#37 Cyrus Safaie: Inside DoorDash's Three-Sided Optimization Problem

Make better business decisions with data and AI—subscribe to The Decision Intelligence Lab Newsletter at ⁠⁠https://decisionintelligencelab.substack.com/⁠⁠. What does it actually take to run a marketplace across thousands of micro-markets in real time? Cyrus Safaie, Former Director of Engineering and AI/ML at DoorDash joins Mike and Vijay to pull back the curtain on how DoorDash balances supply and demand, pays dashers, and predicts delivery times at massive scale. Cyrus shares why the company started with spreadsheets instead of algorithms, how a surprise Netflix boxing match drove a Super Bowl-sized demand spike with zero warning, and why the best automated systems treat human judgment as an input rather than an override. He also previews his new venture: using AI to build operationally heavy companies run by a single domain expert or tiny team. Timestamps 0:00 - Preview 1:01 - Meet Cyrus Safaie 2:00 - What people underestimate about DoorDash: hyper-local markets 3:28 - Scaling by doing things that don't scale 6:48 - Optimizing a three-sided marketplace & the trade-offs 11:20 - Who resolves conflicts between dasher, merchant, and consumer teams 13:00 - Balancing supply/demand and dasher incentives in real time 16:48 - Reactive Real-time decisions and the Netflix Tyson fight 21:30 - Human input vs. human override: how to automate the right way 27:27 - Fixing ETAs by predicting your own model's errors 29:35 - Cyrus's new venture: AI-native operationally heavy companies 33:38 - Where the human stays in the loop: one-expert companies 36:36 - Advice for students: curiosity and "torturing your brain" What You'll Learn \- Why DoorDash is really thousands of tiny, semi-isolated markets and why it optimizes locally before globally \- How DoorDash scaled by starting with spreadsheets and manual judgment before building algorithms \- How a three-sided marketplace balances dashers, merchants, and consumers \- The difference between proactive and reactive dasher incentives, and how real-time mobilization works \- Why human input into automated systems beats human overrides of them \- How DoorDash improved ETAs by building models that predict their own errors \- Cyrus's vision for AI-native, operationally heavy companies run by a single domain expert (or tiny team) plus an AI operating system Follow the show Apple: ⁠https://podcasts.apple.com/in/podcast/the-decision-intelligence-lab/id1811085064⁠ Spotify: ⁠https://open.spotify.com/show/0lFoAVKqJHTYSZNpeN61ou?si=0ae973aab0174b3b⁠ Connect with guest Cyrus Safaie: ⁠https://www.linkedin.com/in/hsafaie/ Connect with hosts Prof. Vijay Mehrotra (University of San Francisco): ⁠https://www.linkedin.com/in/vijay-mehrotra-ba9498/⁠ Prof. Michael Watson (Northwestern University): ⁠https://www.linkedin.com/in/michael-watson-07600a1⁠ About the podcast The Decision Intelligence Lab podcast delivers real-world insights for data professionals, business leaders, and anyone seeking to leverage data & AI for smarter decision-making & successful business outcomes. For business inquiries, email at ⁠[decisionintelligencepodcast@gmail.com⁠](decisionintelligencepodcast@gmail.com)