Three-quarters of organizations have rolled back or shut down AI agents deployed in customer service, according to Sinch research spanning over 2,500 industry leaders. The primary culprits aren't technical limitations but governance failures: customer data exposure (31%), hallucinations and brand risk (22%), and lack of auditability (16%). These aren't theoretical concerns—they manifest as PII surfacing in unintended interactions, agents confidently delivering false information to customers under your brand name, and systems that leave no audit trail when things go wrong. The paradox is stark: 88% of respondents plan full production deployments within a year, and 98% are increasing AI investment despite the rollbacks, suggesting organizations view these failures as implementation problems rather than fundamental viability issues.
The data reveals a counterintuitive pattern that should concern CX leaders already committed to agentic AI. Organizations with mature governance frameworks report rollback rates of 81%—higher than the overall average—because better monitoring surfaces failures faster rather than preventing them. This inverts the assumption that governance solves the problem. Instead, engineering teams are spending 84% of their time building safety infrastructure that should be embedded in communications platforms themselves, creating what Sinch calls a "guardrail tax" that diverts resources from actual customer experience improvement. For teams running Agentforce, Freshdesk's AI agents, or similar platforms, this raises a critical question: are your vendors architecting safety into the core product, or are you inheriting the burden of bolting it on afterwards?
The investment imbalance compounds the issue. Three-quarters of respondents are prioritizing trust, security, and compliance infrastructure (75%) over AI technology itself (63%), yet this hasn't prevented widespread failures. The most popular use cases—chatbots and email responses—remain the riskiest channels for brand exposure. With customer confidence in responsible AI use already fragile and search behaviour shifting away from automated channels, the real risk isn't that AI agents will disappear from customer service, but that poorly governed deployments will erode trust faster than mature implementations can rebuild it. The question for your organization isn't whether to deploy AI agents, but whether your platform vendor has solved the governance problem or simply shifted it onto your team.
AI agents aren’t cutting it in customer service IT Pro