Oracle launches AI spending tool that ties cross-provider costs to finance controls

Enterprise finance teams can now bring AI spending across several providers into the same workflows they use to manage budgets and approvals. Oracle launched Cost Advisor for AI in limited availability on October 27, 2026, connecting usage data with financial controls and recommendations intended to help teams identify where to cut costs.

The managed service initially runs through Oracle Fusion Cloud EPM, its financial planning and governance software. Oracle says it combines AI usage and spending across providers, models, tools, teams and use cases, rather than presenting only a single provider’s bill.

From provider bills to accountable owners

Initial data sources include OCI Enterprise AI and tools from OpenAI and Anthropic. Customers can examine spending by application, workflow, owner, model, user or tool. That gives them several ways to investigate a bill: by the technology being used, the work it supports or the person responsible.

The EPM integration connects those costs to cost centers, budgets, owners and approval workflows. Oracle positions this shared financial record as a basis for forecasting spending, assigning responsibility and deciding which costs need attention—not simply displaying usage totals.

Oracle describes different uses for the same underlying information:

• Finance teams can review budget exposure within their existing planning processes.
• Technology and AI leaders can examine spending across their portfolio of AI work.
• Engineering and cloud-cost management teams can investigate unusual charges and opportunities to improve efficiency.

A cheaper model still needs a quality check

Oracle says Cost Advisor flags budget overruns, unusual spending, waste and charges that have not been assigned to an owner. Teams can then investigate the applications, tasks and groups behind those costs and focus on the areas with the greatest potential savings.

Its announcement illustrates the recommendation process with customer-service ticketing. If requests are running on a higher-cost model, the service can suggest a lower-cost alternative for comparable tasks and estimate the dollar savings. This is Oracle’s example of the intended workflow, not a reported customer savings result.

The team reviews that recommendation and tests whether the alternative meets its quality and response-time needs before deciding whether to switch. After a change, it can monitor spending to see the effect. Oracle’s described process therefore separates finding a less expensive option from deciding that it is suitable for a particular workload.

ChatGPT access is on the roadmap

The initial interface is EPM. For general availability, Oracle plans to make the insights accessible within ChatGPT and business-intelligence tools, including custom dashboards, in 2027. The stated goal is to let more stakeholders access the information through workflows suited to their roles, rather than requiring everyone to use the finance interface.

Those additional access points remain planned features, not part of the current limited-availability offering. Oracle’s future-product disclaimer says its roadmap is not a delivery commitment and that release timing, functionality and pricing can change. Customers evaluating the service now have an EPM-based offering; the wider set of interfaces is a future step.