Skip to main content
Reference

AI-Powered Report Viewer

The AI Report Viewer lets you open, inspect, and iterate on reports created with the AI-Powered Report Builder. From one screen you can see the chart, the raw data, the exact SQL, an expert AI analysis, and a side-chat to ask questions about the results.

image-20250820-203928.png

  1. Go to Operations → Reporting.

  2. Open a report by either:

    • Selecting it from a Custom Group you created when generating reports, or

    • Clicking Build Custom Reports to see your list of generated reports (or to create a new one).

  3. Click any report in the list to open it in the AI Report Viewer.

note

Tip: Use clear names and groups when saving reports so your team can quickly find them later.

Tip: Use clear names and groups when saving reports so your team can quickly find them later.

What you can do in the Viewer​

Across the top of the viewer you’ll see tabs for each part of the report:

Visualization​

View the chart the AI generated for your query (pie, bar, line, etc.).

  • Hover to see tooltips and segment values.

  • Click Download PNG to share the chart with colleagues or stakeholders.

Data​

See the complete table that powers the visualization.

  • Inspect rows and columns to validate results.

  • Click Download Excel to export the dataset for offline analysis or sharing.

SQL​

Review the exact SQL used to generate the data.

  • Great for learning how queries are structured or for adapting them to your own needs.

  • Copy the query with one click.

  • Parameters (e.g., start_date, end_date) are listed under the SQL so you can see the filters that were applied.

Analysis (contextual, not one-size-fits-all)​

When a report is first generated, UltraCart also generates a report-specific analysis prompt tailored to that report’s SQL, parameters, dataset, and chart type. This means the AI isn’t using a generic template—it’s guided by context that’s unique to your report so the write-up focuses on the right dimensions, filters, and business questions.

What the contextual prompt captures

  • The timeframe and filters you chose (e.g., last 90 days, channel = Paid Social).

  • The aggregation and groupings in your SQL (e.g., by SKU, by category, by state).

  • The metrics present (units, net revenue, AOV, margin if available) and any parameters.

  • The visualization type (pie, bar, line) to align commentary with what you see.

Why it matters

  • Produces relevant insights (e.g., concentration, movers/decliners, promo effects) instead of generic commentary.

  • Surfaces actionable recommendations tied to your exact dataset.

  • Ensures consistency: when you Edit a report (change dates, dimensions, filters, metrics, or chart), the system regenerates the analysis prompt and updates the analysis to match the new context.

note

Tip: If you want the analysis to emphasize a theme (e.g., margin over units, retention over acquisition), click Edit and say so—your new prompt and analysis will reflect that priority.

Tip: If you want the analysis to emphasize a theme (e.g., margin over units, retention over acquisition), click Edit and say so—your new prompt and analysis will reflect that priority.

note

Tip: Use the AI Chat to ask follow-ups about anything the analysis mentions; the chat already has the same full context (data, SQL, visualization, parameters, and the generated analysis).

Tip: Use the AI Chat to ask follow-ups about anything the analysis mentions; the chat already has the same full context (data, SQL, visualization, parameters, and the generated analysis).

Edit​

Iterate on the report without starting over.

  • Click Edit to re-engage the AI agent and describe what you want changed (date ranges, dimensions, filters, metrics, chart type, etc.).

  • The agent will regenerate SQL, visualization, and the analysis to match your instructions.

  • Save over the existing report or save as a new one to keep versions.

AI Chat (side panel)​

The side-chat agent automatically inherits the full context of the open report—the data table, the exact SQL, the visualization, the filters/parameters (e.g., date range), and the AI analysis. That means you can ask about any of these without re-explaining. It can cite rows, sanity-check the query, interpret the chart, and expand on the written insights.

Great for

  • Explaining spikes/dips and identifying the drivers (SKU, channel, campaign, region, cohort).

  • Validating the SQL logic and suggesting safer filters or edge-case fixes.

  • Translating the chart/table into plain-English takeaways and next-step tests.

  • Producing quick comparisons (MoM/YoY, channel vs. channel, SKU vs. category).

  • Drafting short summaries you can paste into email/Slack.

note

Note: Chat won’t change the report by itself. If an answer requires different dimensions/metrics, click Edit and tell the agent what to adjust; it will regenerate the SQL, visualization, and analysis.

Note: Chat won’t change the report by itself. If an answer requires different dimensions/metrics, click Edit and tell the agent what to adjust; it will regenerate the SQL, visualization, and analysis.

Prompting tips​

  • Be specific about metric, timeframe, dimension, and comparison.
    Template: “Compare {metric} for {dimension} over {timeframe} vs {baseline}; return a ranked table and 3 insights.”

  • Ask for the format you want: table, bullets, or a short paragraph.

  • If you need new fields, say so (“add COGS and gross margin”) and then use Edit to update the report.

Settings & Token Cost​

Settings​

Click Settings (top-right) to control how AI is used for custom reports.

  • Opt-in to custom reports and set a monthly AI Budget.

  • (Optional) Novice SQL Comments adds instructional comments to generated SQL.

  • Permissions: Changing budgets and some settings requires the Edit Service Plan permission in user configuration. If you see a message indicating you lack permission, contact your account admin.

Token Cost​

Click Token Cost for full transparency into AI usage during your session.

  • Light model tokens are used for predictable tasks.

  • Heavy model tokens are used for deep reasoning and analysis.

  • You’ll see counts for Input, Output, and Cache reads/creates to understand where compute was spent.

Best practices​

  • Name reports clearly (purpose, date range, primary metric) and group them so teams can find them.

  • Validate the data on the Data tab before sharing charts.

  • Use Edit for iteration: ask for new dimensions, filters, or chart styles in plain English.

  • Save as new when exploring big changes so you keep a clean history.

  • Leverage Analysis to turn findings into action items your team can test.

See also​

AI-Powered Report Builder

Was this page helpful?