Salesforce's Agentforce Sales deployment reveals a fundamental gap between AI agent capability and enterprise readiness: the technology works, but only when organisations treat data quality and governance as prerequisites rather than afterthoughts. Kris Billmaier's account of internal testing shows that agents can outperform human sellers on specific tasks—email composition, for instance—and that Salesforce identified over 60% of seller time spent on non-selling activities, creating a clear productivity case. Yet the transition from controlled pilots to live customer-facing work exposed critical vulnerabilities. Early agent emails required "careful tuning" before reaching acceptable quality, demonstrating that production environments reveal failure modes that demos obscure. For CX teams already managing customer data across multiple touchpoints, this signals a hard truth: Agentforce and similar platforms cannot function effectively on CRM data alone. Billmaier emphasises that "nuggets of gold" exist in call recordings, email threads, and unstructured customer interactions—the very signals that most CX platforms struggle to capture, normalise, and feed back into sales workflows. The implication is stark: teams running fragmented data stacks will see agents make poor decisions, not because the AI is flawed, but because the input is incomplete.
The governance framework Salesforce built—testing, quality assurance, live monitoring, permissions, and policy controls—reflects a maturity that most enterprises have not yet achieved. What does this mean for teams already running Agentforce or considering deployment? The lesson is not about agent capability but about organisational discipline. Billmaier's advice to "pick a use case, go deep, and treat the agent as if you would a junior employee" reframes AI implementation as a change management problem, not a technology problem. For CX leaders, this creates both opportunity and risk. The opportunity lies in using agents to reclaim seller time for high-value customer interactions, but only if data governance, agent supervision, and clear performance metrics are in place from day one. The risk is that teams rushing to deploy agents without addressing data quality or establishing monitoring will amplify existing problems—poor data hygiene, inconsistent customer context, and eroded trust. Given the security landscape around AI agents, where credential sharing and multi-turn attacks remain common, the governance question becomes even more pressing for CX professionals responsible for customer data integrity.
Francesca Roche sits down with Kris Billmaier, EVP and General Manager, Agentforce Sales at Salesforce, to discuss what enterprises can learn from moving AI agents out of the demo environment and into real sales workflows. As enterprises push AI agents from polished pilots into live sales environmen