Enterprise AI Adoption Stalls as Trust and Governance Hurdles Delay Deployment

Despite surging investment in enterprise 'AI customer experience', many companies are finding that moving from pilot projects to real-world deployment is where momentum stalls—often at the last mile. The issue is not a lack of ambitious prototypes, executives say, but the difficulty of proving that AI systems can handle edge cases safely enough to satisfy security teams, legal departments, and internal governance committees.

Chris Mina, chief product and technology officer at LivePerson ($LPSN), described the pattern in an interview with theCUBE on the sidelines of Google Cloud Next 2026, according to reporting by SiliconANGLE on Apr. 23 ET. “Companies build impressive proofs of concept and workflows, but the moment they encounter an edge case in production, they stop,” Mina said, arguing that the biggest barrier is not technical capability but insufficient confidence backed by evidence.

Mina put current enterprise AI adoption at “single digits,” even as consumer expectations continue to rise. In his view, front-line business teams still struggle to demonstrate 'trust'—that an AI system will behave reliably across unpredictable customer interactions—and to produce the data required to prove it. When security or legal stakeholders raise concerns, projects can be delayed for extended periods, and in some cases remain stuck just short of commercialization, he said.

LivePerson’s proposed remedy is Syntrix, a testing platform designed to simulate thousands of customer-service scenarios before an AI agent is deployed. The system uses synthetic users and generative test scenarios to stress-test customer questions, exception handling, and failure conditions, producing results that companies can submit as part of internal review and approval processes.

“When security or legal blocks you, you need to show evidence: we tested this many scenarios and we have the outcomes,” Mina said. “In the end, commercialization is about evidence-based confidence, not a demo.”

The company is also emphasizing runtime safeguards. LivePerson said it has developed a so-called 'guardian agent' that monitors 100% of conversations handled by both human representatives and chatbots in real time, continuously assessing whether interactions remain within normal bounds or require escalation. The approach is aimed at reducing risks that have made generative AI contentious in customer support, including hallucinations, inappropriate responses, and the potential for complaints to spiral into reputational damage.

Alongside governance tooling, LivePerson has been modernizing its infrastructure. Mina said the firm recently completed a multi-year migration to Google Cloud, shedding a large portion of around two decades of accumulated on-premises technical debt. For customers, the move is intended to combine access to Google’s Gemini models with hyperscale stability—an attempt to meet the operational requirements of AI contact centers, where scalability, latency, and flexibility in model choice can be decisive.

Industry observers say the company’s focus reflects a broader shift in the enterprise AI market. Early enthusiasm for generative AI was fueled by flashy demonstrations and promises of rapid productivity gains. But as the market transitions to budget execution and production accountability, evaluation criteria are increasingly centered on security, auditability, and clear ownership of operational risk—especially in customer-facing environments where mistakes can directly affect brand trust.

Mina likened the demand trend to “a train that has already started moving,” arguing that consumers now expect more automated, AI-driven service experiences. The critical task for vendors and enterprises, he said, is ensuring those experiences can be delivered with strong controls and defensible safeguards. For companies competing on 'AI customer experience', the next phase appears less about announcing deployments—and more about proving reliability, governance readiness, and real-time oversight at scale.