Oracle adds AI agents for multi-step business analysis to Analytics Cloud
The November update lets agents combine datasets with business instructions and PDF references, while giving users sources and execution records to review.
Loading page…
The November update lets agents combine datasets with business instructions and PDF references, while giving users sources and execution records to review.
Listen to this story
Oracle's November 2026 Analytics Cloud release moves data-preparation AI from a September preview into documented functionality and adds Deep Insights Agents for investigations spanning datasets, company instructions, and PDFs. Agents can revise their plans, show supporting sources, and provide execution records for technical review; administrators control who creates, certifies, and uses them. The features make AI-generated analysis more inspectable, but Oracle's finance and sales scenarios are illustrative workflows, not evidence of customer outcomes.
Data Flow AI Assistant can recommend derived metrics, flags, and transformations, then generate columns from plain-English descriptions.
Deep Insights offers Instant mode for focused questions and high reasoning mode for broader investigations.
Authors and viewers can paste lists into filters; Trellis Scope lets authors choose shared or panel-specific chart scales.
Oracle’s November 2026 Analytics Cloud update brings AI assistance to preparing data and investigating business questions. Detailed in a November 19 overview, the release pairs plain-language requests for filters and formulas with agents that combine multiple datasets and company context. Users can inspect generated logic and review the agents’ work, rather than receive only a finished answer.
The data-preparation assistant has a documented precursor: Oracle’s September release notes listed creating and editing data flows with AI Assistant as a preview. In the November release notes, the corresponding entries no longer carry that label. They describe an assistant that recommends useful derived fields—new values calculated from existing data—and generates columns from plain-English descriptions.
Data Flow AI Assistant accepts everyday-language instructions to add a filter step or generate a formula for a new column. Oracle says it can also suggest metrics, flags and transformations, then create the formulas behind those suggestions. The workflow targets both choosing what to calculate and writing the expression that performs the calculation.
Oracle illustrates the process with a sales-promotion example. An author can ask for a filter that keeps only sales associated with a promotion, then request a formula expressing promotion cost as a percentage of sales. The author can inspect the proposed filter and review the formula before applying them. These are company-provided examples, not measured customer results.
Deep Insights Agents address questions that require several stages of analysis. Oracle’s documentation says an agent can refine its plan and continue investigating when it needs more evidence, before producing a structured report. It can work across multiple datasets and incorporate supplemental business instructions and PDF reference documents into that analysis.
The company’s finance example asks why cash has fallen despite stable working capital. An agent could examine balance-sheet, receivables and payables data alongside company policies. That example puts organizational guidance inside the investigation, rather than limiting the agent to figures in a single dataset. Oracle presents this as a possible workflow, not a demonstrated financial finding.
Oracle distinguishes two modes: Instant supports focused questions, while high reasoning supports broader investigations. Findings can include tables, charts and supporting sources. Users can review agent activity, save responses for sharing and download an execution record for technical review. Administrators can separately control who creates, certifies and uses Deep Insights Agents.
The same update changes how people select and display workbook data. Authors and viewers can paste a copied list into a list filter instead of selecting values individually. Oracle’s example uses a list of cities. Paste Values also works for parameters on the filter bar, in both authoring and consumer modes.
Workbook authors also gain a choice over trellis charts, which split a visualization into panels for groups such as product categories. A shared value-axis range helps readers compare quantities across panels. Independent ranges instead make differences within each group easier to see. The Trellis Scope property switches between those two presentations.
That chart change remains an explicit author decision: turning Trellis Scope on gives each panel its own range. Leaving it off preserves the shared scale for cross-group comparisons.
Loading discussion...
Join the conversation
Explain when company context helps—and when it might skew the answer.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.