Indian enterprises are showing stronger capabilities in monitoring and managing AI systems than their global peers, even as the growing use of AI agents creates new challenges around cost, quality and security, according to the Splunk State of Observability 2026 report.
The report by Cisco surveyed 2,050 ITOps and engineering professionals across 10 countries. It found that Indian organisations are relatively advanced in agent observability, with 51% describing their practices as very effective at monitoring AI model quality, compared with 35% globally. Indian organisations are also less likely to struggle with measuring AI workloads, with 46% reporting difficulty compared with 56% globally.
The focus on observability is also making its way into strategic decision-making. As many as 73% of Indian respondents said telemetry is a key factor in their organisation’s strategic planning process, significantly higher than the global figure of 54%. Telemetry refers to the operational data generated by applications, infrastructure and AI systems that helps companies understand how those systems are performing.
However, India is not immune to the financial and operational challenges created by AI. The report’s India data shows that 43% of respondents cannot accurately control AI costs, compared with 55% globally. At the same time, 87% of Indian respondents consider visibility into AI hallucinations important, while 70% say they are effective at monitoring them.
Indian organisations are also paying greater attention to AI safety and bias. Seventy-four percent consider visibility into AI safety important, while 70% attach similar importance to visibility into AI bias.
Asian enterprises show different stages of AI adoption
The broader Asian picture is more mixed, with Japan and Singapore showing different challenges from India.
Japanese organisations appear to be under greater cost pressure. Sixty-four percent said their total observability costs exceeded infrastructure and operational costs, compared with 57% globally. Japan was also more likely to identify AI infrastructure as a major observability cost driver, at 11% compared with 6% globally.
The country also showed gaps in monitoring AI behaviour. Only 25% of Japanese respondents rated their organisations as very effective at catching AI hallucinations, compared with 34% globally. Just 33% said agentic AI had significantly improved mean time to recovery, below the global figure of 42%.
Singapore, meanwhile, reported stronger operational gains from agentic AI but continued to face challenges in converting that adoption into measurable business outcomes. About 51% of respondents said agentic AI had significantly improved mean time to recovery, above the global 42%. However, only 39% said they had clear business outcomes tied to AI monitoring, compared with 46% globally.
Singapore also lagged the global average in confidence around AI-related root-cause analysis, with 47% saying they were moderately or very confident in isolating such issues, compared with 57% globally. The findings suggest that Asian enterprises are moving beyond simply experimenting with AI agents, but the maturity of their monitoring and governance practices varies considerably.
Global AI adoption is creating a cost and visibility problem
Globally, 87% of respondents said monitoring AI is a priority for their organisation. Yet the report highlights a significant gap between recognising the need for AI observability and actually being able to measure AI systems effectively. More than half, or 56%, said they were unable to measure AI workload metrics such as large language model response quality. Meanwhile, 85% said visibility into AI hallucinations was important, but only 58% said they were effective at monitoring them.
Cost is another growing concern. Fifty-five percent of respondents said they cannot accurately control AI costs. The problem is not only the overall size of AI bills but their unpredictability, with every prompt, rerun and token contributing to spending across multiple models and agents.
The report also points to the growing volume of telemetry generated by AI systems. Sixty-seven percent said the number of sources sending data into their observability platforms had increased. Interestingly, AI infrastructure itself was identified as a top observability cost driver by only 6% of respondents. Overly noisy logs were the biggest cost driver at 31%, followed by high-cardinality metrics at 27% and trace volume at 21%.
The growing use of AI agents is also bringing observability and cybersecurity teams closer together. Globally, 81% said monitoring AI workloads had increased collaboration with security teams, while 77% said ITOps and engineering teams now troubleshoot issues with security teams, up from 62% in 2025. Data sharing between teams also rose from 74% to 84%. For organisations that have moved further ahead, the report identifies three common practices: using AI agents to fix and prevent issues, monitoring model quality, cost, usage, security and privacy, and connecting technical telemetry to business outcomes.
These leaders are more likely to control and attribute AI costs accurately. Ninety-one percent can accurately attribute AI costs, compared with 69% among other organisations, while 76% say they can accurately control AI costs, against 41% of non-leaders.