Three-quarters of AI customer service rollouts fail in production
Nearly three-quarters of enterprises that deploy AI customer communications agents later roll them back or shut them down, according to research from Swedish communications platform Sinch. The 74 percent rollback rate suggests that managing these systems reliably in production is far harder than the initial hype suggested.
Sinch surveyed more than 2,500 AI decision makers across various countries and industries for its AI Production Paradox study. The findings paint a picture of organizations struggling to operationalize AI systems, even when they invest heavily in safety controls.
Better monitoring reveals deeper problems
The most surprising finding: rollback rates actually climb to 81 percent among organizations with fully mature governance frameworks. This counterintuitive result suggests that governance alone cannot fix the underlying issues.
"The most advanced organizations aren't failing less; they're seeing failures sooner," said Daniel Morris, Sinch Chief Product Officer. "Higher rollback rates reflect better monitoring and control, not weaker performance. If governance was the fix, the most mature teams would roll back less, not more."
The data points to a structural problem: the operational cost of running AI safely at scale is far larger than most organizations expect.
Safety spending crowds out development
Eighty-four percent of AI engineering teams spend at least half their time on safety infrastructure, leaving minimal capacity for actual AI development. Most firms rank spending on trust, security, and compliance ahead of AI development itself-75 percent versus 63 percent respectively.
This budget allocation reflects where organizations see their real challenge. Getting AI to work is secondary to getting it to work safely.
Scale and budget offer no protection
Sinch found that rollback rates remain consistent across organizational size, budget, and geography. Company size is not a meaningful protective factor. Rollback is not a symptom of under-investment or being too small to afford proper guardrails.
This consistency across the board suggests the problem is structural rather than resource-based.
Broader staffing plans also stalling
Sinch's findings align with other recent data on AI customer service. Gartner reported in June 2025 that half of organizations expecting AI to significantly reduce customer service headcount would abandon those plans by 2027. Gartner also warned that fully agentless contact centers are not yet technically feasible or operationally desirable.
"Unexpected costs and unintended results" are driving abandonment, according to Gartner analyst Brian Weber. That matches what Sinch is seeing: organizations deploying AI agents only to discover the systems cannot meet their reliability and safety requirements in live environments.
For customer support professionals, the message is clear: AI for customer support remains a work in progress. The gap between pilot projects and production systems is wider than many expected.
Your membership also unlocks:
