Despite token costs coming down, businesses continue to grapple with managing AI costs as usage accelerates. The shift means IT leaders must rethink how to manage their investments while keeping the budget in check.
AI sprawl is contributing to cost control headaches for enterprises, according to a Flexera study. Roughly three-fifths of IT professionals said AI overspend has increased while more than two-thirds said they lacked visibility into AI software usage.
Large language model provider OpenAI published a Tuesday blog post outlining five steps for businesses to control spend, as well as understand AI use and value generation, in the agentic era. The playbook marks the latest vendor move to address rising AI cost concerns.
The company suggested IT leaders should increase visibility into AI use and spend, track model outcomes, incorporate governance, manage AI investments as a broader portfolio and match the product to demand once it has proven its value.
Determining overall value created by AI within an organization involves measurements such as time saved, how decision-making was improved and the number of tasks completed, according to the LLM provider.
Grasping AI’s value
While enterprise confidence in AI is growing, perception of the technology’s value isn’t always aligned across the C-suite and cost concerns are starting to become a top of mind issue for executives.
CIOs and IT decisionmakers reported feeling higher levels of confidence in the technology driving revenue growth than CEOs and board members, according to a Protiviti study.
Meanwhile, companies like Prudential Financial and Shutterstock are implementing stringent AI cost management strategies. Shutterstock CTO and CISO Courtney Totten said understanding AI costs is “no longer optional, it is foundational to business strategy,” during the FinOps X 2026 conference earlier this year.
OpenAI, one of the largest frontier AI model providers, recommended that IT leaders take five steps to invest in and manage AI use and spend more confidently.
First, enterprise leaders should increase visibility into who is using AI, along with which products and models they’re using, the capacity being consumed and what work the usage is supporting.
“Without that visibility, a growing bill is hard to interpret,” OpenAI said in the blog post.
Second, cheaper models may not always produce the best results, meaning they could fail and retry, driving more token usage. OpenAI suggested IT leaders evaluate models based on the work they are being asked to do and measure the full cost of reaching the outcome, including model and tool use, number of attempts and completion rate.
IT leaders can track model cost based on accepted outcomes, especially for priority workflows.
“In customer support, that might be a resolved case,” according to OpenAI. “In engineering, it might be a tested change that passes review. Pair that cost with business value such as time saved, cycle time reduced, revenue protected, risk avoided, or capacity created.”
Governance should also work as the “operating layer that determines which AI work can scale,” the LLM provider noted.
Governance should include outlining what context, data, applications and tools LLMs can access and what actions they’re allowed to take. IT leaders should also designate who approves higher-risk steps for LLMs and how extra capacity is granted for valuable workflows.
And while enterprises can grant access more broadly to everyday functions improving productivity and workflows, IT leaders should also make more strategic bets based on proprietary information, according to the LLM provider.
“The strongest candidates are workflows that repeat at meaningful scale, have clear ownership, and can be measured for quality, risk, and business value,” OpenAI said.
Finally, after establishing a workflow that generates value, IT leaders should “match the product, capacity, and support model to its demand,” OpenAI said.
