43% of Indian Enterprises Struggle to Control AI Costs, Splunk Report Finds

Indian businesses are investing in artificial intelligence, but many are finding it difficult to predict how much AI systems will cost to operate. According to Cisco-owned Splunk’s State of Observability 2026 report, 43% of Indian respondents said their organizations could not accurately control AI costs, highlighting a growing financial challenge as enterprises expand their use of AI models and autonomous agents.

The report points to a wider problem: organizations need to understand not only what their AI systems cost, but also how well they perform, whether they are secure and whether they deliver measurable business value. Although Indian respondents reported stronger capabilities in several areas than the global average, gaps in cost attribution and workload monitoring remain significant obstacles to scaling AI reliably.

Key Takeaways

• 43% struggle to control AI costs: Indian respondents reported difficulties accurately managing AI spending, compared with 55% globally.
• AI monitoring is a priority: 85% of Indian respondents said monitoring AI workloads was important to their organizations.
• Workload measurement remains difficult: 46% of Indian respondents said they could not effectively measure AI workload metrics.
• Unexpected bills are a major concern: 54% said unexpected billing spikes created more operational strain than high baseline costs.
• Security and governance are becoming more important: 81% were concerned about unintended consequences from AI agents.
• Better observability can improve cost accountability: Organizations classified as observability leaders were more likely to attribute AI costs accurately.

What the Splunk Report Says About AI Costs in India

The Splunk State of Observability 2026 report examines how organizations monitor and manage their technology systems as AI adoption grows. This year’s findings focus particularly on the challenges associated with AI agents, which can perform multistep tasks and interact with software, data and other systems.

For Indian enterprises, cost control is emerging as an important operational issue. AI spending can include model inference, cloud computing, data processing, storage, monitoring and the infrastructure needed to run applications. When organizations deploy multiple models and agents, understanding which activities generate costs can become increasingly complicated.

The report found that 43% of Indian respondents could not accurately control AI costs. The corresponding global figure was higher, at 55%.

Report finding	India	Global
Unable to control AI costs accurately	43%	55%
Monitoring AI workloads is a priority	85%	87%
Cannot effectively measure AI workload metrics	46%	56%
Unexpected billing spikes create more strain than baseline costs	54%	60%
Very effective at monitoring AI model quality	51%	35%

Source: Splunk State of Observability 2026. Figures reflect survey responses, not a census of all enterprises.

The results suggest that Indian organizations are comparatively well positioned in some areas, but there is still considerable room to improve financial oversight and operational visibility.

Why AI Costs Are Becoming Harder to Predict

Traditional software systems often have relatively predictable operating patterns once they are established. AI workloads can be more variable because costs depend on usage, model selection, prompt and response sizes, the number of agent actions, and the amount of data processed.

An AI agent may also initiate several operations while completing a single business task. For example, an agent helping a software engineer might inspect a codebase, call a model repeatedly, execute tests and attempt to fix errors. Each step can consume computing resources or generate additional model usage.

At enterprise scale, these costs can accumulate across teams and applications.

Unexpected billing spikes are a key concern

The report found that 54% of Indian respondents experienced greater operational strain from unexpected billing spikes than from high baseline costs. Globally, the figure was 60%.

This distinction matters because companies may be able to budget for a known monthly expense but struggle when usage suddenly increases. A new AI feature, an automated workflow running too frequently, or an agent repeatedly attempting the same task could create spending that is difficult to anticipate without effective monitoring.

Organizations therefore need mechanisms to detect unusual usage early, identify the source and take corrective action before costs escalate.

Measuring usage is not the same as understanding value

A business might know how many AI requests it processes without knowing whether those requests produce useful results.

The report found that 46% of Indian respondents could not effectively measure AI workload metrics. These measurements can include response quality, latency, reliability, usage and other indicators that help teams determine whether an AI system is functioning as intended.

Without these metrics, organizations may continue paying for a system that is expensive, unreliable or poorly suited to the task.

AI Monitoring Is a Business Priority

The survey found that 85% of Indian respondents considered monitoring AI workloads a priority. However, 39% said their organizations lacked clear business outcomes tied to AI monitoring, compared with 54% globally.

This reveals a gap between recognizing the need for monitoring and connecting it to business performance.

Monitoring becomes more useful when organizations can link technical measurements to outcomes such as lower operating costs, faster software development, improved customer service or fewer errors. Without that connection, teams may collect large amounts of operational data without knowing whether their investments are paying off.

For example, a customer-service agent should not be evaluated solely on the number of conversations it handles. A company may also need to measure resolution rates, customer satisfaction, escalation frequency and the cost of each successful resolution.

