic disappeared with the surprise bill.
Rider fare and driver fare, separating in stagesOne formula in 2012. By 2022 both sides are priced independently. Colorado does not close the gap in 2025 — it makes both numbers visible.
A published rate card: base fare, per mile, per minute, plus a surge multiplier the rider had to type to confirm. Uber took a disclosed percentage.
One formulaUpfront rider pricing. One number before booking. The multiplier and the arithmetic leave the rider’s screen.
Rider price detachedRoute-based pricing. Uber confirms it estimates what groups of customers will pay for particular routes at particular times.
Price ≠ formulaDriver surge decoupled. A driver can receive surge on a trip where the rider paid no premium at all.
Surge detachedDriver upfront fares in most U.S. markets. The driver now gets an algorithmic offer before accepting, not a formula afterwards.
Both sides pricedColorado disclosure. For the first time, one U.S. state puts both numbers from the same trip in front of the people who funded it.
Both sides shownThe visible surge multiplier went away. The rate card stopped explaining much about the number on your screen. Uber still showed you the price. It just stopped showing you enough to understand how that price related to the underlying ride.
Then, in 2017, Uber moved further toward route-based pricing. Bloomberg reported that Uber was using machine learning to estimate how much groups of customers might pay for specific routes at specific times.6 This investigation comes from the Customer Hostility Index (CHI), a framework for identifying customer-hostile commercial practices. CHI is not claiming Uber calculates your personal maximum willingness to pay. The evidence does not support that. It does not need to.
The important change is simpler: the rider’s price no longer had to move mechanically with the driver’s compensation.
Then Uber cut another wire. In 2019 it changed driver surge so that a driver could receive surge even when the rider paid nothing extra.7 That can make perfect marketplace sense. Driver supply and rider demand are different problems. But it also means “surge” on one side of the ride no longer tells you much about surge on the other.
By August 2022, Uber said drivers in most U.S. markets were receiving upfront fares too.8 The driver gets an offer before accepting. Uber says that offer can reflect expected time and distance, demand and whether similar trips are more or less likely to be accepted.
Now we have the finished machine. The rider gets a price. The driver gets a price. Uber sits in the middle.
The number everyone argues about
So what does Uber actually take?
This is where nearly every viral Uber argument goes straight into a ditch. “Uber takes 20%.” “Uber takes 40%.” “Uber took 60% of my ride.” All three can be true depending on what the hell you’re measuring.
Suppose Uber charges you $25 and offers the driver $12. Did Uber take 52%? Not necessarily. Taxes, regulatory charges, booking fees, commercial insurance and other costs may sit inside the $13 difference. Uber says that after insurance and other direct transaction costs, its own blended take is around 20–21%.9 Independent analyses comparing full rider payments with driver compensation have sometimes produced figures around 40–50% before those costs are removed.
Both can be mathematically defensible. And that is exactly why the real story is not how much Uber takes on average. The story is that the number changes.
What a screenshot comparison produces. It is arithmetically true and economically incomplete.
Only the last of these is Uber’s. The others are real costs the gap has to cover.
Uber’s own blended figure, after insurance and other direct transaction costs are removed.
Comparisons of full rider payments with driver compensation, before those costs are removed.
The evidence for that is not anecdotal. Binns and colleagues, using roughly 1.5 million UK Uber trips, found a much wider distribution after dynamic pricing replaced the old fixed 25% model, including trips with very high platform shares and some where driver pay exceeded the rider fare.10 FairFare researchers using more than 76,000 Colorado trips found a broad spread of platform fees.11 UCLA researchers using New York City TLC data likewise found the platforms’ share of base fare rising materially between 2019 and 2022.12 The methodologies differ, but the conclusion is consistent: the old fixed-percentage mental model is gone.
A much wider distribution after dynamic pricing replaced the old fixed 25% model — including trips with very high platform shares and some where driver pay exceeded the rider fare.
A broad spread of platform fees across Colorado trips, drawn from the receipts the state’s disclosure law made visible.
The platforms’ share of base fare rising materially between 2019 and 2022 — above 30% on roughly three rides in ten by 2022.
Sometimes Uber’s spread is large. Sometimes it is small. Sometimes it can go negative. What’s remarkable is that the rider and driver cannot evaluate that relationship when they’re making the deal.
Imagine the transaction the way it actually works. Uber tells you the ride costs $37.80. You decide whether the ride is worth $37.80. Your driver gets a separate screen offering $16.25. They decide whether the job is worth $16.25. You do not know their number. They do not know your number when deciding whether to take the trip. Uber knows both.
Most riders do not think they are participating in a two-sided price-discovery system where Uber gets to price the buyer, price the seller and manage the space between them. They think they are buying a ride and Uber is taking a fee. Those are not the same thing.
The countercase
Uber has a legitimate defense
Uber runs a marketplace. Insurance is expensive. Some rides need driver incentives. Some riders get promotions. Some trips lose money. Higher-margin trips can subsidize lower-margin ones. And Uber creates enormous customer value.
CHI gives Uber a Customer Value Index (CVI) of 82 because the product is excellent at what it does. Uber made transportation easier, faster and more predictable. You can land in an unfamiliar city, open one app, see the driver, know when the car is arriving, know your price, track the trip, pay automatically and have a record of the entire interaction.
Anyone who remembers taxi dispatchers saying “twenty minutes” and then apparently entering witness protection understands what Uber fixed.
Uber’s frozen company scores, carried here as context. A high CVI and a customer-hostile pricing architecture are not a contradiction to be resolved — they are the reason CHI separates how useful a product is from how the commercial relationship around it is run. See the full score breakdown →
But Uber’s supermarket defense only goes so far. A supermarket buys milk as a principal, owns it and resells it. Uber’s own U.S. terms describe it as a limited payment collection agent for third-party providers.13 Its Driver Fare Addendum describes Uber as a limited pricing agent.4
Uber wants the legal benefits of being an intermediary while exercising the economic power to set both sides of the transaction. Interesting arrangement.
Disclosure
Then Colorado did something almost comically simple
- Colorado did not ban Uber’s model.
- It did not cap Uber’s fee.
- It did not tell Uber what to charge riders.
- It did not dictate what drivers must earn.
- Colorado essentially said: show them the numbers.
Since February 1, 2025, Colorado riders are shown what they paid and what the driver received after the ride and before tipping. Drivers receive corresponding post-trip information.14
Uber sued.
One of Uber’s arguments was that putting the rider’s payment next to the driver’s compensation could be misleading because the difference includes legitimate costs such as insurance and government fees. That point is factually fair. But the underlying argument is extraordinary: if customers see two accurate numbers from their own transaction, they might misunderstand them.
Here is a revolutionary solution: explain the numbers.
CHI calls this Intellectual Disrespect: withholding, obscuring or controlling useful information on the premise that customers are better off not seeing it because they might misunderstand it. Uber was not arguing that the two numbers were false. It was arguing that customers might draw the wrong conclusion from seeing them together.
Uber sought a preliminary injunction. A federal judge denied it on January 31, 2025. That was not a final ruling that Uber’s pricing system was illegal or that Uber’s constitutional argument was wrong. Uber later voluntarily dismissed the challenge on undisclosed terms.15
What the ruling did and did not decide
Uber sought a preliminary injunction and did not get one. That is not the same as losing the case. The denial was not a final merits determination that Uber’s pricing model was illegal, and it was not a ruling that Uber’s constitutional argument was wrong. Uber later voluntarily dismissed the challenge on undisclosed terms. This investigation does not claim that Uber sued and lost.
And after Colorado turned on the light, the world did not end. Uber kept operating. Drivers kept driving. Riders kept riding. Reporting found the disclosure screens straightforward.16
Consumer leverage
What happens when both sides compare notes?
