DoorDash

Every Dasher who can’t finish a delivery, can’t get paid, or can’t get into the app arrives at Dasher support. With over 1.6 million Dashers around the world served by 24,000 agents, a thirty-second improvement in handling time is worth millions a year and an outage isn’t an inconvenience, it’s somebody’s livelihood.

For our teams, it was even higher stakes. The pandemic meant a 50% reduction in agents, a 50% increase in demand, and a short-term shift from large call centers to a distributed work-from-home agent base. This meant massive, high-speed development of automated tooling, agent assist, fraud prevention tooling and more.

What was in place had been built for a startup, and by the time I arrived DoorDash had stopped being one. The job was to replace it with a fully operational, scalable software stack that could carry millions of users while maintaining the startup pace of innovation and development that had made it a household name.

The answer was production-level automation and operational overhaul. AI-enabled flows built on Amazon Lex and Google Cloud now handle more than half of all support traffic. Deflection, collection and escalation saved an estimated $82M a year. Automating half of anything at that volume is not one launch. It is a long sequence of small corrections against live traffic, which is why the release cadence had to be fixed first.

When you are growing as fast as DoorDash, shipping twice a week is not enough. At one point, we had 90% of our international call center workforce under government-imposed shelter-in-place orders. We couldn’t overcome that change in our capacity with updates on Tuesday and Thursday. We had to figure out a way to build fast and build well.

That meant changing how our teams operated. I rebuilt Dasher Experience around outcomes, empowering teams to deliver product improvements, operational efficiencies, and everything in between. The key wasn’t shipping more, it was building better Dasher support. It worked. Our annual releases went from 107 to 1,049 (an 880% increase) and, more importantly, we saw customer satisfaction go up, deliveries go up, and the number of agent tasks and time on task go down.

In addition to building out support in-house, I led more than $101M of RFPs and implementations across contact-center-as-a-service, workforce management and quality management systems.

Platform purchases at that size are not procurement exercises, they are architecture decisions with a five-year blast radius. They set the ceiling on what the support organization could do long after I left.

Across the six-month peak of COVID, the team designed, built, deployed and iterated 14 mission-critical initiatives needed to combat the simultaneous increase in delivery demand and loss of agents.

That is the clearest test the throughput work ever got. The capacity to ship into a live support platform during a demand shock is the direct dividend of having fixed the way the organization operated first.

Product, Business Applications and Operations: the software, the agent systems, and the processes wrapped around both.