How an AI CRM System Generated 1M+ Recommendations

Salesforce's Revenue Orchestration team built a Sales Agent on Agentforce that processes hundreds of thousands of CRM opportunities nightly within a nine-hour window, generating 1.04 million monthly recommendations for 13,000 sellers. Key engineering challenges included scaling past a 300 requests-per-minute platform limit using a message queue–driven architecture, reducing per-opportunity latency from 1.35 seconds to ~600ms, and ensuring AI recommendations are explainable with citations and secured via layered access controls. The team also built a reusable, persistent platform entity for AI-generated actions shared across multiple agents, and a low-code/no-code modular architecture for enterprise configurability. In production, the system contributes to a 75% reduction in time spent on manual CRM updates.