Andrew Macdonald on Uber’s Membership Strategy, Autonomy and AI

Harry Stebbings welcomes Andrew Macdonald, Uber’s president and COO, and asks how a leader can drive execution while remaining trusted by colleagues. Macdonald says there is no single career formula, but consistent decisions aimed at what he believes is best for the company help build followership.

Macdonald pairs that consistency with deep knowledge of the business. He says colleagues are more willing to follow a difficult decision when they understand the principle behind it, even though a leader will sometimes get the decision wrong.

Asked for an example of being wrong, Macdonald says he initially favored short-term levers such as lower prices and greater driver supply over the longer-term development of Uber One. He and CEO Dara Khosrowshahi debated the tradeoff, and Macdonald says he later changed his mind.

His concern was allocation: a dollar committed to membership benefits could not simultaneously lower the price of a ride or improve marketplace supply. He still regards price, reliability and safety as central to ride-hailing, but now thinks he undervalued the longer-term effect of membership.

Macdonald says Uber One has become an efficient consumer investment on the company’s measures. He believes the effect becomes clearer with longer observation because members can increase usage over time. The performance assessment is the executive’s account, not an independent evaluation of the program.

He describes incremental gross bookings as one measure used to assess money put into the marketplace. An increase in bookings is not the same as profit: in his example, a promotion can generate more transactions without recovering its cost immediately, with the hoped-for return dependent on future customer engagement.

Macdonald says members may consolidate more trips on Uber and use benefits across mobility and delivery, potentially reducing churn. He contrasts that longer tail with a short-term promotion whose effect may dissipate. These are his explanations of observed company behavior, not proof that every member behaves the same way.

Asked how Uber One might approach the scale of established membership programs, he says it needs more consumer value and better understanding of existing benefits. He believes some mobility riders do not fully appreciate what their membership already provides.

A useful benefit should have high perceived value relative to its cost, Macdonald says. That is difficult in ride-hailing because giving a member a ride still requires paying a driver. Unlike a business with unused fixed inventory, the marketplace has substantial variable costs on each transaction.

Stebbings asks what new revenue line might become significant in five years. Macdonald says the company’s existing transaction scale makes the test for significance very large: a new business may need a plausible path to billions in transaction volume to command meaningful attention.

He recognizes the innovator’s dilemma: a large core operation can absorb management time, engineering capacity and marketing resources, leaving little for small ideas. Uber’s growth-bets program is intended to protect new initiatives from that pull.

Macdonald says the program assigns dedicated people to products that have not yet proven product-market fit or unit economics. An employee working on a new idea for only a small fraction of their time is unlikely to move it forward as effectively as a focused team.

Stebbings describes another company’s practice of reviewing experiments weekly. Macdonald likes the short cadence and the discipline of asking for further resources. He warns that an internal startup can become slower and more expensive if it assumes a large parent company will always supply more people and cash.

Uber’s advantage, he says, is distribution through an app used by a large consumer base. But access is not automatic: new products compete for limited screen space, marketing and customer communication. A company still has to decide which products deserve attention and demonstrate value.

Asked what could prevent Uber from reaching more users, Macdonald points to price. He says many ordinary transportation trips occur at prices below a typical private ride-hailing trip, so reaching more people and more frequent use may require a lower average transaction cost.

He names public transport, bikes and scooters as possible components of a broader mobility mix. Macdonald predicts that over a longer horizon fewer people may need to own cars because several services could meet different travel needs. The timing and extent of that change are his forecast.

Macdonald calls autonomous vehicles strategically important because he expects the experience to improve and become attractive in more situations. He acknowledges current limitations in speed, pickup points, weather and geography. His expectation that future systems will be safer or preferred by most riders is a forecast, not a demonstrated guarantee.

He says Uber is investing through several channels, including stakes in companies, vehicle commitments, infrastructure and data collection. He describes autonomy as a major standalone investment area and says he expects a long-term return, while the actual economics remain uncertain.

Stebbings asks whether Uber would be better positioned if its earlier in-house autonomous effort had continued. Macdonald credits founder Travis Kalanick with recognizing the technology’s importance early, while saying he was not involved in starting that work.

Macdonald says predictions about autonomous deployment once ran well ahead of reality. He resists assuming that simply retaining Uber’s former unit would have made the company a leader. A counterfactual outcome cannot be observed, and technological progress required more time than many forecasts suggested.

He recalls the divestiture of Uber’s autonomous unit during the pandemic, when mobility demand had collapsed and the core business was losing money. In his account, the unit was not leading the field, and management focused on strengthening the core. He says he would still welcome a leading internal autonomy capability today.

Asked what fraction of trips could be autonomous in five years, Macdonald says it is hard to predict. He cites Uber’s large existing weekly trip volume and says autonomous trips are starting from a small share. He also expects the human-driven business to change, making the denominator a moving target.

