The algorithm knows exactly how little it can pay staff. That is the point

Sarah O'Connor put it well in the Financial Times last August. Dynamic pricing for consumer goods - concert tickets, hotel rooms, surge fares - is easy enough to accept as market logic. Supply and demand, real-time data, faster adjustment. Fine. But what happens when you apply that same logic to wages? What if your pay rate varied trip by trip, order by order, based on factors an algorithm understood and you did not?

This is not a thought experiment. It is what a growing body of peer-reviewed research now documents in detail - and the findings are instructive for anyone who manages people, not just those who manage gig platforms.

The mechanics are worth understanding. When Uber operated a fixed commission model, the relationship between passenger price and driver pay was transparent and structurally aligned: if Uber charged more, the driver earned more. Dynamic pricing broke that alignment. Fares and driver pay now vary independently, set by algorithms that incorporate location, time, probability of cancellation and other factors the platform does not disclose. The driver sees the offer. They do not see what the passenger paid. They cannot calculate what percentage Uber kept.

Oxford University researchers Reuben Binns and Jake Stein spent two years analysing data subject access requests from 258 UK Uber drivers, covering more than 1.5 million trips between 2016 and 2024. Their finding, published at the ACM Conference on Fairness, Accountability and Transparency in June 2025, is precise: after dynamic pricing was introduced in London in early 2023, average pay per hour fell from £22.20 to £19.06 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 per driver working hour - the revenue generated for the platform while a driver was on trip, minus the driver's labour cost - rose 38%.

What the algorithm is actually doing

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

That is not metaphor. Len Sherman, adjunct professor at Columbia Business School, analysed decade-long trip histories from three veteran US drivers with 50,000 combined rides and found that Uber's average take rate in many US cities has now passed 50%. More specifically, he analysed 100 near-identical trips by a single driver - same vehicle, same route, same driver, same distance - and found that Uber's reported insurance and operating expenses on those trips varied from $13.75 to $50.00. His regression found that the rider price and the driver's pay predicted the variance; the risk profile of the trip did not. The higher the passenger fare, the higher the reported costs. The lower the driver's pay, the higher the reported costs still.

The Oxford team found the same structural pattern in their UK data. The higher the passenger's fare, the less the driver earned per minute in absolute terms.

What these two teams have independently documented is a system that probes, in real time, the minimum price any active driver will accept on any given trip. It adjusts continuously. It accumulates data on individual driver behaviour - acceptance rates, patterns, threshold responses - and uses that data to calibrate future offers. As Uber's own CEO described it to investors: "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."

Why HR leaders should care about this

The immediate response is to say this is a gig economy story, not an HR story. It involves contractors, not employees. It is a regulatory question for platform workers, not a workforce management question for people directors.

That response is harder to sustain than it was two years ago.

Nearly one in five Canadians - approximately six million adults - reported working in the gig economy in 2025, according to an H&R Block Canada survey. The ILO estimates the gig economy now covers up to 435 million workers, 12.5% of the global labour force. As HCAMag reported when the ILO opened binding negotiations in June 2026, Human Rights Watch's report Algorithms of Exploitation, examining working conditions across nine countries, found "a serious problem with transparency and accountability around how algorithms are used to determine pay and performance."

The algorithmic pay model is not confined to Uber and Deliveroo. It is spreading into warehouse picking, last-mile delivery, care work scheduling and freelance professional services - industries where many workers are not contractors. They are employees. Which means they fall under your remit.

The information asymmetry is the point

The UK Low Pay Commission's investigation, cited in the FT piece, found "frequent complaints about the opacity of pay, with platform workers across several sectors agreeing this had worsened over time." One worker put it directly: "They've upped the prices, the delivery fee, but we... actually get less money."

This is algorithmically engineered information asymmetry. The platform knows, in real time, what a particular worker's acceptance threshold is, what comparable workers nearby are accepting, how long a given driver has been waiting, and what happens to acceptance rates when pay drops below a certain level. The worker knows only the number in front of them and a deadline to accept or reject.

For HR professionals, this framing has a specific resonance. The questions you would ask about any compensation system - is it transparent? Is it predictable? Does it reflect performance and effort in ways employees can understand and plan around? - are exactly the questions these systems are designed to make unanswerable. The Oxford team found that models trained on pre-2023 Uber trip data could not predict post-2023 driver earnings at all. The relationship between trip characteristics and pay had been structurally severed.

Veena Dubal, law professor at the University of California Irvine, has called this the "gamblification" of work: algorithmic decision-making that undermines economic stability by transforming the basic terms of how workers are paid. In this model, as Dubal has written, the house always wins.

Canada's regulatory response

Ontario's Digital Platform Workers' Rights Act came into force on 1 July 2025, introducing minimum wage protections, pay transparency requirements and the obligation to disclose pay calculation methods within 24 hours of sign-up. Fines reach $500,000 for serious violations.

But Jennifer Scott, president of Gig Workers United, told CBC it falls short: "This minimum wage isn't a real minimum wage because it only applies to just a little over half the time that we're at work." The Act covers active work time but not standby - waiting for the next trip. The Oxford study found that in most months since 2023, UK Uber drivers spend more time waiting for jobs than carrying passengers. That waiting time costs the platform nothing. It costs the worker everything.

British Columbia went further in September 2025, amending its Employment Standards Act and Workers Compensation Act to extend protections to app-based workers. ILO Convention No. 193 on platform economy standards was adopted in June 2026 with 406 votes in favour - and as HCAMag reported, Canada has not yet ratified it.

The EI system is another gap. As HCAMag reported when the EI shortfall was examined, the programme was designed for stable employment and has not adapted for gig and variable work. The workers most exposed to algorithmically induced income instability are often the ones excluded from the safety net built to catch them.

What this means for how you design work

For most HR leaders the immediate question is not what to do about Uber drivers. It is what the algorithmic pay model represents as a precedent - and whether the same logic is entering your own organisation through the back door.

Several things are worth watching. Performance management systems that incorporate real-time algorithmic scoring increasingly resemble the dynamic pay architectures the gig research documents - variable compensation tied to metrics the employee cannot fully observe or predict. The growing use of AI in scheduling and task allocation gives employers the same kind of behavioural data platforms use to calibrate minimum acceptable offers. The expansion of zero-hours contracts and on-demand staffing arrangements imports the information asymmetry problem into nominally employed workforces.

The Oxford paper does not argue that algorithmic management is inherently destructive. The same data infrastructure that enables the extraction documented here could theoretically enable better, more predictable work allocation - if the incentives were different. The architecture is neutral. The incentives, as currently structured, are not.

For HR leaders, the relevant question is not whether your pay system is sophisticated. It is whether the people affected by it can understand how their pay is determined, predict what they will earn from a given amount of work, and trust that the system is not simply running a continuous experiment in finding the floor of what they will accept. Those are design choices. They sit, in the broadest sense, within the scope of every people function in every organisation now building AI into how work is allocated and paid.