DoorDash has introduced an AI agent named Vera, designed to assist over 10,000 employees with business-related inquiries. The tool processes data from more than 200,000 datasets, which collectively amount to 350 petabytes. It handles tasks ranging from simple metric lookups to complex strategic analysis.
The company evaluates Vera using a proprietary test consisting of over 1,900 questions across 19 business areas. Through internal optimization of data organization, the agent's pass rate increased from 43% to 90%. DoorDash noted that structuring data proved more effective than relying solely on model upgrades.
For high-stakes tasks like forecasting and cause-and-effect analysis, Vera now requires human approval before executing its proposed plan. This adjustment addresses previous issues where the agent provided shallow responses. Performance on these complex tasks currently fluctuates between 64% and 100%.
The deployment of Vera is part of a broader AI strategy at DoorDash. While the company continues to refine the agent, it is still evaluating whether the faster access to insights directly improves core business performance metrics.
