Uber's COO says distribution, not technology, wins the robotaxi era

1. Short-termism on membership — ▶ 3:15

   The running debate with CEO Dara Khosrowshahi is short-term versus long-term levers. Running mobility, McDonald capped capital for Uber One. Given $40 million next quarter, his gut was to put it into price or driver supply, because ride sharing is "price, reliability, and safety" and will be even with AVs. He now says he was wrong. On many metrics Uber One is one of the more successful membership programs in the world, within "spitting distance" of Amazon Prime and Costco.

2. Why membership compounds — ▶ 6:54

   Uber judges every dollar by incremental gross bookings. A $1 incentive returning $2 of revenue is still negative ROI at a 7.5% profit margin, so the real test is lifetime value. Members ride more, and their cohorts keep riding more as they consolidate mobility on Uber and use Uber Eats instead of DoorDash or Deliveroo. They also churn less. Price promotions fade faster. The gap to Prime is value: mobility riders get about 5% cash back, comprehension is low, and Uber has no spare inventory to give away because every free ride still pays a driver.

3. Size blocks new bets — ▶ 9:54

   At nearly $250 billion in gross bookings, a new product must show a path to multiple billions of GMV within a few years to matter, which McDonald admits constrains thinking. Uber's answer is Growth Bats: of 2,000 people on a business, 100-150 work only on new things, rather than 2% of someone's time. He likes the weekly cadence of Revolut's experiments model. Corporate teams "get fat on resources" and move slower than startups. The advantage is distribution into 200 million monthly users, though every new product fights for CRM support and app placement.

4. Price caps user growth — ▶ 14:40

   The inhibitor to 500 million users is price, which McDonald concedes his IR team won't love. Going from about 6 trips a month per user to 25 means a much cheaper average trip. UberX at $35 each way in New York is "still a luxury product", and most transport happens at lower price points. The fix is cheaper modes such as London trains, bikes and scooters. The privately owned car is the most inefficient asset anyone owns, sitting idle 98% of the day. He expects a world of no car owners in 15-20 years, not 5.

5. Autonomy is existential — ▶ 17:59

   AVs are a better product in many use cases and eventually all of them: private, a place to work or sleep, and eventually safer. McDonald concedes today's gaps in speed, pickup points and weather, but says "autonomy is as bad as it's ever going to be today." Without it, Uber's core would face an existential problem. Autonomy is Uber's largest single area of investment, spread across equity stakes, vehicle purchase commitments, autonomous infrastructure and data-collection fleets. He says Uber is confident in the long-term ROI.

6. Why ATG was sold — ▶ 21:05

   Uber's secret autonomy effort began before McDonald's involvement, and he credits the foresight to Travis Kalanick. He rejects the idea that staying in would have meant control. When ATG was divested, mobility had lost 84% of its topline in 3 weeks, the core was burning billions, and Uber was trailing in autonomy, whether behind the field or just Waymo. Focus turned the core into cash flow and got the company public. He still concedes that leading AV players with control of its destiny would be a good thing.

7. Trips versus dollars — ▶ 23:10

   McDonald won't forecast the 5-year AV mix. Uber does about 300 million trips a week, versus a few million AV trips a month, so even triple-digit monthly growth stays a drop in the bucket. Human-driven trips are growing faster than the US average in San Francisco and LA, where Uber has no AVs. India and Brazil, two of the three largest countries by volume, have low fares and will take decades. The counter he accepts is that AVs could dominate San Francisco, LA, DC, Miami, New York, Boston and Chicago, a big chunk of dollars. "That's kind of the ultimate question."

8. Fixed assets need utilization — ▶ 25:20

   Investors fear a few AV winners will squeeze Uber's share, since today Uber benefits from fragmentation. McDonald's reply is that McDonald's and Starbucks own stores and apps yet still work with delivery marketplaces on terms that work for both, because expensive fixed assets need high utilization. He expects Waymo and Tesla to put cars on Uber even if they run their own apps. "In the end, distribution wins." China already has four or five AV companies, which he expects to reach eight or 10, so one global winner seems unlikely to him.

9. China's silver medal — ▶ 29:06

   McDonald ran China for only a few months before the DiDi deal. In the final weeks, Uber was burning $52 million a week on price subsidies, because investment gave leverage at the negotiating table. Travis's rule was to raise more than all competitors combined, but that was never going to hold with SoftBank backing the market. Uber also couldn't operate on WeChat, which he likens to competing in the US without email or a phone number. He says a US company becoming China's largest mobility service was never plausible, so a local exit was always the outcome.

10. AI ROI is belief-based — ▶ 34:45

    The headlines about Uber blowing through its AI budget in 4 months came from usage growing vertically against a budget set in November. McDonald's examples: a pod of 30 AI engineers paired with business teams cut weekly pricing-dollar allocation from 15 hours to 2, forecasting from 8 hours to 2, and marketing QA from two weeks to two days. Asked whether ROI is material outside coding and support, he concedes it is hard to quantify and requires a top-down, belief-based stance. His levers are tighter targets, such as 2% headcount growth instead of 10%, and combined headcount-and-compute budgets.