CHI is not telling anyone to take a ride off-platform. The point is narrower: with both numbers visible, the two people funding the trip have enough information to decide for themselves whether Uber’s continued presence in it is worth what it costs.
Suppose Uber quotes you $32 and your driver says Uber is paying them $17.
Before the alternative — what leaving the platform costs
Staying on-platform gets you Uber’s insurance structure, trip tracking, payment handling, support and dispute mechanisms. Taking the ride off-platform may mean giving those protections up, may violate platform rules, and may create additional risk for both rider and driver. That risk belongs in the calculation.
Both knowingly accept the additional risk.
More than Uber’s offer.
Less than Uber’s price.
But now you can actually make the calculation. You could decide that $32 is worth paying for everything Uber provides. Or you and the driver could agree on $22: the driver makes $5 more, you save $10, and both of you knowingly accept the additional risk.
Congratulations. You have effectively converted Uber into a driver-retrieval service.
Uber found the driver. Uber established the route. Uber told you what the ride was worth to you. Uber told the driver what the ride was worth to them. Once the two of you compare those numbers, you can finally ask whether Uber’s continued presence in the transaction is worth the difference.
That is exactly why the opacity matters. Maybe Uber’s spread is worth every cent. But shouldn’t the two people funding it be allowed to know what it is before deciding?
CHI Lexicon · emerging pattern
Two-Sided Spread Opacity
CHI calls this structure Two-Sided Spread Opacity: a platform materially controls the price one side pays and what the other side receives for the same transaction, monetizes a variable spread between them and leaves the participants unable to evaluate that relationship when they agree to the deal.
The platform materially controls the price one side pays.
It also materially controls what the other side receives for the same transaction.
It monetizes a variable spread between the two.
Neither participant can evaluate that relationship when they agree to the deal.
Uber did not invent the underlying economic idea, and CHI did not discover Uber’s pricing split. Researchers and journalists including Len Sherman, Binns and colleagues, Veena Dubal, Calo and Rosenblat, and Bloomberg have documented important pieces of it.
What CHI does is put those pieces in front of the customer and ask the question that actually matters: now that you know how it works, what are you going to do differently?
Two-Sided Spread Opacity is recorded as Uber’s primary pattern on the company assessment. See it applied → · Browse the CHI Lexicon →
Because the next time Uber quotes you $41 for an airport ride, you will know one number. Your driver will know another. Uber will know both.
Maybe your driver is getting $30. Maybe $24. Maybe $18. Maybe the spread is entirely justified by insurance, fees and the cost of running the platform. Maybe it is not.
You have taken hundreds of Uber rides without knowing. And once you understand that, the natural reaction is not some abstract concern about algorithmic marketplace transparency.
It is much simpler:
How many fucking times did I get screwed?
Evidence
Sources
Numbered markers in the text link here. Uber’s own filings, contracts and marketplace pages are preferred over reporting about them; court records and statute link to their primary text.
