Purpose
This Frequently Asked Questions document is intended to provide clear and practical answers to common questions about the AI Narrative Analysis (NAR) solution. It is designed to help customers, partners, and implementation teams better understand the capabilities and intended use of the product. This FAQ should be used as a complementary resource alongside configuration guides and product documentation.
Questions
1. What kind of questions can I ask NAR about my commentary?
NAR is designed to answer questions that are directly related to financial performance, commentary, and analysis. Typical questions focus on summarizing commentary, identifying key drivers, explaining changes in results, or highlighting important trends and insights. Users commonly ask NAR to generate executive ready summaries, explain favorable or unfavorable variances, consolidate commentary across entities, or surface the most important takeaways for a given period. Questions that fall outside of financial or commentary related analysis are intentionally filtered out to ensure the solution remains focused and reliable.
2. Can NAR be plugged directly into my Workflow process?
Yes. NAR can be embedded directly into an existing Workflow as a Workspace step, allowing users to generate AI driven analysis as part of their standard close or reporting process. When deployed in this way, NAR can inherit Workflow context such as time, scenario, and entity, reducing the need for manual POV selection. NAR can also be deployed as a standalone dashboard for ad hoc analysis, giving customers flexibility to choose the deployment model that best fits their business processes.
3. How does NAR incorporate variance data into my AI Analysis?
NAR can incorporate variance data when the variance settings are enabled within the Cube Commentary Genesis block designer page. When enabled, users can choose to compare either two Scenarios or two Time periods, while all other dimensions are determined based on how the Cell Member Script and Aggregation Filters are configured.
When variance analysis is enabled, NAR evaluates the underlying financial data alongside the commentary and automatically highlights the most material favorable and unfavorable movements. The generated analysis ties narrative explanations directly back to actual financial results, ensuring the output is grounded in data rather than purely textual interpretation.
4. How does NAR handle multiple different languages?
NAR supports multiple languages out of the box, including all major global languages used in business, finance, and reporting. Users can simply enter prompts in their desired language, and NAR will generate analysis in that same language, regardless of the language used in the underlying commentary. Additionally, users can also explicitly specify which language they would like the output to be generated in. This allows global teams to work within a single solution without needing separate workflows or manual translation processes.
5. Where do my comments have to be stored for NAR to pull from?
NAR pulls commentary directly from the Cube using inherent OneStream functionality. As long as comments are stored in standard Cube annotation intersections and are properly referenced through the block’s member script and aggregation filters, NAR can access and analyze them. No special data structures or custom storage locations are required beyond standard OneStream commentary functionality.
6. Can I implement NAR prior to uploading my commentary to the Cube?
Yes. NAR can be fully implemented and configured before commentary data is available in the Cube. This allows customers to set up dashboards, parameters, workflows, and user experiences in advance. Once commentary is loaded or submitted to the Cube, NAR will immediately begin generating analysis without requiring additional configuration.
7. Is NAR able to “weight” certain comments more than others when formulating an analysis?
NAR does not apply explicit weighting rules to individual comments through configurable scoring or prioritization. Instead, it evaluates commentary based on the context provided through the Cube Commentary configuration and the data included in the analysis.
When variance analysis is enabled, commentary associated with more material favorable or unfavorable movements naturally carries greater influence in the generated output because the analysis is grounded in the underlying financial variances. This allows NAR to emphasize the most impactful drivers without requiring manual weighting logic.
When variance analysis is toggled off, NAR focuses solely on the content and themes present in the commentary itself. In this mode, all included comments are evaluated more evenly, and the analysis emphasizes recurring topics, explanations, and qualitative insights rather than prioritizing comments based on financial impact.
In both cases, customers can influence which comments are included by configuring member scripts, aggregation filters, and POV selections within the Cube Commentary block.
8. Can NAR be customized differently for different user groups?
Yes. NAR can be configured differently for various user groups by leveraging standard OneStream capabilities such as dashboards, parameters, workflows, and security access. Because NAR is built on Genesis and existing OneStream constructs, these customizations align naturally with how customers already manage user access and reporting experiences.
9. Does NAR store or persist AI generated output?
NAR itself does not automatically persist AI generated output unless the user chooses to save it. When users are satisfied with the generated analysis, they can save the output back to the Cube as commentary using standard OneStream functionality. This allows AI generated insights to be reused in reports, dashboards, workflows, or future analysis, while still keeping the user in control of what is ultimately stored and shared.
10. How is data security handled for AI generated analysis?
NAR inherits standard OneStream Platform security and does not introduce additional security requirements or access models. AI generated analysis is governed by the same permissions that control access to the underlying data and commentary. Users can only generate analysis for data they are authorized to view, and all interactions occur within the existing OneStream security framework. No additional roles, groups, or custom security configurations are required.