Key Takeaways

  • Trust and safety within a high-volume logistics system is a constant system-building problem.
  • Combating 'offline fraud,' such as false claims of non-delivery, requires building proprietary internal signals.
  • Proactive measures, like analyzing chat for verbal altercations, can prevent physical incidents before they escalate.
  • For public safety threats, rapid, integrated alert systems with law enforcement offer critical response times.

The Method

Tony Xu, CEO of DoorDash, frames trust and safety not as a peripheral concern, but as a core engineering challenge. He describes DoorDash's operation as a "large state machine." This view means trust and safety aren't solved by a single fix; they require continuous system building to monitor and respond to evolving threats.

DoorDash distinguishes between online fraud and more challenging 'offline fraud.' Online fraud, like payment card issues, often benefits from partnerships with companies like Stripe, which provide vast datasets for detection. However, offline fraud, such as a consumer falsely claiming an order was never dropped off, requires a different approach.

“On the flip side, there's also offline fraud, which is very tricky to catch. For example, let's say that a consumer suggested that an order was never dropped off,” Xu states. To combat this, DoorDash builds its own signals, like precise mapping of delivery locations, to verify claims and detect patterns of abuse. This internal data generation is key.

Beyond fraud, physical safety is paramount. Xu detailed proactive systems. One example is "SafeChat," a feature developed a few years ago. "We noticed that prior to any physical altercation that may, unfortunately, happen between audiences, 90-something percent of the time, it's always preceded by a verbal altercation," Xu explains. Detecting these precursors allows intervention before incidents turn physical.

DoorDash also built a rapid alert system for major public safety threats. Xu recounts an incident where a Dasher reported a shooting in Manhattan. “Within minutes of a Dasher seeing the shooter walk in to the building, we have an alert system that doesn't just alert all Dashers and consumers and merchants within that vicinity, but also all local law enforcement. Actually, we were the first to report that to the NYPD who try their best to respond in time.”

Where This Breaks Down

DoorDash's method for trust and safety relies heavily on scale and internal resource investment. Building proprietary signal detection systems, like precise mapping for delivery verification or natural language processing for SafeChat, demands significant engineering talent and data volume. A smaller startup might not have the billions of orders per year that provide the data density needed to train such systems effectively or the budget to develop them from scratch.

This approach also requires a high tolerance for operational complexity. Maintaining a "large state machine" means constant iteration, monitoring, and adaptation to new fraud vectors and safety threats. It is a never-ending battle, not a one-time solution. While effective for DoorDash's scale, attempting to replicate this without similar resources can lead to overengineering and diverted focus for early-stage companies.

What to Do With This

This week, analyze your business's "offline" incidents – any problems that existing third-party fraud tools or standard software cannot directly verify or prevent. For the single most damaging or frequent incident type, map out 3-5 specific internal data points you could collect that would uniquely signal its occurrence or precursor. Don't outsource this thinking. Start building the system to gather those signals, even if it's just a manual log initially. This forces you to own your trust and safety data.