DoorDash is a technology company that connects people with the best in their cities through door-to-door delivery. It connects consumers with their favorite restaurants, empowers local businesses to expand their reach and generates new ways for people to earn, work and live. With over 450,000 merchants in its network, thousands of food orders are processed daily, and its Global Operations team manages the ecosystem of merchants, Dashers and customers.

Managing an ecosystem of merchants, Dashers and customers is an operational challenge. Each time a customer places an order through the DoorDash app, the merchant receives it digitally or over the phone. With more than 450,000 merchants in the network, a portion prefer to receive order details by phone, and all of those calls were previously handled by an offshore call center.

As DoorDash grew, so did order volumes and the cost of that operation. Rising revenue was dampened by inefficient and ballooning call center costs. Demand also fluctuates dramatically, with most orders placed at meal times and on weekends, so the call center had to be overstaffed to cover peaks, leaving unused agent minutes when demand was low and driving agent turnover.

DoorDash wanted a high-quality solution that could handle erratic phone order traffic, minimize escalations and continue to lower costs, while placing orders faster to improve customer satisfaction.

DoorDash's Global Operations team used Replicant's voice AI to place 100% of eligible phone orders to restaurants on behalf of DoorDash customers. Within six weeks of launch, Replicant's Thinking Machine went from making zero outbound calls a day to more than 35,000.

Because the Thinking Machine handles hundreds of calls concurrently and scales elastically with demand, DoorDash could meet order volumes minute by minute, regardless of how demand shifted through the day or across seasons. DoorDash pays only for the minutes the Thinking Machine is on the phone, so it no longer carries the cost of unused agent capacity, and legacy telephony and internal tool maintenance costs were eliminated.

Replicant continuously retrained its machine learning models so the Thinking Machine communicated orders more clearly and succinctly to merchants. These improvements reduced hang-up rates and re-engaged merchants who had previously preferred a human agent. Through testing, Replicant also identified the optimal callback delay and frequency when calls reached voicemail or a busy line, which improved the placement rate and avoided failed orders caused by missed connections.

In the first six weeks the Thinking Machine's placement rate was 89% as effective as the offshore agents; by the third month it reached 94%, within 5 percentage points of the call center's 85% benchmark placement rate.

• 94% success rate, within 5 percentage points of the offshore call center's 85% benchmark placement rate
• 49% reduction in average handle time
• Scaled from zero to more than 35,000 automated outbound calls a day within six weeks of launch
• 16% fewer calls escalated to DoorDash's corporate customer service team
• Legacy telephony and internal tool maintenance costs eliminated; DoorDash pays only for minutes on the phone
• Merchants preferred the AI caller because the Thinking Machine is clear in noisy kitchens