AI Agents Drive Unexpected Business Costs, Splunk Report Finds

AI agents are creating new challenges for businesses trying to manage their technology budgets, with unpredictable usage charges becoming a bigger financial concern than consistently high costs. According to Splunk’s State of Observability 2026 report, 60% of surveyed organizations say sudden spikes in AI bills put more financial strain on their business than a higher but predictable baseline cost.

The problem is becoming more pressing as companies adopt agentic AI systems that can perform tasks independently. These agents may consume vastly different amounts of computing resources and tokens depending on the task, its complexity, and the model being used, making monthly expenses harder to forecast.

The report found that 55% of organizations struggle to control their AI costs effectively. Meanwhile, 57% are concerned that their telemetry expenses—the costs of collecting and analyzing operational data—are now higher than the infrastructure they monitor.

Agentic AI Makes Cost Forecasting More Difficult

Consumption-based pricing is adding uncertainty to AI budgets. Unlike fixed subscription plans, where businesses pay a predictable amount per user or seat, usage-based models can produce fluctuating bills as AI workloads change.

Agentic AI adds another layer of complexity because the amount of work an agent performs can vary significantly from one task to another. A simple request may require relatively few tokens, while a complex workflow involving multiple steps and model calls can consume considerably more.

For finance teams, this creates a challenge: even when an organization understands its overall AI spending, predicting the next bill can be difficult. The report suggests that unexpected cost spikes are a greater source of financial pressure than consistently high expenses, highlighting the appeal of more predictable pricing structures.

Observability and Security Teams Are Working More Closely

Despite the challenges surrounding AI costs, the growing use of AI monitoring tools appears to be improving collaboration between observability and security teams.

Splunk’s report found that 81% of respondents say monitoring AI workloads has strengthened collaboration between these teams. In addition, 73% say they can accurately trace problems back to their underlying security causes, up from 65% in the previous year’s report.

The findings point to a closer relationship between operational monitoring and security analysis, as organizations work to understand not only how their AI systems perform but also where risks and failures originate.

Cisco Chief Product Officer Jeetu Patel highlighted the importance of this shift, saying, “This report shows that observability and security data are converging, and that teams need to collaborate more than ever.”

Greater information sharing can help organizations investigate incidents more effectively and understand how AI workloads affect both system performance and security.

Tracking AI Costs at the Model and Agent Level

The report also highlights the role observability teams can play in helping businesses understand the return on their AI investments.

According to Splunk, 91% of observability leaders say they can accurately attribute AI costs to individual models, agents, and workloads. Among organizations that do not qualify as observability leaders, the figure drops to 69%.

This level of visibility can help businesses identify which AI agents and models consume the most resources, investigate unexpected spending increases, and make better-informed decisions about how AI systems are deployed.

Without detailed cost attribution, organizations may struggle to determine whether a particular AI workflow is delivering enough value to justify its expense. Tracking usage at a more granular level can make it easier to assess costs alongside performance and business outcomes.

Businesses Want More Predictable AI Pricing

The growing use of consumption-based pricing presents a dilemma for businesses. Companies may value the flexibility of paying for what they use, but unpredictable demand can make financial planning more difficult.

Splunk’s findings suggest that many organizations would prefer a pricing model with fewer surprises. A fixed, seat-based subscription can make budgeting easier because costs are generally more stable, whereas usage-based AI services can generate sudden increases when workloads expand.

That does not necessarily mean consumption-based pricing is unsuitable for businesses. Rather, the findings highlight the need for stronger monitoring, clearer usage reporting, and better forecasting as AI adoption grows.

For organizations deploying autonomous AI agents, understanding the relationship between usage, performance, and cost is becoming an important part of managing AI successfully.

As agentic AI becomes more common, businesses will need to balance the benefits of automation with the financial uncertainty that comes with variable workloads. Better observability may help them gain control over spending while strengthening collaboration between finance, operations, and security teams.