He says Uber operates in many countries with different labor costs, fares and infrastructure. In his view, autonomous service may take longer to become economical in lower-fare markets such as India and Brazil. Those timelines are his predictions and depend on technology, regulation and local economics.

Stebbings notes that autonomous trips could represent a modest global share by volume while accounting for a significant portion of bookings in higher-fare cities. Macdonald agrees that adoption in major US markets could matter financially even before autonomous vehicles dominate trips worldwide.

Asked whether Waymo or Tesla is the larger threat, Macdonald says he expects more than two successful technology providers, while believing both named companies could succeed. He acknowledges an investor concern that a small number of vehicle providers could gain bargaining power over marketplaces.

Macdonald compares the problem with food delivery: large merchants may have their own customer channels but can also use marketplaces to increase utilization. He argues that autonomous vehicle owners might similarly value access to Uber’s demand. This is a strategic thesis, not a confirmed agreement with every provider.

He says distribution can matter even when a technology partner has its own app, because expensive vehicles need utilization. At the same time, he acknowledges that a world with only one successful autonomous provider could create a problem for Uber. His expectation of several winners is a forecast.

The conversation moves to China. Macdonald says Uber exited its China business in 2016 and explains that he oversaw the business for only a few months before the agreement with Didi. He says other people led the negotiations, so his perspective is that of an operating executive rather than the principal negotiator.

He recalls that both companies were heavily capitalized and competing intensely while a deal was being discussed. In his account, market share and ongoing investment affected negotiating leverage. These are his recollections of a competitive period, not an independently verified reconstruction of the talks.

Macdonald says Uber was spending roughly $52 million a week on price subsidies in China near the end of negotiations. He uses that figure to illustrate the intensity of competition. The amount is his retrospective account and is not verified by the interview.

He says access to local distribution channels was a major challenge and recalls a period when Uber could not operate on WeChat. He also says local partners helped the business. His description concerns his company’s experience at the time and should not be generalized to all foreign companies or Chinese platforms.

Asked whether he was pleased to exit, Macdonald says finishing behind a competitor was disappointing, but he did not think Uber would ultimately become the largest mobility provider in China. He cites competitive and geopolitical factors as his personal assessment, not as an established account of all causes.

He says the hardest part of the exit was its effect on employees who had committed to Uber’s China team. In his account, the company sought roles in its global organization for as many people as possible. He remembers the experience as emotionally difficult despite believing the transaction was a comparatively good outcome for Uber.

Macdonald says Uber exceeded the AI budget it had planned for the whole year within the first few months. He describes the overrun as a sign of demand for the tools, not a financial emergency, while acknowledging that it prompted a closer look at where the spending was going.

He says an earlier comment about the difficulty of linking consumer-facing features directly to AI spend was interpreted by some listeners as an admission that Uber could find no return, or did not understand what it was doing. Neither was his intended point: the company was still learning to measure benefits that appeared across many activities.

The problem, in his telling, is attribution. A new capability may contribute to a better product without producing a clean line between a particular model bill and a single outcome. He argues that leaders should acknowledge that ambiguity rather than pretend every use can be evaluated in the same way.

Uber paired roughly 30 AI engineers with business and general-and-administrative teams to examine specific processes. Macdonald says the team looked for recurring work where people spent substantial time gathering information, preparing decisions or checking outputs.

He gives several internal examples: a weekly pricing-allocation process that he says fell from 15 hours to two, financial forecasting from eight hours to two, and marketing quality assurance from two weeks to two days. These are the company's reported process estimates, not independently verified measurements.

Those gains do not automatically translate into an equal reduction in staff or a neatly identifiable profit figure. Macdonald says employees can use the time for more valuable work, and that isolating the effect of a tool from the rest of a person's job is difficult.

His proposed management response is to make the future staffing plan reflect the productivity leaders expect, rather than demand a precise attribution for every saved hour. That still leaves room for teams to grow where the business and the work justify it.

On budgeting, Macdonald suggests bringing the money for people and AI compute into a shared conversation. A manager who can choose between additional staff and tools may make better decisions than one who receives a fixed headcount allowance while technology spending sits elsewhere.

The comparison also requires visibility into costs and the value of the task being performed. A more capable model is not necessarily needed for every job; routing different tasks to different tools can help teams spend in proportion to what they gain.

Stebbings raises the risk of measuring workers simply by how much they spend or use. Macdonald agrees that either extreme can be misleading: high usage may be wasteful, while low spending may also be a mistake if a tool could meaningfully improve the work.

He still sees a role for adoption metrics as a way to identify teams that might not have discovered useful tools. The goal, he argues, is to understand the output and the task, not reward a usage number in isolation.

Asked about the workforce in five years, Macdonald expects the same scope of work could be done with fewer people. He does not predict a specific total headcount: Uber may also enter new areas and create work that does not exist today.

