20VC: Uber President on The Untold Uber Stories: Travis, China and Self-Driving | Why Autonomy Is Existential | How to Beat DoorDash to #1 in Food with Andrew MacDonald

Summary

Andrew MacDonald gives a rare inside view of Uber’s operating model, from membership economics and product selection to autonomy, China, delivery, and AI. He says Uber One is one of the company’s most efficient levers because it lifts frequency and LTV through cross-use of mobility and Eats, while Uber’s growth-bets framework exists to protect new products from being swallowed by the core business. On autonomy, he argues the threat is existential because autonomous vehicles are already better in many use cases and will keep improving, though adoption will take much longer in markets like India and Brazil. He also says Uber is applying AI first to internal operations, with concrete time savings in pricing, forecasting, and QA, and expects the company to do more with fewer people over time. The China stories underscore how brutal marketplace competition can be, but also how Uber’s distribution and persistence shaped the outcome.

Notes

• Uber is already operating at massive marketplace scale: MacDonald says the core platform handles about 300 million trips per week and 200 million monthly consumers.
• He frames ride-hailing around three durable product levers: price, reliability, and safety.
• He says AI should eventually reduce Uber’s required headcount, even if the near-term effect is hard to isolate.
• Uber One is described as one of the world’s more successful membership programs, approaching the scale benchmark set by Amazon Prime and Costco.
• MacDonald says Uber evaluates membership and marketplace incentives by incremental gross bookings, not just headline ROI.
• He says membership works by increasing engagement, ride frequency, and cross-usage across mobility and Uber Eats.
• He admits he was too short-term oriented on membership and on allocating dollars to price cuts versus longer-term retention and product-building.
• The recurring trade-off inside Uber is whether each incremental dollar should go to lower prices, membership, or new business units.
• Uber’s innovation structure is meant to solve the classic innovator’s dilemma inside a very large incumbent platform.
• He says Uber uses dedicated Growth Bets resources so new products are not starved by the core mobility and delivery businesses.
• Distribution is Uber’s core advantage because a new product can be plugged into 200 million monthly users and scaled quickly if it works.
• He says new products only matter if they can plausibly become multibillion-dollar GMV businesses.
• He argues price is the biggest blocker to reaching 500 million users and higher ride frequency.
• The growth thesis is to lower average transaction cost so riders use Uber more often, moving usage from roughly six trips per month toward 25.
• He points to cheaper modes such as trains, bikes, and scooters as part of the solution, not just price cuts within UberX.
• He calls the privately owned car an extremely inefficient asset, saying it sits idle about 98% of the day.
• Autonomy is existential because it is already a better product in many cases and improves over time.
• He says autonomous vehicles are still slower than human drivers in many situations today, but the trajectory is favorable.
• Uber is making its largest standalone investment in autonomy through equity, purchase commitments, autonomous infrastructure, and data-collection fleets.
• Travis Kalanick saw autonomy years before 2016 and pushed Uber toward it early.
• When Uber divested ATG, the company was in COVID stress, mobility revenue had fallen 84% in three weeks, and the core business was not generating cash.
• Uber’s AV business is still tiny: a few million trips per month versus 300 million core trips per week.
• India and Brazil will likely delay autonomy for decades because fares are far below autonomous operating costs.
• He expects more than two winners in autonomy and cites China as evidence that the market stays fragmented.
• Uber believes autonomous fleet operators will still want its distribution because they need utilization for expensive fixed assets.
• China’s AV market already appears to have four or five winners rather than one dominant player.
• Uber exited China in 2016.
• MacDonald says the China exit still produced a better outcome than most Western tech companies achieved in that market.
• He recalls the final weeks of the Didi fight as a subsidy war that burned $52 million per week in China.
• The China market was unusually chaotic, including a merger anecdote where about 200 employees were on both payrolls.
• Uber’s China competition was hampered by losing WeChat access for a period, making operations much harder.
• Uber’s AI push is mostly about internal productivity, not just consumer-facing features.
• Uber formed a 30-person pod of AI engineers paired with business teams to redesign workflows process by process.
• AI cut one pricing-allocation workflow from 15 hours to 2 hours.
• AI cut a finance forecasting workflow from 8 hours to 2 hours, while improving precision by adding nuance.
• AI also shortened marketing QA from two weeks to two days.
• MacDonald says AI savings are hard to measure because freed time usually gets reabsorbed into other valuable work.
• He proposes combining headcount and compute into one budget pool so trusted operators can allocate between engineers and inference spend.
• Uber is using smart routing internally to choose different models for different tasks.
• The company also tracks AI usage and cost with internal dashboards so employees can see consumption.
• He argues frontier models should not be used wastefully for trivial lookup tasks.
• Visibility into AI cost and usage can make workers more conscious of the value of their actions.
• He warns that blunt AI leaderboards can cause people to optimize the metric instead of the outcome.
• Uber wants every support agent using the AI assistant, so adoption metrics matter when the tool directly improves work quality.
• He expects the biggest AI impact to land in producing functions like support, sales, analytics, and content generation.
• Uber’s physical-world business makes it less easy for frontier labs to disrupt directly.
• He worries more about AI products that redirect the consumer front end, such as queries beginning in ChatGPT or Claude.
• A simple query like ‘get me an Uber’ is not a threat, but comparison and aggregation apps could shift the customer relationship.
• Uber’s managed transactions are operationally complex, with multiple failure points between booking and completion.
• He argues chat is not the right interface for every booking flow because some experiences are more visual and more managed.
• Uber wants to be where consumers already are, including on large AI platforms.
• Transactions can originate in another channel and still complete in the Uber app.
• Agent-mediated booking will need clear operational responsibility, such as who handles hotel follow-up messages.
• Delivery Hero was attractive because delivery was growing fast, strong in local brands, and helped expand Uber geographically.
• Delivery Hero’s brands in Argentina, Korea, and the Middle East have strong consumer mindshare and are not easy to displace.
• MacDonald has recently been directly running delivery in addition to mobility.
• He says the delivery business is complex, with a three-sided marketplace and more variables than mobility.
• He says Uber can get stressed and may run teams too hard, but it tends to perform best in crises.
• Travis Kalanick is described as a creative problem solver who could shift a room’s thinking in minutes.
• He summarizes Dara Khosrowshahi’s philosophy as: management comes from an org chart, leadership comes from the heart.
• MacDonald says he stayed through both the worst and best periods because it felt like the right thing for Uber, regardless of CEO.
• Rachel Whetstone’s advice to him was to say yes and jump at the next adventure.
• He says voice is his most-used AI feature by far, and estimates he uses OpenAI’s voice engine about 50 times more than anything else.