- [1]Court recordUber Technologies, Inc. v. Moss, No. 1:25-cv-00096 (D. Colo.) — complaint filed 11 January 2025The trip that opens this investigation, and Uber’s own use of it. In arguing that the Colorado disclosure was misleading, Uber’s complaint set out a ride where the rider paid $14.38 and the driver received $6.06, with $2.24 of mandatory insurance plus the Colorado pre-arranged-ride fee, the booking fee and airport and city fees sitting inside the difference. That is why this page does not treat the $8.32 as profit.↑ back to text
- [2]SEC filingUber Technologies, Inc. — Annual Report on Form 10-K, FY2025 (filed 13 February 2026)Uber’s revenue-recognition language: “Depending on the market where the trip is completed, the service fee is either a fixed percentage of the end-user fare or the difference between the amount paid by an end-user and the amount earned by Drivers.” The residual definition appears in Uber’s 10-Ks from FY2020 onward — it is what makes the spread a documented structure rather than an inference.↑ back to text
- [3]Company documentationService fee — Uber Marketplace pricingUber’s own definition: the service fee “is the fee drivers pay Uber, and it varies from trip to trip. It’s the difference between what a rider pays and what a driver earns on a trip, excluding tips, tolls, and certain fees, taxes, and surcharges,” and “if the rider payment is the same or less than the driver earnings, Uber does not receive a service fee.” The companion marketplace page carries the “distinct” language quoted in the article: “How much a rider pays for a trip is distinct from how much a driver earns for that trip.”↑ back to text
- [4]ContractUber — Driver Fare Addendum (Rasier LLC), effective 30 September 2024The contract sentence quoted in the article: “The Rider Fare will not match the Driver Fare amount.” It defines the Service Fee as Rider Fare minus Driver Fare, gratuity, Direct Fees and taxes, tolls and surcharges; states the Driver Fare “may… be adjusted based on marketplace factors, such as supply and demand”; and makes Uber the driver’s limited pricing agent, recommending the Rider Fare the driver charges. Rate cards survive only as a fallback where no pre-trip Driver Fare is given.↑ back to text
- [5]Company documentationUber Newsroom — “Upfront Fares: No Math and No Surprises,” 23 June 2016The arrival of rider upfront pricing: uberX upfront fares “started rolling out in April in US cities and India,” and “instead of surge lightning bolts and pop-up screens, riders are given the actual fare before they request their ride.” Uber’s current rider-facing explainer is at uber.com/us/en/ride/how-it-works/upfront-pricing, where the price is described as reflecting estimated time and distance plus demand patterns for that route at that time.↑ back to text
- [6]ReportingBloomberg (Eric Newcomer) — “Uber Starts Charging What It Thinks You’re Willing to Pay,” 19 May 2017The 2017 confirmation of route-based pricing: Uber “applies machine-learning techniques to estimate how much groups of customers are willing to shell out for a ride,” a break from calculating fares “using a combination of mileage, time and multipliers.” The article follows Bloomberg in saying groups of customers; Uber denies individual personalisation. Bloomberg is paywalled — the reporting was syndicated by Fortune on 20 May 2017. Uber’s own route-based pricing page is at uber.com/us/en/marketplace/pricing/route-based-pricing.↑ back to text
- [7]Company documentationUber Under the Hood (Sabrina Ross) — “Shedding More Light on New Driver Surge,” 30 August 2019Uber’s own account of the redesigned driver surge as exact dollar amounts keyed to the driver’s location rather than the rider’s multiplier — including the sentence this passage rests on: “a driver may receive surge on a trip even if the rider doesn’t pay anything extra.”↑ back to text
- [8]Company documentationMarketplace pricing — Uber (“As of August 11, 2022 drivers in most US markets will receive upfront fares”)Uber’s own statement of the driver-side rollout, in the article’s own terms: most US markets, since August 2022. The Markup reported the earlier wave of 24 cities on 1 March 2022, with Uber’s statement that drivers “will make less money for longer trips but should earn more on shorter trips.” Uber describes the offer as reflecting estimated time and distance, demand, and whether a trip “is generally more or less likely to be fulfilled by drivers, based on aggregate patterns.”↑ back to text
