Oracle bets on agentic apps in Fusion suite to ‘fully’ automate business processes

Oracle is recasting its Fusion Cloud Applications suite as something that not just flags insights for humans to act on. The debut of Fusion Agentic Applications, an upgraded set of applications that embeds AI agents directly into transactional business workflows, is designed to make decisions without human intervention.

Where its earlier generative AI-based copilots and task-specific agents stopped at recommendations and singular tasks, the new agents are designed to operate with shared context, memory, and built-in governance to run business processes end-to-end, a shift that Natalia Rachelson, SVP of product management for Oracle Fusion Applications, says will let CIOs cut back on manual oversight and accelerate execution.

The new applications, Rachelson said, also mark Oracle’s shift from a workflow- and assistance-based approach to an outcome-driven, agent-led execution model.

“We are delivering Fusion Agentic Applications designed to target specific outcomes for customers. For example, a customer could use an agentic application to reduce DSO (days sales outstanding) by X%; or source parts and optimize cost, quality delivery time; or reduce cost accounting variance by X%; or pay all invoices over $X amount within 15 days,” Rachelson told CIO.com.

To ensure that these applications meet the desired outcome, Oracle is also offering optional deterministic guardrails and flexible approval frameworks, including human-in-the-loop approval controls, which Rachelson said will allow enterprises to choose the level of autonomy they are comfortable with.

Execution gains come with trade-offs

The new set of applications could deliver value for CIOs and enterprises, but there are caveats.

“Oracle’s built-in agents can abstract the manual tasks away from the lowest levels of effort, freeing up scarce resources to focus on higher value-added tasks,” said Scott Bickley, advisory fellow at Info-Tech Research Group.

“The move from isolated AI agents to agentic applications signals a step-up level of advancement as Oracle moves from task-level automation to full process automation, yielding a business outcome.  This is significant as the ROI potential based on labor rationalization and increased throughput becomes scalable at this level of execution,” Bickley added.

However, the analyst warned that Oracle’s claim of “full process automation” is a “lofty endeavor” as the complexity involved in multi-step business process automation is materially higher than that of task-level agentification.

More so because in production environments, things don’t always go as planned, there are data hiccups, and sometimes human employees need to step in, said Robert Kramer, principal analyst at Moor Strategy and Insights.

In fact, HFS Research’s leader of executive research Ashish Chaturvedi, says “supervised autonomy” is the choice of operation for the new applications across most enterprises in the next 24 months: “While agents will handle the 80% of routine execution, humans will handle exceptions, judgment calls, and anything with reputational or financial risk.”

Having said that, though, Chaturvedi expects Oracle Fusion’s existing customer base to adopt the new applications, as switching costs of not adopting are low because they are built-in and the switching costs of leaving Oracle to get a similar offering are astronomical.

Developers’ gains come with complexity, too

There are caveats for developers, too, analysts say, especially with the natural language-driven custom Agentic Applications Builder capability that has been added to its existing AI Agent Studio as part of the agentic applications recast.

While the capability removes the material volume of custom integration, UI work, and custom orchestration code for development teams, it shifts the burden to process engineering, security design, evaluation, model governance, and change management, according to Info-Tech Research Group’s Bickley.

More so because Oracle’s templates are not customizable, Bickley pointed out, adding that any significant variation from the packaged template or out-of-the-box agentic application requires significant modification or a net new agent developed.

In fact, according to Kate Leggett, principal analyst at Forrester, the long-term maintenance of custom agents and agentic applications is a burden that most developers or CIOs might initially overlook.

Moving to ‘action units’ from traditional SaaS pricing

With the move to agentic applications in Fusion, Oracle has also introduced a new pricing model, in the form of action units, moving away from traditional SaaS pricing, which is a per-user, per-month subscription model.

“Action Units are Oracle’s single consumption metric, each worth approximately one cent, that are consumed at different rates depending on the type of AI work executed, from simple queries to document generation to video creation, all drawing from one shared pool across every Fusion pillar,” Oracle’s Rachelson said.

“Each customer gets an allotment of action units to help them get started, but once they run out of action units, they will need to pay for additional action units,” Rachelson noted, adding that the previously released generative AI and AI agents embedded in Fusion continue to be packaged at no additional cost.

That shift in pricing, however, is likely to require a reset in how CIOs budget for and track AI usage, analysts say.

“Actual AI cost is in flux, and there is short-term pain as CIOs must quantify usage and ROI. Eventually, AI should be directly tied to outcomes. Outcome pricing will require high effort in vendor management and internal financial modeling,” said Forrester’s Leggett.

In fact, HFS Research’s Chaturvedi sees this short-term pain delaying contracts or purchase decisions: “CIOs have been burned before by vendors who promised outcomes and then argued about measurement when the bill came due.”

“Early adopters from Oracle’s inner circle will move fast because they have the executive air cover and the Oracle relationship capital to cut through friction. The broad installed base will take 12 to 18 months longer, waiting for proof points and contract templates to mature before they wade in,” Chaturvedi added.

In the long term, though, Leggett said that consumption-based pricing will offer better cost control than rigid, legacy seat-based licensing for CIOs as it provides a clearer line of sight into what they are paying for and supports more predictable budgeting.

To help CIOs make sense of those costs, Oracle is also introducing ROI tracking tools in the form of an Agent ROI dashboard inside AI Agent Studio that tracks time saved, cost savings, and productivity gains per agent across workflows, teams, and business functions.

Analysts, however, remain skeptical.

“There’s real demand for ROI visibility. However, metrics like time saved or productivity gains can be misleading without context. CIOs will still need independent validation of value,” said Stephanie Walter, leader of the AI stack at HyperFRAME Research.

Chaturvedi, too, says there’s a “credibility” issue with the dashboard, comparing it to a similar metric released by Salesforce refcently: “Essentially, the vendor selling you the agent is also the one measuring the agent’s value. The fox is auditing the henhouse.”

The 22 new agentic applications and the builder capability, according to Rachelson, are expected to be released in April as part of release 26b of Fusion Cloud Applications.

There are also plans to add new applications along with a separate marketplace for such applications, Rachelson added.