These measures can help decision-makers determine whether an AI deployment is producing enough value to justify its expense.

AI Agents Introduce New Security and Governance Risks

AI agents create additional monitoring requirements because they can take actions rather than simply generate text. Depending on their permissions, they may access enterprise information, interact with applications or execute workflows.

The Splunk report found that 81% of Indian respondents were concerned about unintended consequences from AI agents. Globally, 83% expressed this concern.

Such risks can include incorrect actions, unauthorized access, unexpected data exposure and automated decisions that produce undesirable results. An agent that behaves unpredictably may create financial costs as well as security or operational problems.

The report also found that 89% of Indian respondents said monitoring AI workloads had increased collaboration between observability and security teams, compared with 81% globally.

This points to a growing need for technical operations and security teams to work together. Monitoring should cover not only whether a system is running but also what it is doing, which resources it is accessing and whether its actions comply with organizational rules.

How Leading Organizations Manage AI Costs Better

The report identifies practices associated with stronger observability outcomes. These include using AI agents to help fix and prevent problems, monitoring model quality alongside cost and security, and connecting technical telemetry to business impact.

The difference in cost attribution is notable. Among organizations classified as observability leaders, 91% could accurately attribute AI costs, compared with 69% of non-leaders.

This suggests that organizations with more mature monitoring practices are better equipped to identify which workloads, teams or applications drive spending.

The report also found that 54% of observability leaders reported at least $5 million in annual business revenue from observability, compared with 28% of non-leaders. This is an association reported by the study, not proof that observability alone caused the difference.

For enterprises, practical steps include setting usage budgets, tracking model consumption by application, alerting teams to unusual spending and measuring whether AI-generated outputs meet quality requirements.

What Indian Businesses Should Do Next

Organizations looking to expand AI adoption can begin by making costs and performance visible at the level where decisions are made.

First, businesses should assign AI spending to specific products, teams and workflows. Aggregated cloud bills can obscure which applications are responsible for rising costs.

Second, they should establish alerts and limits for usage-intensive workloads. These controls can help teams investigate unexpected increases before they affect budgets.

Third, companies should evaluate model quality alongside price. A cheaper model may require more retries or human correction, potentially increasing the total cost of completing a task.

Finally, AI monitoring should include security, access permissions and business outcomes. A system that is inexpensive but unreliable or unsafe is not necessarily a successful investment.

The Bigger Picture

The Splunk findings highlight a shift in enterprise AI priorities. Early adoption often focuses on experimentation and demonstrating what AI can do. As organizations move toward production deployments, financial predictability, security and measurable results become more important.

India’s comparatively stronger results in some observability measures suggest that local enterprises are developing capabilities to manage AI workloads. However, the remaining gaps show that deploying AI at scale requires more than access to advanced models. Companies need operational systems that connect usage, cost, quality and business value.

As AI agents become more deeply embedded in business processes, the ability to monitor and govern them could become a competitive advantage. Enterprises that understand the full cost and impact of their AI deployments will be better positioned to expand successful applications while limiting waste and risk.

FAQs

1. What percentage of Indian enterprises struggle to control AI costs?

According to the Splunk State of Observability 2026 report, 43% of Indian respondents said their organizations could not accurately control AI costs.

2. How does India compare with the global average?

The global figure for organizations unable to control AI costs was 55%, compared with 43% in India. However, 46% of Indian respondents still reported difficulty effectively measuring AI workload metrics.

3. Why are AI costs difficult to manage?

Costs can vary with model usage, data processing, cloud infrastructure and the number of operations performed by AI agents. Without detailed monitoring, companies may struggle to identify the sources of unexpected spending.

4. What is AI observability?

AI observability involves monitoring AI systems to understand their usage, performance, output quality, costs and security. It helps organizations identify problems and assess whether AI workloads are delivering expected results.

5. What can companies do to control AI spending?

Businesses can track costs by application and team, establish usage budgets and alerts, monitor model quality, investigate unexpected billing spikes and connect AI spending to measurable business outcomes.

Looking Ahead

The next phase of enterprise AI adoption will increasingly depend on whether companies can operate AI systems predictably, rather than simply deploy them. As usage grows, cost attribution and workload monitoring will help businesses distinguish useful automation from expensive experimentation.

For Indian enterprises, the findings provide a practical reminder that AI investments need clear ownership, measurable outcomes and effective controls. Organizations that combine cost visibility with quality and security monitoring may be better equipped to scale AI while keeping operational risks and spending under control.