11. Agents and the front end — ▶ 48:20

    Uber has refused aggregators that wanted real-time APIs for every car and price, because McDonald wants Uber to be the front end, where about 80% of people start. He would fulfil a ChatGPT or Claude request like "get me my usual Uber", but a cheapest-price comparison across Uber, Lyft and Waymo is a different matter. His defence is that rides are managed transactions: pickup, driver interaction, lost items, payment. Brian Chesky's point about chat not suiting every purchase made sense to him. The worst case "hasn't played out yet. I'm not saying it won't."

The robotaxi era will be settled by utilization, not by who builds the best driver. Autonomous cars are expensive fixed assets, and a fixed asset has to be kept busy. Andrew McDonald points to McDonald's and Starbucks. They have spent billions on stores and supply chains. They run their own apps and take walk-in orders. They still work with delivery marketplaces, because high utilization matters more than pride. He expects Waymo and Tesla to behave the same way. Even if they run their own robotaxi apps or work with Uber's competitors, he thinks they will put cars on Uber's network, because Uber brings 200 million monthly consumers. "In the end, distribution wins."

This only holds if there is no single winner. McDonald says there won't be one. China already has four or five AV companies, a number he expects to reach eight or 10. Investors push the opposite case: a few strong AV players get leverage and squeeze Uber's share of every dollar. He concedes that Uber benefits from fragmentation today. His bet is that it persists. If one player reached the finish line alone, he says plainly, that would be a problem.

Scale buys him time. Uber does about 300 million trips a week and a few million AV trips a month. Those AV trips will grow triple digits month after month and still be a drop in the bucket. The human-driven business keeps growing too. In San Francisco and Los Angeles, where Uber has no AVs, it is growing faster than the rest of the US. Uber also operates in 75 countries, and two of its three largest by volume are India and Brazil, where fares are so low that autonomy needs decades to match human labor costs. Asked when most trips will be autonomous, McDonald says he cannot tell you. Probably not until AVs reach those two countries.

Distribution still has to be fed. McDonald's target is 500 million users, up from 200 million, riding 25 times a month instead of 6. The blocker is price. An UberX at $35 each way in New York is "still a luxury product", and most transport happens far below that. So Uber adds cheaper modes, such as trains in London, bikes and scooters, and expects autonomy to take cost out of the car.

Membership is the lever he now trusts most, and he admits he held it back for too long. Running mobility, his instinct was to put each spare dollar into price or driver supply, because ride-hailing is price, reliability and safety. But a membership dollar compounds. Members ride more next month. Their cohort keeps riding more over time. They consolidate mobility on Uber, use Uber Eats instead of DoorDash or Deliveroo, and churn less. A price promotion fades within weeks. That is the logic of a company deepening its moat before someone else's technology tests it.

Size is its own constraint. Uber is approaching $250 billion in gross bookings, so a new product only matters if it can reach multiple billions within a few years. McDonald calls it the innovator's dilemma: the core swallows attention, engineers and marketing. Uber's answer is Growth Bats, dedicated teams of 100-150 out of 2,000 people, rather than 2% of someone's time. Distribution helps here, but every new product also wants the Uber Eats masthead and CRM support.

Distribution is no guarantee, and Uber has the scars. In China it burned $52 million a week on subsidies in the final weeks of the negotiation. It could not operate on WeChat. It took what McDonald calls the silver medal, and he says a US company winning there was never plausible. At home he defends selling its internal autonomy group. Mobility had lost 84% of its topline in 3 weeks, the core was burning billions, and Uber was trailing in autonomy. He grants that controlling its own destiny today would be good.

Inside the company, AI is delivering, though McDonald is careful about how much he can prove. A pod of 30 engineers paired with business teams cut a weekly pricing-dollar allocation process from 15 hours to 2. Forecasting went from 8 hours to 2, and marketing QA from two weeks to two days. The interviewer pressed that, outside coding and support, enterprise ROI is still immaterial. McDonald did not dispute it. He concedes it takes a top-down, belief-based approach, because the time saved gets filled with other work and never shows up as a clean headcount cut. His fix is tighter targets, such as 2% headcount growth instead of 10%, and combined pools for headcount and compute so leaders trade one for the other. He thinks Uber could do everything it does today with fewer people in 5 years. He won't say it will have fewer.

Two objections stay open. The first is trips versus dollars. India and Brazil protect trip counts, not the dollar pool. If autonomy takes the majority in San Francisco, LA, DC, Miami, New York, Boston and Chicago, that is a big chunk of bookings. McDonald accepts this and calls it "the ultimate question." He would not put a 5-year number on it.

The second is the front end. Distribution is only an advantage if people start at Uber. About 80% do today, and McDonald has refused aggregators who wanted real-time APIs for every car and price. He would fulfil a ChatGPT or Claude request like "get me my usual Uber", but he does not want riders migrating to comparison apps. His defence is that a ride is a managed transaction: pickup, the driver interaction, a lost item, payment. Those things go wrong in ways a chat box cannot absorb. It is a fair point about today. He is careful to say the worst case "hasn't played out yet" and that he is "not saying it won't." If agents come to own the first tap, the case that Waymo and Tesla need Uber's 200 million users gets much harder to make.