- [9]Company documentationUber Under the Hood — “Understanding Uber’s Share of Driver Earnings,” 27 January 2026The source of Uber’s own take figure: subtract platform-participant direct transaction costs from revenue, divide by gross bookings, and the result is “21% (as of Q3’25).” Uber put the US figure at “around 20%… adjusted for insurance costs” in its 16 June 2026 comment on Consumer Reports. The components of the calculation are not disclosed.↑ back to text
- [10]Empirical studyBinns, Stein, Datta, Van Kleek & Shadbolt — ACM FAccT 2025: algorithmic pricing across approximately 1.5 million UK Uber trips (258 drivers)Peer-reviewed, and the strongest distributional evidence in the article. After dynamic pricing replaced the fixed 25% model, the median platform take rose to about 29%, many trips ran at 50–60%, some exceeded 100% — that is, the driver was paid more than the rider’s fare — and “the higher the fare charged to the customer, the higher Uber’s take rate.” Also at ACM DL doi:10.1145/3715275.3732099. UK data, not US.↑ back to text
- [11]Empirical studyFairFare (Calacci, Nagaraj Rao et al.; Princeton CITP / Penn State) — 76,625 Colorado trips from 45 drivers, 2017–2024The Colorado trip-level analysis: a mean platform fee of about 30% with a range from 10% to 80%, peaking near 44% in 2021–22 and falling to about 24% in 2024 on the study’s narrower service-fee definition. Its need to drop negative-take trips is independent corroboration that Uber’s spread can run below zero. Background from Princeton’s Center for Information Technology Policy, 19 June 2024 and 25 March 2025.↑ back to text
- [12]Empirical studyUCLA Labor Center — New York City TLC trip data, February 2023The platforms’ share of the base fare roughly doubled, from 9.0% in February 2019 to 20.7% in April 2022, with about 30% or more taken on roughly three rides in ten by 2022. Measured on the pre-tax base fare and pooled across Uber and Lyft — the cleanest US trip-level series available, because New York City publishes both sides.↑ back to text
- [13]ContractUber — U.S. Terms of UseThe legal characterisation the article places against Uber’s economic power over both sides: charges “will be owed directly to third-party providers, and Uber will collect payment of those charges from you, on the third-party provider’s behalf as their limited payment collection agent.” Uber also states that it is not a common carrier and does not transport passengers.↑ back to text
- [14]StatuteColorado SB24-075, codified at C.R.S. § 8-4-127 — signed 5 June 2024; disclosure provisions effective 1 February 2025The law behind this section. § 8-4-127(11)(d) requires the rider to be shown, on a single prominent screen before the tip prompt, the total the consumer paid and the total the driver received; (11)(b)–(c) require corresponding post-task information to the driver. It does not cap the spread, does not require Uber’s own share to be stated, and does not show the driver the rider’s price before accepting. Administered by the Colorado Department of Labor and Employment’s Division of Labor Standards and Statistics under INFO #23B. Not HB24-1129, which is the companion delivery-network bill at C.R.S. § 8-4-126.↑ back to text
- [15]Court recordUber Technologies, Inc. v. Moss, No. 1:25-cv-00096 (D. Colo., Judge Daniel D. Domenico) — preliminary injunction denied 31 January 2025; action voluntarily dismissed November 2025Uber filed a single First Amendment compelled-speech count on 11 January 2025. The preliminary injunction was denied on 31 January 2025 on reported grounds of standing, Uber’s year-long delay and unproven irreparable harm — there was no merits ruling on the First Amendment claim and no determination that Uber’s pricing architecture was unlawful. No appeal has been located. Uber voluntarily dismissed the action around 11 November 2025 after the parties resolved their dispute, on undisclosed terms (Bloomberg Law, 1 February and 12 November 2025).↑ back to text
- [16]ReportingWestword — “We Tested Uber’s Updates,” 14 February 2025The contemporaneous test of the disclosure in operation. Westword found the Colorado rider app showing the rider’s total and the driver’s earnings post-trip and before the tip prompt, and described the screens as “clean and easy to understand.” Uber relocated in-app tipping to after trip completion and withdrew four Uber Pro perks in Colorado, neither of which the statute requires; one driver reported that passengers “actually tip better.”↑ back to text
Every source above resolves to its primary document. Two are behind paywalls at the publisher: Bloomberg’s 2017 route-based pricing story, which was syndicated, and Bloomberg Law’s reporting of the Colorado docket. Prior work on Uber’s pricing split by Len Sherman, Binns and colleagues, Veena Dubal, Calo and Rosenblat, and Bloomberg is credited in the text; this investigation does not claim to have discovered the split.
Continue