How Salesforce Pilots Its Own Software
Salesforce tests its own unfinished software internally before release through its Customer Zero model, using 75,000 employees to stress-test products in real-world scenarios and refine them based on direct feedback.
The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.
Salesforce’s Customer Zero model involves 75,000 employees using new software in daily operations before public release, filing support tickets and stress-testing tools in real-world scenarios. This approach, led by Andy White of Salesforce on Salesforce Technology, treats internal teams as demanding users rather than passive testers, embedding new tools into genuine workflows and measuring performance against business outcomes. The Digital Enterprise Technology (DET) organization acts as both an advocate for internal users and a business operator, ensuring feedback directly informs product development cycles.
The Customer Zero process prioritizes greenfield projects—areas without existing solutions—allowing pilots to iterate without disrupting legacy systems. Liz Aloisi, Customer Zero & Agentforce Transformation Lead, notes that rapid adoption of AI agents led to ‘agent sprawl,’ prompting a shift toward fewer, better-built agents scaled deliberately. Product-market fit is critical; pilots work best when internal users mirror external customers, ensuring feedback is relevant and actionable.
Pilots follow a structured path: ideate use cases, configure agents, test, deploy to small release rings, and supervise closely before expanding. Jason Perez’s TechForce team, which handles IT support for 75,000 employees, piloted an AI agent for support requests, starting with simple knowledge questions. Early handle rates reached the low 30s, and iterative improvements addressed real workflow challenges, such as lost device reports, leading to the development of Agentforce Graph for precise data matching.
The Help Agent, launched on help.salesforce.com in October 2024, exemplifies Customer Zero’s real-time iteration. After over 4 million conversations, it resolves nearly 70% of inquiries without human intervention, with a target of 80% by year’s end. Internal testing of features like inline citations improved adoption by 27%, validating that solutions proven internally translate effectively to external customers. Performance targets are dynamic, with teams raising standards as agents improve, ensuring continuous product refinement.