He points to analytics, customer support and sales as examples of functions that could change. His distinction is between the staffing required for a fixed set of tasks and the size of an evolving company; the former, he believes, is likely to fall.

Uber works with frontier AI labs, but Macdonald sees its own competitive position as grounded in a physical-world marketplace, local operations and the requirements of running transportation and delivery services. That is his assessment of Uber's circumstances, not a guarantee that its partners will never compete with it.

In his view, matching a rider with a driver is only one part of the service. The company also has to operate across local markets and handle what happens before and after the match. Those practical responsibilities shape how he thinks about the threat from a new consumer interface.

AI agents could nevertheless alter how customers choose a ride. Macdonald distinguishes between a request to an assistant to get an Uber and a request to find the cheapest available ride: the latter can shift the comparison and the customer relationship to an intermediary.

That raises a question Uber has confronted before with price-comparison products. Macdonald says the company has historically been reluctant to provide the real-time data needed for those experiences because it values the relationship with customers in its own app.

A transaction is more involved than displaying a price. Someone has to manage pickup details, payment, communication with the driver, lost items and other problems. Macdonald says those responsibilities need to be considered alongside the question of where a customer begins a booking.

For an assistant to be genuinely useful, he argues, it must do more than hand a customer a quote and disappear. The companies involved have to decide where the transaction is completed and who helps if it goes wrong.

He says Uber is open to partnerships with large AI interfaces, including OpenAI, while expecting negotiation over data, transaction flow and responsibility for resolving problems. The interview does not establish that any particular arrangement has been finalized.

On the proposed Delivery Hero transaction, Macdonald describes a chance to reach markets more quickly and build on established regional brands. He discusses the potential deal as a proposal subject to regulatory and shareholder processes, not a completed acquisition.

He points to businesses in Korea, Latin America and the Middle East as examples of the geographic footprint under discussion. Local brands and operational knowledge, in his view, can be valuable even when Uber brings global technology and experience.

Macdonald sees a connection between the delivery businesses and Uber's existing mobility network. He argues that a broader relationship with customers can make the overall platform more useful, while acknowledging that each market has its own competitive and operating conditions.

He thinks delivery could eventually become a larger business, though that is a forecast rather than an established result. The strategic appeal, as he frames it, is the combination of reach, frequency and the opportunity to serve consumers across more needs.

Macdonald has also taken direct responsibility for Uber Eats while serving as president and COO. He says the additional role has made him conscious of the burden on colleagues and of the need to avoid treating long hours as the only way to solve a problem.

Food delivery is a three-sided marketplace: consumers, couriers and restaurants must all find the service useful. He describes it as having more dimensions to manage than a ride marketplace, with restaurant operations and the quality of the food adding complexity.

In the United States, he recognizes DoorDash as the leader and praises both the company and its CEO, Tony Xu. Competition, he says, requires Uber to learn from a strong rival and improve the experience it offers to all sides of the marketplace.

Stebbings asks whether Uber's path in delivery might have been different under Travis Kalanick. Macdonald is cautious about the counterfactual: it is impossible to know with confidence what another leader would have done or what the result would have been.

Asked about past acquisition possibilities, he similarly avoids claiming to know whether a DoorDash deal could have happened. On Uber's Postmates acquisition, he notes that he was not running delivery at the time and cannot give a definitive account of the decision.

He does see value in the regional recognition of brands such as Postmates. For a marketplace expanding across different cities, a familiar local brand can contribute a customer relationship that takes time to build.

In the quick-fire discussion, Macdonald expresses personal optimism about future advances in longevity and medicine. His hopes about future treatments are speculative, not clinical evidence or a promise that particular diseases will be cured.

Asked about his own AI use, he highlights voice interaction with OpenAI tools while saying he appreciates the capabilities of other systems as well. The answer reflects his personal habits rather than a comprehensive evaluation of the products.

On Kalanick, Macdonald emphasizes creative problem-solving and the discipline of making reasoning explicit. A solution becomes more useful to an organization when people can explain the principle behind it and apply that principle to a different problem.

He recalls learning to break difficult decisions into their component parts instead of accepting the first framing of a problem. The lasting lesson was a way of thinking that a team could reuse, not merely a single decision made by a founder.

Macdonald describes Dara Khosrowshahi's leadership as combining a clear business head with care for people. He distinguishes the authority of a position on an organizational chart from the trust that makes colleagues willing to follow someone.

He credits different leaders with different strengths and says both approaches have shaped him. His long tenure at Uber, he suggests, also reflects loyalty to the people he works with and the opportunity to keep learning as the company changes.

The advice he recalls from Rachel Whetstone is to say yes to opportunities. Taking on an unfamiliar assignment can create learning whether it succeeds or fails, and can open paths that are hard to see in advance.

Stebbings and Macdonald close by thanking one another for the conversation. The interview returns to a theme running through the discussion: operating a large platform requires choices about people and execution as well as technology and strategy.