Swedish communications provider Sinch surveyed more than 2,500 AI decision-makers across countries and industries for its "AI Production Paradox" study. The research found that roughly three quarters (74%) of companies withdrew or shut down AI-based customer communication systems that had already been deployed in production due to governance and oversight issues.
Even the most mature organizations report high withdrawal rates
Sinch highlights that the withdrawal rate rises to 81% among organizations that have "fully mature protection and governance frameworks." Daniel Morris, Sinch's product director, said this suggests governance alone does not solve the problem: "The most advanced organizations do not make fewer mistakes, they detect mistakes sooner. A higher withdrawal rate reflects better monitoring and control, not poorer performance." Morris added that the data points to deeper issues.
Trust, security and compliance prioritized over AI development
According to Sinch, 84% of AI engineering teams spend at least half of their time building security infrastructure, leaving less capacity for developing AI features. A company spokesperson said 75% of respondents ranked trust, security and compliance among their top three priorities, while AI development itself was prioritized by 63%.
Sinch interprets these findings as many organizations recognizing that the core challenge is not merely making AI work, but making it work safely. The company warned that the operational cost of running AI safely at scale is substantially higher than many organizations expect.
Size is not a shield
Sinch also noted that withdrawal rates remain consistently high across regions and industries, indicating that organization size or budget does not provide real protection: "Withdrawal is not a symptom of underfunding or a lack of proper protective systems." As a communications provider, Sinch added that AI customer agents are often deployed on communication infrastructures that were not designed specifically for AI systems.
Industry context and Gartner's outlook
Sinch's results align with numerous reports that replacing customer service staff with AI has not met expectations in many companies. Gartner indicated in June 2025 that half of the organizations that expected substantial headcount reductions from AI in customer service would abandon those plans by 2027. Brian Weber, vice president and analyst in Gartner's customer service and support practice, has said fully agentless contact centers are not yet technically viable or operationally desirable.
Conclusion
Sinch's study warns that operating AI-driven customer service at scale is more complex and costly than initial hype suggested. High withdrawal rates and the large share of engineering effort devoted to security indicate that most organizations are prioritizing reliability, trust and compliance over aggressive automation plans.
